How Vendar finds the top in every area
Every top list answers one question: a service, a city and a buyer need. An agency is ranked only on work a source shows it has done, scored from those sources, and every position shows the facts behind it. Read the short version first; the full rule set is below, and the same rules are served as JSON at /ai/methodology.
The short version
Every top list answers one question, such as “SEO agencies for online stores in Los Angeles”. An agency gets a place only if a source we can link to shows it has done that work. Then seven things are scored, out of 100 points. Proof of work and fit carry 55 of them.
- Proof of work for your question30Case studies and client reviews that show this exact kind of work, weighted by who wrote them.Coalition Technologies: 30.0 of 30
- Fit to your city, budget and business25Office or service area in the city, a price inside the usual band, the right type of client.Coalition Technologies: 25.0 of 25
- How central this service is15A specialist ranks above an agency for which it is a side offer.Coalition Technologies: 5.3 of 15
- Track record10Years in business and client retention where documented.Coalition Technologies: 9.4 of 10
- Openness10Published prices, team size and how the work is run.Coalition Technologies: 7.5 of 10
- Outside reviews5Rating on platforms with 10 or more reviews, discounted when there are only a few.Coalition Technologies: 4.7 of 5
- Recognition5Awards, certifications and press on file; the agency's own word counts half.Coalition Technologies: 5.0 of 5
Minus penalties for a profile that is flagged, stale, unsupported by any independent source or not fully reviewed (-5, -5, -3, -2 points). Confidence is shown next to the score and never changes the order. No one pays for a place.
For “Which SEO agencies have documented ecommerce-SEO work for businesses in Los Angeles?” we looked at 131 agencies. 39 have documented work and are ranked; 92 are listed below without a rank. Coalition Technologies is #1 with 87/100: 15 third-party claims back its ecommerce seo work, and its weakest point is how central this service is (5.3 of 15).
3 of 8 data stages are fully in force; the rest are partly built and say exactly what runs today.
An AI reader checks every claim against the page it cites and must quote it; code confirms the quote is really on that page.
Recommended, Insufficient data, Shortlist with checks, Ranked, Not ranked yet: set by the best position and how far its evidence checks out.
Service × city × need. A company can rank first for local SEO in Los Angeles and be unranked for enterprise SEO in the same city. A builder can lead wooden houses near Phoenix and not appear for modular homes.
Case studies, customer reviews, permits, registry entries, measurements: each with a source URL, a date and a verification status. No claim, no rank.
No paid boost in the organic order and no hidden curation. An override, if one ever exists, is labelled on the page with a reason.
The page defines the service, the city and the need, and publishes the rubric: what counts as evidence for this question.
Service on the profile, market served, at least one outside claim that documents work for this need; in regulated trades, a licence confirmed on the registry. Everyone else is listed below, unranked.
The track's weighted criteria computed from countable facts, minus penalties. Confidence is computed separately.
Each position shows its points per criterion, the claims behind them with sources, and its limits. The machine feed carries the same.
The full rule set18 sections: every weight and how it is computed, the data stages and what runs today, eligibility, penalties, confidence, rubrics, a live position claim by claim, and how a profile verdict is built.Open ↓Close ↑
One real position, end to end
One question, one pool
“Which SEO agencies have documented ecommerce-SEO work for businesses in Los Angeles?”
Typical reason a vendor is not ranked: No ecommerce SEO claim on file yet.
Seven criteria, minus penalties
- Evidence for this query30.0 / 30
68 ecommerce SEO claims on file: 15 third-party, 32 first-party, 21 profile signals; outside claims from 8 distinct sites
- Fit to this query25.0 / 25
Headquartered in Los Angeles (Los Angeles, CA, USA)
- Capability depth5.3 / 15
15 of 82 listed services are SEO services
- Reliability9.4 / 10
In business since 2009
- Transparency7.5 / 10
Pricing disclosed (From $1,800/month)
- Reputation4.7 / 5
281 reviews at 4.7 on Google, Trustpilot
- Market credibility5.0 / 5
8 recognitions with a source on file. 8 confirmed by an outside source: Ordergroove partner, Klaviyo partner, Teamwork Commerce partner, Professional Developers partner and 4 more
The claims behind Coalition Technologies
68 of 322 claims on file name ecommerce seo work; 15 are third-party, 13 of them documents rather than directory listings. Each claim carries its weight (1.0 base for a checked third-party claim, plus 0.15 for the rubric's strongest terms and 0.15 for a number); each further claim from the same source counts half.
- third-partychecked against its page· documents workweight 1.30
Coalition Technologies used DataFeedWatch to create and manage a Shopping feed for a B2B eCommerce store. The case study reports an overall 451% increase in ROAS.
datafeedwatch.com
- third-partychecked against its page· documents workweight 1.20
Essential Sports Nutrition's SEO team used the client's Shopify account and Google Analytics to conduct a thorough SEO analysis of its Shopify website; the final analysis was deli…
shopify.com
- third-partychecked against its page· documents workweight 1.20
AMIClubwear needed to upgrade from Liquid checkout to the new Checkout Extensibility, and Coalition provided a fixed price with a fast delivery date.
shopify.com
- third-partychecked against its page· documents workweight 1.20
White Oak Ambience said Coalition updated and redesigned its website and fixed back end SEO issues.
shopify.com
- third-partychecked against its page· documents workweight 1.20
Through SEO efforts for client Shopping in Japan, a business that started as a small local shop selling primarily on eBay, Coalition Technologies grew revenue by 100% and traffic…
magazine.retail-today.com
- third-partychecked against its page· documents workweight 1.20
After switching from Volusion to Shopify, IronVapin sought SEO consultations and contacted Coalition Technologies.
shopify.com
Confidence, basis and the order
High confidence · Backed by an outside source. Confidence is shown next to the score and never changes the order.
- #1Coalition TechnologiesHigh confidence87
- #2GR0Moderate confidence76
- #3IcepopHigh confidence73
Ties break on evidence points, then third-party claims, then name. The listing shows every position this way: open the ranking or its machine feed.
1. One method, every area
An area is a family of services with one kind of buyer and one kind of evidence: marketing, development, design, construction. Every area runs the same six steps: question, eligibility, score, confidence, order, explanation. An area is admitted to ranking only through a published area profile: which track sets the weights, which ranking variants apply, which services it covers, what counts as a document for this kind of work, and which registry or instrument is checked. The profile is on this page; a page without one stays on the transitional model and says so.
- A top list is one question: a service, a city and a need. There is no overall best agency or best builder across questions.
- Top N means the first N ranked positions by score. Confidence is shown next to each; a lower-confidence position is shown, never hidden and never promoted.
- A page may call its ranking a top list only when it passes the indexing gate: at least five ranked vendors, three in a small city. Below that it is a ranking in progress and says so.
- The top is recomputed whenever evidence changes, and the page shows the data date it rests on.
- Admission to the top is by evidence under the area's track; no vendor pays for a position and no editor places one without a label.
2. Two tracks: B2B and B2C
The buyer decides what evidence matters. A company hiring an agency for a year reads case studies and asks for references; a person building a house does it once, with their own money, and cannot check the work in advance. The B2B track is in force on every evidence-ranked listing. The B2C track is published here as a proposal so that its rules are public before the first consumer listing opens; no page uses it yet.
B2B: agencies and studios
version 2026.11.9 · In force on every evidence-ranked listing since 2026-11-05.
A company picks a partner for months of work. It can read case studies, ask for references and compare proposals, so documented results for similar clients, depth in the service and fit to budget and client type decide.
Evidence and fit carry 55 of 100 points because the engagement is long and specific: the question is whether this agency has done this work for this kind of client at this budget. Capability depth gets 15 because a side offer is a risk over a year of work. Reputation gets 5: review platforms for agencies are thin and self-selected.
- Case studies that name the client, the service and a result
- Third-party client reviews that describe the service and the outcome
- Projects and portfolio entries with a live result
- Awards, certifications and media mentions from outside the agency
- Service pages, price lists and positioning statements: they say what is offered, not what was done
- Directory profiles and self-set focus percentages
- Employer reviews and social media followers
Full rule set: eligibility, how each criterion is computed, penalties
- The vendor lists the page's service on its profile.
- The vendor serves the page's market: headquarters or a listed service region in the city. When a need page has fewer than ten ranked vendors from the city, agencies from other cities join it only if they are headquartered in the same state or have published work (a case study, a project, a client review) that names the city or its state. A line saying a company serves the whole country is not enough. They are ranked by the same score, geography counts 0.3 for them, and each position says where the agency works from. Mobile development is bought remotely, so its city pages take agencies from anywhere in the United States under the same condition: documented work for the need. The same holds for SEO needs bought remotely: international, enterprise, consulting, recovery, marketplaces, startups, B2B SaaS, ecommerce, AI search visibility, audits, technical and content-led SEO. Local SEO, small business, lead generation and high-ticket services stay with agencies from the city and its state.
- A need page that still has fewer than ten ranked agencies after that is filled from the city's own agencies whose sources say they offer the need but document no project for it yet. They always stand below every agency with documented work, are labelled "Offers this service; no documented project on file yet", carry lower confidence, and do not count toward the threshold that opens a page to search engines.
- For an intent page, at least one claim on file supports the intent under the page's rubric, and at least one of those claims has a source address outside Vendar's own profile fields; the agency's own site counts. Our own service listings and profile tags describe the intent but document nothing, so they add weight to a vendor that already has outside evidence and never earn a rank on their own. Vendors without outside evidence appear below the ranking, unranked, never padded into it.
- For an intent page, at least one of those claims documents work the vendor has done: a case study, a project or portfolio entry, or a client review. Service pages, price lists, positioning statements and directory listings say what a vendor offers, not what it did; they add weight to a vendor that already has documented work and never earn a documented position. They can only fill a page that is short of ten, below every documented position and labelled as such (see below).
- A position may rest on case studies the agency published on its own site. Own published case studies count toward eligibility; independent confirmation adds to the score and spares the 3-point penalty for no independent source. Such a position is labelled "Agency's own published cases" until an independent source documents the same kind of work.
- Filtered views (an entry price, a pricing model, a team size) are not rankings of their own: they show agencies ranked for at least one need in the city, each with the score of its best need, and keep only those that meet the condition. An agency that does not publish the fact is listed below with the reason.
- Draft and needs-data profiles are never listed.
Claims on file that support the page's exact service and intent, weighted by who made them and whether they were checked.
- Third-party claims (client reviews, press, awards) that name the intent: 1.0 verified, 0.8 claimed
- First-party claims that name the intent: the vendor's own site and platforms where the vendor writes about itself (partner directories and portals, case-study hosting, paid release wires): 0.8 verified, 0.6 claimed
- Directory listings that only name the service: 0.4, with no bonuses
- Profile signals (listed services, best-fit tags, editorial use cases): 0.25; a claim whose speaker is unknown: 0.3
- Bonus 0.15 when the claim uses the rubric's strongest terms, and 0.15 when it carries a number
- Three fully weighted claims reach the maximum; the total is capped at 30 points
How closely the vendor matches the market, the budget band and the client type the page is about.
- Geography (half of the criterion): headquarters in the city 1.0, city listed as a service region 0.75, an agency from another city admitted by the regional rule 0.3
- Budget (30%): monthly floor disclosed and inside the intent's band 1.0, disclosed but above it 0.5, only a project floor or hourly rate 0.6, nothing disclosed 0.2
- Client type (20%): profile names the intent's buyer types 1.0, adjacent wording 0.7, otherwise 0.3
How much of what the vendor does is this service and this intent, as opposed to a side offer.
- Share of the vendor's listed services that belong to the service family (half)
- Share of the vendor's evidence that supports the intent, scaled so that 40% of claims counts as full depth (half)
Operating history and retention where they are documented.
- Years in business from the founding year, full marks at ten years; an unknown founding year counts as 0.3 (60%)
- Client retention rate where documented, otherwise a neutral 0.3 (40%)
How much decision-relevant information the vendor discloses.
- Any pricing signal disclosed (50%)
- Team size disclosed (25%)
- Workflow, onboarding or reporting cadence documented (25%)
Client review volume and rating on open review platforms: Google, Trustpilot and BBB. Paid agency directories are not counted.
- A platform counts only with a rating and 10 or more reviews, the same threshold as the profile; a platform with fewer reviews is shown as too few to count and adds nothing
- Average rating out of five across the counted platforms, weighted by review count and discounted by their combined review volume (17 reviews count for about 60% of the volume factor, 100 reviews for 100%)
Recognition on file with a source: awards, certifications, partner programs and media mentions, weighted by who confirms it.
- Awards, certifications, partner programs, third-party validations and media mentions named in the claims on file, each counted once; one claim can name several ("Meta Business and Google Featured Partner" counts as two)
- Confirmed recognition counts 1: its source is the program's own partner directory, the Inc. 5000 list, the award's site or independent press. Claimed recognition counts 0.5: its source is the agency's own site or a page where the agency enters its own data, such as a website builder profile, a self-listed agency directory or a press release wire
- Full marks at four, and only with at least two confirmed; with fewer, the criterion stops at three quarters
- Only recognition with a source address counts, and not when its source page does not support it; an award or listing from a paid agency directory never counts
- -5Profile flagged for review. An editor has flagged the profile; the flag is shown and the score is reduced until it is cleared.
- -5Evidence not re-checked in 18 months. No claim on file has been checked against its source in the last 18 months, or none has been checked at all. Profile verification, which agencies can pay for, does not count here.
- -3No independent source on file. Minus 3 when no independent source describes the agency's work for this page's need; the agency's own published case studies do not lift it. Who wrote the text decides, not the address: a client, the press or another independent writer counts, a page retelling what the agency says does not, and repeats of one case count once.
- -2Profile below decision-ready status. The profile has not completed the decision-ready review.
B2C: consumer services
version 2026.10-draft · Published as a proposal. No listing is ranked by this track yet; the first candidate areas are wooden houses and modular homes.
A person buys one job, often once, with their own money and little way to check the work in advance. Legal standing, protection if something goes wrong and the experience of other customers weigh more; capability depth matters less, because the job is the whole relationship.
Documented work keeps 30 points: a builder who has finished houses like this one is still the strongest signal. Trust and compliance take 20 because a lapsed licence or no insurance turns a good price into a loss, and the buyer cannot check this alone. Reputation rises to 15 because consumer reviews are many and describe the actual job. Capability depth is dropped: a roofer who also fits windows is not a risk to a buyer who needs a roof.
- Completed jobs with a date, a location and photos
- Customer reviews that describe the job, on independent platforms
- Permit, inspection and registry records
- Licence entries, insurance certificates and written warranty terms
- Price lists, service pages and claims such as 'we build any house'
- Social media followers and stock photos
- Employer reviews and directory profiles
Full rule set: eligibility, how each criterion is computed, penalties
- The vendor lists the page's service on its profile.
- The vendor serves the buyer's location: a base in the city or a stated service radius that covers it. Region-wide coverage counts only when the city pool is short.
- Where the trade requires a licence or registration, it is confirmed on the official registry and in force. An unconfirmed or lapsed licence keeps the vendor out of the ranking; the vendor is listed below it with the reason.
- At least one record on file documents a completed job of the kind the page is about, and at least one of those records comes from a source outside Vendar: a customer review that describes the job, a permit or inspection record, a dated project with a location. Price lists, service pages and our own tags describe an offer, not a job.
- Draft and needs-data profiles are never listed.
Completed jobs of the kind the page is about, each with a source: a dated project with a location and photos, a customer review that describes that job, a permit or inspection record.
- Third-party records (customer reviews that describe the job, permit and inspection records, settled warranty cases): 1.0 verified, 0.8 claimed
- First-party records (a dated project with location and photos, a signed handover): 0.8 verified, 0.6 claimed
- Profile signals (listed services, tags): 0.25
- Bonus 0.15 when the record uses the rubric's strongest terms, and 0.15 when it carries a number: an area, a delivery time, a price
- Three fully weighted records reach the maximum; the total is capped at 30 points
Legal standing and the buyer's protection if something goes wrong, confirmed on official sources.
- Licence or registration confirmed on the official registry where the trade requires one (40%); in a regulated trade this is also the eligibility gate
- Liability insurance or bonding documented (25%)
- Written warranty terms published (20%)
- Written contract and payment schedule published (15%)
How closely the vendor matches the buyer's location, the price band of the need and the kind of buyer.
- Service area (half of the criterion): base in the city 1.0, city inside a stated service radius 0.75, region-wide coverage only 0.3
- Price (30%): a published price or band inside the need's band 1.0, published but above it 0.5, quote on request only 0.2
- Buyer type (20%): the profile names the need's buyer, such as a homeowner or a family building a first house 1.0, adjacent wording 0.7, otherwise 0.3
What customers say, discounted for volume, age and concentration on one platform.
- Average rating out of five, discounted by review volume (17 reviews count for about 60% of the volume factor, 100 reviews for 100%)
- Reviews on two or more independent platforms earn a bonus; one platform cannot carry the criterion
- Reviews older than three years count at half weight
- A documented answer to a negative review counts for the vendor; a pattern of deleted or disputed reviews does not
How much the buyer can learn before asking for a quote.
- Prices, price bands or a cost calculator published (50%)
- A named owner or responsible person and a physical address (25%)
- Process, timeline and what is included documented (25%)
Operating history and responsiveness where they are documented.
- Years operating from the registration date, full marks at ten years (60%)
- Response time to a quote request where measured, otherwise a neutral 0.3 (40%)
- -10Unresolved regulatory complaint or judgment. A complaint upheld by a regulator or a court judgment against the vendor is on file and not shown as resolved. The record and its source are shown.
- -5Signs of review manipulation. Bursts of reviews in a short period, repeated text across reviews, or a platform flag on the profile. The signal and its source are shown.
- -5Profile flagged for review. An editor has flagged the profile; the flag is shown and the score is reduced until it is cleared.
- -5Evidence not re-checked in 18 months. No claim on file has been checked against its source in the last 18 months, or none has been checked at all. Profile verification, which agencies can pay for, does not count here.
- -3No independent source on file. Minus 3 when no independent source describes the agency's work for this page's need; the agency's own published case studies do not lift it. Who wrote the text decides, not the address: a client, the press or another independent writer counts, a page retelling what the agency says does not, and repeats of one case count once.
- Capability depth (15 points in B2B) is dropped in B2C: a roofer who also fits windows is not a risk to a buyer who needs a roof. The points move to trust and compliance (new, 20) and reputation (5 to 15).
- Compliance is both a gate and a criterion in B2C: the registry decides who is in; insurance, warranty and contract terms decide how many of the 20 points a licence holder gets.
- B2C reputation adds review age decay and a multi-platform bonus, because consumer reviews are many and concentrated on one platform at a time.
- B2C adds penalties for regulator complaints and court judgments, which exist for consumer trades and rarely for agencies.
3. How the top is determined in each area
Five ways to set an order. An area declares which it uses; measurement and the compliance gate run on top of evidence ranking, never instead of it.
| Variant | Rule | When used | What is published |
|---|---|---|---|
Evidence-ranked In force | Positions are set by the track's score, computed from sourced claims under a rubric published on this page, minus penalties. | The default for every area where vendors document their work: case studies, projects, customer reviews, official records. | The rubric, the weights, and for each position its points per criterion, the claims behind them with sources, and its limits. |
Measured Proposed | Vendar or an independent instrument measures an outcome itself and records it as a verified claim: the speed of a site a developer delivered, the search visibility of a client an agency names, a delivery time. Measurements add to the evidence criterion up to its cap and never replace documented work. | Areas where the result of the work is public and can be measured without the vendor's help. | The instrument, the date and the raw value of every measurement, next to the claim it supports. |
Compliance-gated Proposed | An official registry decides who is eligible; the B2C score then sets the order among those admitted. A registry entry is a gate, not a point: holding a licence does not rank a vendor above another licence holder. | Regulated consumer trades: construction, electrical and gas work, medical, legal and similar. | The registry checked, the date and the entry for every admitted vendor, and the reason for every exclusion. |
Profile-ranked Transitional | These pages are still ordered by the profile decision score, which averages two profile-completeness checklists, plus route-fit boosts and, on some routes, a curated order. They are being migrated to the evidence-weighted model intent by intent; a page states which model it uses. | City pages, hubs and intent pages that do not have a published rubric yet. | The page states that it uses this model and names the model version. |
Listed, unranked In force | Vendors without a qualifying claim appear below the ranking in alphabetical order, with no rank number and no score. | The tail of every page, and whole areas that have neither a rubric nor a profile score yet. | The reason the list is unranked, next to the list. |
Area profiles
Marketing
Nine SEO needs are ranked by rubric. Other marketing services stay on the profile-ranked model until their rubric is published here. Measured search visibility is proposed, not applied.
Evidence-ranked + Measured
SEO, Paid ads and performance marketing, Reputation management (ORM, SERM), Link building, Community marketing
Which SEO agencies have documented local-search work for businesses in Los Angeles?
- A case study that names the client, the service and a result
- A third-party client review that names the service and the outcome
- Measured search visibility of a client the agency names (proposed)
Instrument (proposed): rank and visibility checks on named client domains.
Development
Vendors and services are in the dataset; listings are unpublished while rubrics for web, ecommerce, mobile, custom software and AI development are written. Measured site performance is proposed.
Evidence-ranked + Measured
Web development, Ecommerce development, Mobile development, Custom software, AI development, MVP and product, DevOps and cloud
Which studios have shipped Shopify Plus stores for retailers in New York?
- A portfolio entry that names the client, the stack and a live URL or app store listing
- A client review that names the project type and the outcome
- Measured performance and uptime of a delivered site (proposed)
Instrument (proposed): page speed and uptime checks on delivered sites.
Design
In the taxonomy. No vendors are scored and no rubric is written yet.
Evidence-ranked
UI design, UX design, Product design
Which studios have designed and shipped a B2B SaaS product for a startup in Chicago?
- A portfolio case with the brief, the outcome and a live product
- A client review that names the project
- A design award from outside the studio
Construction
First B2C area. In the taxonomy; the rubric, the registry list and the price bands are drafted in the manual and not yet applied.
Compliance-gated + Evidence-ranked
Wooden houses, Modular homes
Which builders have completed wooden houses for families near Phoenix and hold a licence in force?
- A completed build with a date, a location and photos
- A permit or inspection record for that build
- A homeowner review that describes the build
- A licence entry, an insurance certificate and written warranty terms
Registry: the state contractor licence board; the entry, its status and the check date are shown.
4. Where the data comes from
The score is only as good as the claims under it. A profile is built from the vendor's site and the outside sources the collector can reach, and every fact keeps its source. Each stage says whether it is in force today and, where it is only partly built, what runs today and what does not.
Candidates for a question are found across the web, not from one list: search results, maps, business registries, professional directories, review platforms, LinkedIn, trade associations and the clients named in other vendors' case studies. Each candidate records where it was found.
Today candidates come from research batches per service and city and from web search. Maps, business registries, LinkedIn and trade association lists are not read yet.
Output: A candidate pool per service and city, with the discovery source on every entry.
An AI agent reads the vendor's own site and every outside source it can reach: review platforms, directories, registries, news and media, app stores, code repositories, court and regulator records, job postings. Every fact is stored as a claim with the source URL, the source class, the date it was seen and how it was extracted. Nothing is stored without a source, and the agent never fills a gap with an inference.
Today the collector reads the vendor's own site and the outside pages templated web searches find: client sites, press, awards, certifications, interviews, partner listings and review platforms. Registries, court and regulator records, app stores, code repositories and job postings are not read yet; LinkedIn and social networks are excluded as sources.
Output: Claims: text, source URL, source class, provenance, date seen, extraction method, verification status.
Sources that describe the same company under different names, domains or addresses are merged into one profile; a company that is really two is split. Matching uses the domain, the legal name, the address and the phone number together, never the name alone.
Today duplicates are caught when vendors are imported, by domain and name similarity. Merging across legal name, address and phone, and splitting a company that is really two, are not built.
Output: One profile per company, with every alias and every source it appears on.
Claims are mapped to profile fields (services, markets served, price signals, client types, team, founding year) and to the rubric terms of each question. Where sources disagree, for example team size on the site, on LinkedIn and on a directory, the disagreement is recorded and lowers confidence instead of being resolved by guess.
Mapping claims to profile fields and rubric terms is in force. Comparing a field across sources (team size on the site, a directory and a registry) is not built yet. What is in force is a consistency check of the profile itself: its counts, posture, caveats and questions must agree between the page and the feed, or the profile is marked needs_review.
Output: A structured profile with a consistency record per field.
Every claim is checked against the page it cites, without a human queue. A verifier model reads the saved page and answers whether the page supports the claim, with a verbatim quote; code then looks for that quote in the page text, and without it nothing is confirmed, whatever the model said. A claim with numbers, and any claim the verifier doubted or could not confirm, is read again by an arbiter model from another family. A new claim is published only when every model that read it approves; a claim already on file keeps the better of the two readings that code could ground. Checks repeat every evidence cycle (90 days). A human can overturn any decision, and the overturn is logged next to the models' verdicts.
Output: On every claim: confirmed, partly confirmed, not supported or unreachable, with the quote, the models, the prompt version and the date.
The profile is scored on every question it is eligible for under the area's track, and confidence is computed separately. A buyer's filters narrow the pool; the score sets the order inside it.
Output: A rank, a score out of 100, points per criterion, penalties, confidence and limits for every eligible question.
Pages and machine feeds are generated from the same records. Each position shows the claims behind it with their sources; each page shows its data date and the model version it uses.
Output: A human page and a JSON feed per question, with identical numbers.
Profiles are re-collected on a cadence set by how fast a signal changes, and immediately on an event: a vendor update, a buyer report, a dead link, a regulator record. Every change is logged, and the score history stays visible.
Today every change the collector or an editor makes is written to the profile's change log, and pages show the new data date. The score history per question is not shown yet.
Output: A change log per profile and a new data date on every page the change touched.
Source classes
| Source | Examples | Provenance | Used for |
|---|---|---|---|
| Official registries | Business registries, licence boards, court records, regulator complaint databases, permit and inspection records | official | Existence, legal standing, licence status, penalties; the eligibility gate for regulated trades (proposed: not read today) |
| The vendor's own site | Service pages, case studies, portfolio, pricing, team, process and contract pages | first_party | What is offered, documented work, price signals, transparency |
| Review platforms | Google, consumer review sites, professional directories with client reviews | third_party | Reputation, documented work described by the customer, review manipulation signals |
| Professional directories | Agency and software directories, trade association member lists | third_party | Discovery, team size, service focus as the vendor states it (a listing that only names the service weighs 0.4) |
| Listings the vendor writes | Platform partner directories and portals, case-study hosting sites, paid press-release wires | first_party | The vendor's own word on another domain, never an independent source; client reviews on a partner directory stay independent |
| Media and awards | News, trade press, award lists, conference programs | third_party | Market credibility, independent confirmation of a case |
| Clients' own properties | The client sites, apps and stores a vendor names in its case studies | third_party | Confirming a case exists; the target of measurements |
| Professional networks and code | LinkedIn company pages and roles, code repositories, app store listings, job postings | third_party | Team size and roles, technology, delivered apps, capacity and hiring signals (proposed: LinkedIn and social networks are not read as sources today) |
| Vendar measurements | Page speed and uptime of delivered sites, search visibility of named clients, response time to a quote request, registry re-checks | third_party | Measured evidence that needs no cooperation from the vendor (proposed) |
| Buyer reports through Vendar | What a buyer tells us 30 and 90 days after a shortlist: hired or not, delivered or not, on budget or not | third_party | Closed-loop outcomes, the strongest claim class once it exists (proposed) |
| Vendor-submitted evidence | Documents and links a vendor sends through its profile, and its replies to disputes | first_party | Filling gaps; counted as claimed until checked, never as verified on submission |
Matching a recommendation to a query
A buyer asks with a service, a city and a need, and narrows with filters. Filters decide who is in the pool; the score decides the order.
- A recommendation answers one question: a service, a city and a need, plus the buyer's filters. The same profile is scored on every question separately.
- Filters are gates, not points. Service family, service and need; city, service radius or national; budget band; client type; industry; engagement model; start date; language. A filter removes vendors from the pool and never adds points, except where a criterion already measures it, such as geography and budget inside Fit.
- Inside the filtered pool the order is the track's score, highest first; ties break on evidence points, then third-party claims, then alphabetically.
- A short pool is never padded with unranked vendors; a page below the indexing gate says it is a ranking in progress. Widening a short pool in labelled steps, nearby cities first, then national coverage, is proposed and not built.
- Every recommendation carries its limits: what is missing or unverified about each vendor for this question, and the questions a buyer should ask before outreach.
- A buyer never sees a recommendation the machine feed cannot reproduce: same pool, same order, same numbers.
5. How often the data is refreshed
Decided on 2026-10-28: the tiered cycle. Fast signals are re-collected every month, evidence every three months, and every profile is fully re-verified once a year. The readiness report already counts a verification as fresh for 90 days, which is the evidence cycle.
The stale-verification penalty stays at 18 months until the first full evidence cycle has run. From 2027-01-28 it tightens to six months, twice the evidence cycle, with a new model version. A listing whose data date is older than three months says so at the top from the same date.
The tiers
- Fast signals, monthly: licence status, site alive, review rating and count, pricing pages, team page.
- Evidence, each base cycle: case studies, portfolio, services, client types, press and awards.
- Full re-verification, yearly: registry, address, ownership, every claim a rank rests on.
- Event-driven, within days: a vendor update, a buyer report, a dead link, a regulator record, a dispute.
Options that were considered
- Every month
For: Freshest data; licence lapses and review bursts are caught early.
Against: Highest collection and verification cost; most changes found are noise.
- Every two months
For: Half the cost of monthly; still inside most review and pricing cycles.
Against: Awkward against quarterly reporting; a licence lapse can sit for eight weeks without the fast tier.
- Every three months
For: Matches how often agencies publish cases and change pricing; cheapest full pass.
Against: Too slow for licence status and reviews on its own.
- Tiered: monthly fast signals, quarterly evidence, yearly full check (chosen)Chosen
For: Spends collection where signals change; catches lapses monthly without re-reading every case study monthly.
Against: Needs each field tagged with its tier; the page must show two dates where they differ.
6. Beyond the directories
The leading directories in B2B and B2C share four weaknesses: one rating reused everywhere, reasons hidden inside the platform, paid placement in the order, and no view of time. Each row says what they do, what Vendar does, and honestly whether it is in force or proposed.
| Principle | Directories | Vendar | Status |
|---|---|---|---|
| One question, not one rating | One rating per company reused across every category and city. | A company is scored on each question separately; it can lead local SEO in one city and be unranked for enterprise SEO in the same city. | In force |
| Every point has a source | A star rating and a review count; the reasons stay inside the platform. | Each position shows its points per criterion and the claims behind them, with the URL, the source class, the verification status and the date. | In force |
| No paid position | Sponsored placements and paid tiers inside the organic order on most directories. | Paid visibility never changes the order; a featured placement carries a label, no rank number and a paid flag in the feed. | In force |
| Confidence and limits next to the score | A thin profile looks as certain as a thick one. | Confidence is computed separately and shown; each position lists what is missing or unverified and what to ask before outreach. | In force |
| Negative evidence counts | Complaints, lapsed licences and dead client sites rarely reach the profile. | Regulator complaints, judgments, lapsed licences, dead case-study links and review manipulation signals are recorded with their source and subtract points or close the gate. | Partly in force |
| Independent measurement | Everything comes from the vendor or from reviewers. | Vendar measures what it can without the vendor: speed and uptime of delivered sites, search visibility of named clients, response time to a quote request, licence re-checks. Each measurement is a dated claim with the instrument named. | Proposed |
| Closed-loop outcomes | Reviews are self-selected and arrive at the vendor's request. | Buyers who shortlisted through Vendar are asked at 30 and 90 days whether they hired, whether it delivered and whether it stayed on budget. The answers become the strongest claim class and feed a per-question hire-and-deliver rate. | Proposed |
| Cross-source consistency | Each platform shows its own copy of the facts. | Team size, founding year, address and prices are compared across the site, LinkedIn, directories and registries; a disagreement is shown and lowers confidence instead of being hidden. | Proposed |
| History you can read | A profile shows today; yesterday is gone. | Every profile keeps a change log and a score history per question; a page shows what changed since the last data date. | Proposed |
| Price bands per need and city | Hourly rate bands self-declared by the vendor. | Typical price bands per need and city are built from disclosed prices, quotes buyers report and reviews that name a price, and shown so a buyer knows what is normal before asking. | Partly in force |
| Right of reply, in public | Disputes are handled privately or not at all. | A vendor can dispute a claim or add evidence from its profile; the dispute, the evidence and the resolution are logged on the profile, and submitted evidence stays claimed until checked. | Proposed |
| Built for AI assistants | Pages for people; scraping for everyone else. | Every listing and profile has a JSON feed with the same numbers, citation guidance and required caveats, so an assistant can recommend and cite without reading marketing pages. | In force |
| Divergence from the incumbents, explained | Lists that cannot be compared. | For each question Vendar records where its top differs from the leading directories' top and publishes the reason: the evidence the others count that we do not, and the evidence we found that they lack. | Proposed |
7. What a position does and does not say
A position reflects documented work for the page's question, not the quality of an agency as such. Without a public case or a visible campaign the result of an agency's work cannot be known from outside. An agency with no public evidence is not worse: there is nothing to confirm it with, so it is listed below the ranking instead of being ranked low.
- About 20 of 100 points rest on what the vendor says about itself without an outside check: years in business, team size, retention, prices.
- The weights favour older, larger and local vendors: full marks at ten years and around 100 reviews, awards count toward market credibility, a vendor that works remotely gets 0.3 on geography, an undisclosed price gets 0.2.
- The weights 30/25/15/10/10/5/5 are set by judgment and are not yet calibrated on client outcomes.
- Reviews on directories are often collected by the agency itself.
- A claim's weight in the evidence criterion still follows its source address: a person who worked at or through the agency writing on their own site, or a third-party page retelling what the agency says, weighs as third-party there. The position's basis, its label, the penalty for no independent source and the page's independence count already treat them as the agency's word, as the profile does.
- Market credibility still counts recognition the agency reports about itself, at half weight; which pages count as the agency's own data is a list of hosts kept by hand.
- Claims are checked by AI models, not by people. Code confirms that the quoted words are on the page; it cannot confirm that the page itself tells the truth. A case study on an agency's own site stays the agency's word even when the check confirms it.
- Confidence is shown next to the score and does not change the order.
8. Next model version: AI-graded evidence
Drafted 2026-10-28. It becomes the ranking model only after calibration against a hand-labelled reference set. Until then 2026.11.9 stays in force and its rules are not changed retroactively.
A model grades each claim against the rubric and must quote its evidence. Code computes every score. A grade whose quote is not found verbatim in the stored source snapshot is discarded.
Semantic match of a claim to a need
Replaces strict keyword matching as the condition of admission.
- The model receives the claim, the fragment of the stored source snapshot and the rubric text, and answers in a fixed schema: documents_work, intent_match, the rubric elements covered, a verbatim supporting quote and a confidence.
- Code checks that the quote appears verbatim in the stored snapshot; if it does not, the judgment is rejected.
- Rubric terms stay as a fast pre-filter and as the explanation of a position, not as the admission condition.
- When the model and the terms disagree, in either direction, an arbiter model from another family reads the claim; while the two models disagree, the claim does not count for the need.
- Vendor text is data only: instructions inside pages are ignored, and an attempt to steer the model flags the profile.
Quality of a documented case
Each sub-criterion is graded 0 to 3, and every grade carries a verbatim quote.
- Client specificity: named, industry, size
- Baseline and result: before and after, numbers, period
- Scale of the result: by bands relative to the baseline, not bare percentages
- Link between work and result: what the agency did and how it produced the outcome
- Verifiability: public client, outside data, confirmation in a review
- Freshness is computed by code from the date, never by the model.
- Code turns the sub-grades into a quality multiplier of about 0.5 to 1.5 that scales the claim's weight inside Evidence for this query. The model never outputs a score.
- Three weak cases no longer reach the 30-point cap; only strong, quantified results do, so the order among strong agencies is set by evidence rather than by secondary criteria.
- Reproducibility: a fixed model and prompt version are recorded on every grade; grades are recomputed only when the methodology version changes.
- Calibration: a reference set of 50 to 100 hand-labelled claims; a new model or prompt goes live only when its agreement with the reference set is above the threshold.
- Two models from different families decide what moves positions; code grounds every quote they rely on. A human can overturn any decision, and the overturn is logged.
- Transparency: sub-grades with their quotes on the vendor page; when the model is in force, this page names the model and prompt version used.
- Cost: one grade per claim, cached by the content hash of the source.
9. Who is eligible for a listing (B2B track, in force)
- The vendor lists the page's service on its profile.
- The vendor serves the page's market: headquarters or a listed service region in the city. When a need page has fewer than ten ranked vendors from the city, agencies from other cities join it only if they are headquartered in the same state or have published work (a case study, a project, a client review) that names the city or its state. A line saying a company serves the whole country is not enough. They are ranked by the same score, geography counts 0.3 for them, and each position says where the agency works from. Mobile development is bought remotely, so its city pages take agencies from anywhere in the United States under the same condition: documented work for the need. The same holds for SEO needs bought remotely: international, enterprise, consulting, recovery, marketplaces, startups, B2B SaaS, ecommerce, AI search visibility, audits, technical and content-led SEO. Local SEO, small business, lead generation and high-ticket services stay with agencies from the city and its state.
- A need page that still has fewer than ten ranked agencies after that is filled from the city's own agencies whose sources say they offer the need but document no project for it yet. They always stand below every agency with documented work, are labelled "Offers this service; no documented project on file yet", carry lower confidence, and do not count toward the threshold that opens a page to search engines.
- For an intent page, at least one claim on file supports the intent under the page's rubric, and at least one of those claims has a source address outside Vendar's own profile fields; the agency's own site counts. Our own service listings and profile tags describe the intent but document nothing, so they add weight to a vendor that already has outside evidence and never earn a rank on their own. Vendors without outside evidence appear below the ranking, unranked, never padded into it.
- For an intent page, at least one of those claims documents work the vendor has done: a case study, a project or portfolio entry, or a client review. Service pages, price lists, positioning statements and directory listings say what a vendor offers, not what it did; they add weight to a vendor that already has documented work and never earn a documented position. They can only fill a page that is short of ten, below every documented position and labelled as such (see below).
- A position may rest on case studies the agency published on its own site. Own published case studies count toward eligibility; independent confirmation adds to the score and spares the 3-point penalty for no independent source. Such a position is labelled "Agency's own published cases" until an independent source documents the same kind of work.
- Filtered views (an entry price, a pricing model, a team size) are not rankings of their own: they show agencies ranked for at least one need in the city, each with the score of its best need, and keep only those that meet the condition. An agency that does not publish the fact is listed below with the reason.
- Draft and needs-data profiles are never listed.
The B2C eligibility rules, including the licence gate, are in the B2C track card above and in the machine feed.
10. The B2B score: seven criteria, one formula
Evidence for this query
30 pointsClaims on file that support the page's exact service and intent, weighted by who made them and whether they were checked.
- Third-party claims (client reviews, press, awards) that name the intent: 1.0 verified, 0.8 claimed
- First-party claims that name the intent: the vendor's own site and platforms where the vendor writes about itself (partner directories and portals, case-study hosting, paid release wires): 0.8 verified, 0.6 claimed
- Directory listings that only name the service: 0.4, with no bonuses
- Profile signals (listed services, best-fit tags, editorial use cases): 0.25; a claim whose speaker is unknown: 0.3
- Bonus 0.15 when the claim uses the rubric's strongest terms, and 0.15 when it carries a number
- Three fully weighted claims reach the maximum; the total is capped at 30 points
Fit to this query
25 pointsHow closely the vendor matches the market, the budget band and the client type the page is about.
- Geography (half of the criterion): headquarters in the city 1.0, city listed as a service region 0.75, an agency from another city admitted by the regional rule 0.3
- Budget (30%): monthly floor disclosed and inside the intent's band 1.0, disclosed but above it 0.5, only a project floor or hourly rate 0.6, nothing disclosed 0.2
- Client type (20%): profile names the intent's buyer types 1.0, adjacent wording 0.7, otherwise 0.3
Capability depth
15 pointsHow much of what the vendor does is this service and this intent, as opposed to a side offer.
- Share of the vendor's listed services that belong to the service family (half)
- Share of the vendor's evidence that supports the intent, scaled so that 40% of claims counts as full depth (half)
Reliability
10 pointsOperating history and retention where they are documented.
- Years in business from the founding year, full marks at ten years; an unknown founding year counts as 0.3 (60%)
- Client retention rate where documented, otherwise a neutral 0.3 (40%)
Transparency
10 pointsHow much decision-relevant information the vendor discloses.
- Any pricing signal disclosed (50%)
- Team size disclosed (25%)
- Workflow, onboarding or reporting cadence documented (25%)
Reputation
5 pointsClient review volume and rating on open review platforms: Google, Trustpilot and BBB. Paid agency directories are not counted.
- A platform counts only with a rating and 10 or more reviews, the same threshold as the profile; a platform with fewer reviews is shown as too few to count and adds nothing
- Average rating out of five across the counted platforms, weighted by review count and discounted by their combined review volume (17 reviews count for about 60% of the volume factor, 100 reviews for 100%)
Market credibility
5 pointsRecognition on file with a source: awards, certifications, partner programs and media mentions, weighted by who confirms it.
- Awards, certifications, partner programs, third-party validations and media mentions named in the claims on file, each counted once; one claim can name several ("Meta Business and Google Featured Partner" counts as two)
- Confirmed recognition counts 1: its source is the program's own partner directory, the Inc. 5000 list, the award's site or independent press. Claimed recognition counts 0.5: its source is the agency's own site or a page where the agency enters its own data, such as a website builder profile, a self-listed agency directory or a press release wire
- Full marks at four, and only with at least two confirmed; with fewer, the criterion stops at three quarters
- Only recognition with a source address counts, and not when its source page does not support it; an award or listing from a paid agency directory never counts
11. Penalties
| Penalty | Points | When it applies |
|---|---|---|
| Profile flagged for review | -5 | An editor has flagged the profile; the flag is shown and the score is reduced until it is cleared. |
| Evidence not re-checked in 18 months | -5 | No claim on file has been checked against its source in the last 18 months, or none has been checked at all. Profile verification, which agencies can pay for, does not count here. |
| No independent source on file | -3 | Minus 3 when no independent source describes the agency's work for this page's need; the agency's own published case studies do not lift it. Who wrote the text decides, not the address: a client, the press or another independent writer counts, a page retelling what the agency says does not, and repeats of one case count once. |
| Profile below decision-ready status | -2 | The profile has not completed the decision-ready review. |
The B2C track adds two penalties: an unresolved regulator complaint or judgment (-10) and signs of review manipulation (-5). See the B2C card above.
12. Confidence is separate from the score
Rule
Confidence is shown next to the score and never mixed into it. Profile completeness and source reliability as recorded on the profile, then capped by what this position rests on: without an independent source for the question, or on a single documented claim, it is at most moderate; with no documented work for the question, it is lower.
Labels
- High confidence: confidence score 0.85 and above.
- Moderate confidence: 0.65 to 0.85.
- Lower confidence: below 0.65 or not disclosed.
13. How the order is set
- Rank order is the score, highest first. Ties break on evidence points, then on the number of third-party claims, then alphabetically.
- There is no hidden ordering: no hand-curated slug lists and no paid boost in the sort on any listing.
- If an editor overrides a position, the position is labelled as an editorial placement with a written reason and ordering_basis set to editorial. No such placements exist on evidence-ranked pages today.
- Every ranked position explains itself from its score components and cites the claims behind them, with their sources.
14. Evidence rules
- Paid agency directories that sell placement are not a source. Their reviews are not cited, never support a position and do not count toward reputation; a profile may show how many reviews such directories hold as one unnamed figure, for context only. Ratings come from open review platforms: Google, Trustpilot and BBB.
- Provenance is recorded per claim: third_party, first_party, internal (a Vendar profile field) or unknown.
- Verification status is recorded per claim: verified, claimed, internal_signal or unverified.
- Every record on file has a role: a claim states a fact; a source reference is a link without a statement (an official site, a directory profile); a research note is Vendar's own remark. Only claims count as evidence. A source reference never earns weight, however its title reads, and a research note never reaches a page or a feed.
- Every claim with a source URL is checked against that page by an AI reader and the result is recorded: confirmed, partly confirmed, not supported, or page unreachable. A claim the page does not support stops counting toward any position until a later check confirms it; a claim whose page cannot be read still counts and says so.
- Third-party claims outweigh first-party claims; profile fields are the weakest signal and are labelled as such.
- A position may rest on case studies the agency published on its own site; such a position is admitted and labelled "Agency's own case studies only" on the listing, the profile and the feed (evidence_basis: own_published), carries the 3-point penalty for no independent source, until an independent source documents the same kind of work. Its confidence is at most moderate.
- In the evidence criterion, a claim's weight follows its source address: the vendor's own site and platforms where the vendor publishes about itself (platform partner directories and portals, case-study hosting sites, paid release wires) are the vendor's own word, not an independent source, whatever domain they sit on. Client reviews on a partner directory stay third-party. Whether a position counts as independently confirmed (its basis, its label, the penalty for no independent source and the page's count) is decided by who wrote the text, the profile's rule below, since model 2026.11.9.
- On the profile and in its feed the rule is stricter: who is speaking is decided by who wrote the text. Every claim falls into one of four classes: independent (a client, the press, a rating written by a third party); linked to the agency (a person who worked at or through it, writing in the first person); a listing the agency writes (the platforms above, or another page retelling what the agency says); and the agency itself. Only the first class counts as independent there (section 18).
- A claim counts for an intent only when the rubric's terms appear in the claim or its source. The rubric is published on the page.
- Claims that only describe the company's location or history are not intent evidence.
- Each claim is marked by whether it documents work (describes_work in the machine feed): case studies, projects, portfolio entries and client reviews do; service pages, pricing, positioning statements, review-platform profiles and employer reviews do not.
- Claims from the same source count with diminishing weight: each further claim adds half as much as the one before it. Everything an agency says about itself, on any page, is one source; outside documents are told apart by the site that published them. The number of reviews on one platform, or of pages on an agency's own site, cannot carry a position.
- Evidence documented on two or more different independent sites earns a small bonus over the same amount of evidence from one site.
- Review platforms that compete with Vendar stay in the machine feed with their source URL, because a claim must be checkable, but pages do not name or link them; where other sources exist, they are listed first.
Verified means Vendar confirmed the company, its service and its market on official sources and stamped the profile with the date. A participation status or a partner relationship does not make a profile verified.
15. Need rubrics in force
A need page is ranked by this model only when its rubric is published here. The rubric text appears on the page and in the feed word for word.
SEO · Local SEO
rubric 2026.09 · minimum 1 claimA claim counts as local-SEO evidence when a source on file names local search work: Google Business Profile or map-pack results, service-area or local landing pages, local citations, multi-location programs, or reviews that describe local rankings. Official case studies and third-party reviews carry the most weight. A Vendar service listing or profile tag is our own label, not a document: it adds a weak signal to a vendor already carrying outside evidence, and on its own it does not earn a rank. A service share an agency sets on its own directory profile, such as a Clutch focus percentage, is the agency's claim about itself and weighs as little as a service listing. Agencies with no outside claim on file are listed below the ranking, not in it.
Terms that qualify a claim: local seo, local search, local pack, map pack, google business profile, google my business, gmb, gbp, google maps, service-area page, service area page, service-area pages, service area pages, local landing page, local landing pages, local citations, near me, multi-location, multi location, local rankings, local visibility, local lead generation. Budget band for fit: up to $6,000/month.
SEO · Technical SEO
rubric 2026.09 · minimum 1 claimA claim counts as technical-SEO evidence when a source on file names engineering-side search work: crawl budget and indexation fixes, Core Web Vitals or page-speed results, site architecture and internal linking, canonicalisation, structured data, JavaScript rendering, log-file analysis, or a platform migration that preserved rankings. Client reviews describing an audit that shipped fixes carry the most weight. A Vendar service listing or profile tag is our own label and never earns a rank on its own.
Terms that qualify a claim: technical seo, core web vitals, page speed, pagespeed, crawl budget, crawlability, indexation, indexing issues, site architecture, internal linking, canonical, structured data, schema markup, javascript rendering, log file, site migration, seo migration, robots.txt, xml sitemap, site speed, lighthouse.
SEO · Ecommerce SEO
rubric 2026.09 · minimum 1 claimA claim counts as ecommerce-SEO evidence when a source on file names store-side search work: category or product-page optimisation, faceted navigation and parameter handling, product schema, a named commerce platform (Shopify, Magento, BigCommerce, WooCommerce), catalogue migrations, or organic revenue and transaction results. Reviews quantifying revenue or transactions carry the most weight. A Vendar service listing or profile tag never earns a rank on its own.
Terms that qualify a claim: ecommerce seo, e-commerce seo, shopify, magento, bigcommerce, woocommerce, product page, product pages, category page, category pages, faceted navigation, product schema, catalog, catalogue, online store, organic revenue, transactions, dtc, direct-to-consumer.
SEO · B2B SaaS SEO
rubric 2026.09 · minimum 1 claimA claim counts as B2B SaaS evidence when a source on file names work for a software business: qualified pipeline, demo or trial signups from organic, product-led or bottom-of-funnel content, comparison and alternative pages, integration or use-case pages, or a named SaaS client with a result. Reviews naming pipeline or signups carry the most weight. A Vendar service listing or profile tag never earns a rank on its own.
Terms that qualify a claim: b2b saas, saas seo, b2b seo, demo request, demo requests, free trial, trial signups, pipeline, mqls, sqls, qualified leads, product-led, bottom of funnel, comparison page, alternatives page, integration pages, use case pages, software company.
SEO · AI search visibility
rubric 2026.09 · minimum 1 claimA claim counts as AI-search evidence when a source on file names work aimed at generative answers: appearing or being cited in AI Overviews, ChatGPT, Perplexity or Claude answers, answer-engine or generative-engine optimisation, entity and knowledge-graph work, llms.txt and machine-readable feeds, or a measured change in AI citations. Reviews or case studies naming a specific assistant carry the most weight; a service page that merely offers the term does not. A Vendar service listing or profile tag never earns a rank on its own.
Terms that qualify a claim: ai search, ai overview, ai overviews, answer engine, answer-engine optimization, generative engine optimization, geo optimization, llm visibility, llms.txt, chatgpt, perplexity, claude, copilot, ai citations, ai assistant, entity optimization, knowledge graph, ai visibility.
SEO · SEO Audit
rubric 2026.09.1 · minimum 1 claimA claim counts as SEO-audit evidence when a source on file describes an audit the agency delivered: a technical, site, content or backlink audit, what it found, and ideally which recommendations were implemented and what changed afterwards. Case studies and reviews in which the audit led to shipped fixes and a measured result carry the most weight. Accessibility and paid-search audits are different services and do not count. A Vendar service listing or profile tag is our own label and never earns a rank on its own.
Terms that qualify a claim: seo audit, seo audits, technical audit, technical audits, technical seo audit, technical seo audits, site audit, site audits, website audit, website audits, content audit, content audits, backlink audit, backlink audits, link audit, audit findings, audit recommendations.
SEO · Small Business
rubric 2026.09 · minimum 1 claimA claim counts as small-business evidence when a source on file describes SEO work for a small, owner-led or single-location business: who the client was, what the agency did, and ideally what changed in rankings, calls or leads. Client reviews from small-business owners and case studies with a named small client carry the most weight. A share of small-business clients that an agency sets on its own directory profile is the agency's description of itself and weighs as little as a service listing. A Vendar service listing or profile tag never earns a rank on its own.
Terms that qualify a claim: small business, small businesses, small-business, smb, smbs, family-owned, family owned, owner-operated, owner-led, single-location, single location. Budget band for fit: up to $4,000/month.
SEO · SEO for Startups
rubric 2026.09 · minimum 1 claimA claim counts as startup evidence when a source on file describes SEO work for an early-stage company: a seed or Series A–B business, a venture-backed product, or a launch built from little or no organic traffic, and ideally what changed in traffic, signups or pipeline. Case studies and reviews naming the stage or funding of the client carry the most weight. A share of startup clients that an agency sets on its own directory profile is the agency's description of itself and weighs as little as a service listing. A Vendar service listing or profile tag never earns a rank on its own.
Terms that qualify a claim: startup, startups, start-up, early-stage, early stage, seed-stage, seed round, pre-seed, series a, series b, venture-backed, vc-backed, y combinator, techstars. Budget band for fit: up to $7,000/month.
SEO · SEO Consulting
rubric 2026.09 · minimum 1 claimA claim counts as SEO-consulting evidence when a source on file describes advisory work rather than done-for-you execution: consulting or coaching for a client's in-house team, SEO training and workshops, fractional or interim SEO leadership, or a strategy and roadmap the client's own team implemented. Reviews from the people who were advised carry the most weight. A Vendar service listing or profile tag never earns a rank on its own.
Terms that qualify a claim: seo consulting, seo consultant, seo consultants, consulting engagement, seo advisor, seo advisory, advisory engagement, fractional seo, in-house team, in-house seo, seo training, team training, trained our team, trained the team, seo workshop, seo workshops, seo coaching, seo roadmap.
SEO · Enterprise SEO
rubric 2026.11 · minimum 1 claimA claim counts as enterprise-SEO evidence when a source on file names search work at enterprise scale: a named large company or Fortune 500 client, sites with very many pages, programs across several markets or brands, or governance with in-house teams and stakeholders. Case studies naming the client and the scale carry the most weight. Small-business and single-location local work does not count here.
Terms that qualify a claim: enterprise seo, enterprise, fortune 500, fortune 100, global brand, multi-market, multi-brand, large-scale, large site, millions of pages, thousands of pages, governance, stakeholders, in-house team, publicly traded.
SEO · Content-led SEO
rubric 2026.11.1 · minimum 1 claimA claim counts as content-led SEO evidence when a source on file names content work done for search: a content strategy, topic clusters or pillar pages, an editorial calendar, articles or guides produced at volume, or organic traffic growth tied to content. A result that names the content and the traffic it earned carries the most weight. Local listings work does not count here.
Terms that qualify a claim: content strategy, content marketing, content seo, seo content, editorial calendar, editorial program, topic cluster, topic clusters, pillar page, pillar pages, blog posts, blog program, content production, content hub, thought leadership content, articles per month.
SEO · International SEO
rubric 2026.11.1 · minimum 1 claimA claim counts as international-SEO evidence when a source on file names search work across countries or languages: hreflang or international site structure, localized or translated sites, programs in several named markets, or results reported per country or language. A case naming the markets and the outcome carries the most weight. Work in one city or one country does not count here.
Terms that qualify a claim: international seo, multilingual seo, multilingual, multi-language, multiple languages, hreflang, localization, localisation, localized sites, translated sites, global seo, multiple countries, multi-country, international markets, country sites.
SEO · SEO Recovery
rubric 2026.11 · minimum 1 claimA claim counts as SEO-recovery evidence when a source on file names a recovery the agency delivered: from a manual action or penalty, a core or helpful-content update, a migration that lost rankings, deindexation, or a traffic drop, ideally with the before and after. A case that names what was lost and what came back carries the most weight.
Terms that qualify a claim: penalty, manual action, algorithm update, core update, helpful content, traffic drop, lost traffic, lost rankings, recovered, recovery, recover, deindexed, reindexed, disavow, toxic links, migration loss.
SEO · SEO for Lead Generation
rubric 2026.11.1 · minimum 1 claimA claim counts as lead-generation evidence when a source on file names inquiries won through organic search: qualified leads, phone calls, form submissions, booked consultations or meetings, or pipeline and sales opportunities. A result with the number of leads and the period carries the most weight. Traffic alone, without leads, does not count here.
Terms that qualify a claim: leads, lead generation, qualified leads, inbound leads, form submissions, form fills, phone calls, booked calls, booked appointments, booked consultations, pipeline, inquiries, enquiries, sales opportunities, mqls, sqls.
SEO · SEO for High-Ticket Services
rubric 2026.11 · minimum 1 claimA claim counts as high-ticket evidence when a source on file names search work for a business that sells high-value services: a law firm or attorney, a medical, dental or cosmetic practice, financial advisory or wealth management, real estate, or premium professional and B2B services. A result in cases, consultations or revenue for such a client carries the most weight. Ecommerce and low-ticket retail do not count here.
Terms that qualify a claim: law firm, attorney, lawyer, legal, personal injury, medical, dental, dentist, plastic surgery, cosmetic, med spa, clinic, healthcare, financial advisor, wealth management, real estate, luxury, high-ticket, high-value, consultations.
SEO · SEO for Marketplaces
rubric 2026.11 · minimum 1 claimA claim counts as marketplace-SEO evidence when a source on file names search work for a two-sided marketplace, a directory, classifieds or another listing platform: programmatic or templated pages at scale, inventory or listing pages, faceted navigation across many listings, or indexation of user-generated pages. A case naming the platform and the scale carries the most weight.
Terms that qualify a claim: marketplace, marketplaces, directory, listing pages, listings, inventory pages, programmatic seo, programmatic, classifieds, user-generated, templated pages, platform seo, two-sided.
SEO · SEO Agencies within a $6,000 Monthly Budget
rubric 2026.11 · minimum 1 claimThis page is not a ranking of its own. It shows the agencies ranked for at least one published SEO need in this city, each with the score of its best documented need, the same score as on its profile, and filters who is shown. Shown: agencies whose published monthly or project entry price is $6,000 or less. An agency that does not publish an entry price is listed below the ranking, because the budget cannot be checked.
Terms that qualify a claim: seo, organic, organic traffic, organic search, search rankings, rankings, search engine, search visibility, keywords, first page. Budget band for fit: up to $6,000/month.
SEO · SEO Agencies on a Monthly Retainer
rubric 2026.11 · minimum 1 claimThis page is not a ranking of its own. It shows the agencies ranked for at least one published SEO need in this city, each with the score of its best documented need, the same score as on its profile, and filters who is shown. Shown: agencies that state a retainer or mixed pricing model. An agency that does not state its pricing model is listed below the ranking.
Terms that qualify a claim: seo, organic, organic traffic, organic search, search rankings, rankings, search engine, search visibility, keywords, first page.
SEO · SEO Agencies with an In-house Team
rubric 2026.11 · minimum 1 claimThis page is not a ranking of its own. It shows the agencies ranked for at least one published SEO need in this city, each with the score of its best documented need, the same score as on its profile, and filters who is shown. Shown: agencies whose stated team size is ten people or more, so the work is not carried by one freelancer. An agency that does not state its team size is listed below the ranking.
Terms that qualify a claim: seo, organic, organic traffic, organic search, search rankings, rankings, search engine, search visibility, keywords, first page.
Mobile Development · iOS App Development
rubric 2026.11 · minimum 1 claimA claim counts as iOS evidence when a source on file names an iPhone, iPad or Apple Watch app the company built or rebuilt, native work in Swift, SwiftUI or Objective-C, or an App Store or TestFlight release, ideally with the client named. An app store listing counts only when it names the company as the developer; user ratings of an app are not client reviews of the company. App store optimization is marketing and does not count.
Terms that qualify a claim: ios app, ios apps, ios application, iphone app, ipad app, apple watch app, swift, swiftui, objective-c, app store, testflight, ios development, native ios.
Mobile Development · Android App Development
rubric 2026.11 · minimum 1 claimA claim counts as Android evidence when a source on file names an Android phone, tablet or Wear OS app the company built or rebuilt, native work in Kotlin, Java for Android or Jetpack Compose, or a Google Play release, ideally with the client named. A Google Play listing counts only when it names the company as the developer; user ratings of an app are not client reviews of the company.
Terms that qualify a claim: android app, android apps, android application, kotlin, jetpack compose, google play, play store, android development, native android, wear os.
Mobile Development · React Native Development
rubric 2026.11 · minimum 1 claimA claim counts as React Native evidence when a source on file names an app the company built, rebuilt or maintains in React Native or Expo, ideally with the client and the platforms it shipped on. A technology listed on the company's services page without a project does not count.
Terms that qualify a claim: react native, expo.
Mobile Development · Flutter Development
rubric 2026.11 · minimum 1 claimA claim counts as Flutter evidence when a source on file names an app the company built, rebuilt or maintains in Flutter or FlutterFlow, ideally with the client and the platforms it shipped on. A technology listed on the company's services page without a project does not count.
Terms that qualify a claim: flutter, flutterflow.
Mobile Development · Mobile MVP Development
rubric 2026.11 · minimum 1 claimA claim counts as MVP evidence when a source on file names a minimum viable product, prototype, proof of concept or first release of an app the company built for a startup or a new product, ideally with the launch, the funding round or the first users that followed. Marketing work for a startup (SEO, ads) does not count.
Terms that qualify a claim: mvp, minimum viable product, prototype, proof of concept, first version, app launch, launched the app, pre-seed, seed round, startup.
Mobile Development · Mobile App Modernization
rubric 2026.11 · minimum 1 claimA claim counts as modernization evidence when a source on file names an existing app the company took over and rebuilt, rewrote, migrated to a new stack or framework, re-architected or upgraded, ideally with what changed after (stability, ratings, speed, cost). A website redesign does not count.
Terms that qualify a claim: modernization, modernize, modernized, rebuild, rebuilt, rewrite, rewrote, re-architect, migrated, migration, legacy app, legacy system, refactor, refactored, took over the app.
Mobile Development · Mobile Backend and API Integration
rubric 2026.11 · minimum 1 claimA claim counts as backend and API evidence when a source on file names a backend, REST or GraphQL API, cloud setup (Firebase, AWS, Azure) or an integration with another system (payments, CRM, EHR, ERP, hardware) the company built for an app, ideally with the client named.
Terms that qualify a claim: api, apis, rest api, graphql, backend, back-end, firebase, aws, azure, microservices, integration, integrations, integrated, stripe, payment integration, real-time, bluetooth.
City pages, hubs and intent pages that do not have a published rubric yet. These pages are still ordered by the profile decision score, which averages two profile-completeness checklists, plus route-fit boosts and, on some routes, a curated order. They are being migrated to the evidence-weighted model intent by intent; a page states which model it uses.
16. Paid visibility, dates and indexing
Paid visibility
Paid visibility never changes the order of a listing. The sort key carried a package boost of up to 0.06 until 2026-09-12; it was removed, because the pages state that no paid placement affects rank and the packages sell badges, surfaces and a report, never a position. If featured placements are sold, they carry a Featured label, no rank number, and a paid flag in the machine feed. No vendor pays for visibility today.
Dates
Every date shown is a data date. A listing's updated date is the newest verification date among its ranked vendors and the evidence claims their ranks rest on; the same value appears in the page text, the structured data and the machine feed.
Indexing gate
An intent page is offered to search engines when its ranked list has at least five evidence-backed vendors (three when the city has fewer than twenty vendors for the service). Only agencies from the city itself (headquarters or a listed service region) with documented work count toward it: agencies filled in from other cities and agencies that only offer the need do not. A page whose list rests mostly on agencies from other cities (fewer local agencies with documented work than that threshold, or local agencies under half of the list) is titled “agencies serving” the city instead of “best agencies in” it and says in its first line how many are local and where the rest serve it from; below the threshold it stays on the city page and in its feed and is kept out of search engines until enough local agencies document the work; otherwise the same national list would stand behind six city pages. Below the threshold even with outside agencies the page stays published with noindex, keeps its URL, links and feed, and is left out of the city page's list of rankings. A page whose top ten repeats 70% or more of another need's top ten in the same city, or has the same top three, also stays noindex while the stronger of the two is offered; it opens when the lists differ. The same holds across cities: of the pages for one need whose top tens share 70% or more, only the one with the most local agencies is offered to search engines.
17. What every ranked position shows
18. From positions to a profile verdict
Profile decision · 2026.11.7-profile
In forceA vendor profile does not have a score of its own. It carries the score of its best position, a verdict built from that position and from how far its evidence can be trusted, and two profile scores that say what is on file and how much of it checks out. The profile page and the profile feed read the same evidence ledger, so their numbers match.
One evidence ledger
- Every claim on file is counted once, in one ledger, by who wrote it: independent, linked to the agency, a listing the agency writes, or the agency itself.
- Documented work is counted separately from offers. The headline numbers are always about work: "71 cases of work, 23 from outside sources" means 23 of those 71.
- The speaker counts add up to the total, and independent work is always a subset of work; a profile where they do not is not published as decision-ready.
- Since model 2026.11.9 a position reads who wrote the text for its basis and for the penalty for no independent source; claim weights in the evidence criterion still follow the source address (section 14). Moving the weights onto the ledger is the next step.
Score and verdict
The profile score is the score of its best position: the rank and the question are named next to it. A vendor with no position has no score and says why.
- Recommended
- Ranks in the top three for at least one need, the trust score is 60 or more, and the record carries what a decision needs: a published price with a unit (monthly floor, project minimum or hourly rate), both fit blocks (who it suits and who should look elsewhere, written by an editor or derived from the record) and a confidence value.
- Insufficient data
- Would be Recommended on rank and trust, but one of the three requirements is missing. The verdict names what is missing; the agency can add it through the claim form, and the verdict updates with the data. Since model 2026.11.6.
- Shortlist with checks
- Ranks in the top three for at least one need, but the trust score is below 60: much of the evidence is the vendor's own word or not yet checked.
- Ranked
- Admitted to a ranking on documented work, outside the top three for every question it appears in.
- Not ranked yet
- No published ranking admits the vendor on documented work yet; the profile shows what is on file.
The verdict text names two things separately, because they are different: whether the claims match the sources they cite (the check against the source page), and who those sources are (how many work examples come from an independent source, with the ledger's numbers). A claim can match its source exactly and still be the agency's own word. For example: "Ranks in the top three for at least one need. Most of its claims match the sources they cite. 1 of its 18 work examples comes from an independent source."
Profile scores
- Completeness (out of 100): identity 20, fit 15, commercial terms 20, documented work 30 (cases naming a client and a result, target five; work claims overall, target ten), reputation on two or more platforms 15.
- Trust (out of 100): independent sources among work claims 30 (full marks at 40%), claims checked against their page 25, numbers backed by a verbatim quote 20, claims seen within a year 15, claims not contradicted by their page 10.
- Neither score moves a position. The trust score decides between "Recommended" and "Shortlist with checks"; the three data requirements decide between "Recommended" and "Insufficient data".
- Confidence: an editor's rating when there is one; otherwise it is rated from the evidence on file (posture and confirmed quotes) with the same words and the same caps, and the feed says which (decision.confidence.source).
Evidence posture and caveats
- Strong: at least five independent work claims, at least a quarter of all documented work, four or more sites and ten or more claims. Confidence is capped at 0.95.
- Adequate: independent proof of work exists but falls short of strong. Capped at 0.78.
- Thin: no independent proof of work, fewer than two sites or fewer than three claims. Capped at 0.62.
- Weak: no source-backed claim at all. Capped at 0.45.
- While independent work is under half of all documented work, the profile, its risks and the feed's caveats carry the same sentence with the numbers, for example "23 of 71 documented work examples come from independent sources". A strong posture and a caveat about few independent sources cannot appear together.
Risks
Risks derived from data come first: the share of independent work, results with a number but no year, and the weakest criterion of the best position. Editorial risks follow them. A line that contradicts the data, such as "buyers who need public pricing" next to a published price, is dropped.
Consistency between the page and the feed
- Every profile feed is checked against its page each time it is built: counts, speakers, posture, caveats, the questions to ask before signing, pricing statements, and the absence of internal fields.
- A blocking disagreement sets the profile's status to needs_review in the feed and lists the reasons under quality.consistency, so an assistant sees the problem instead of a confident verdict.
- The consistency status does not change a position in a listing today; the listing penalty for a flagged profile applies only when an editor flags it.
See it applied
The Los Angeles local SEO listing is ranked by the B2B track: the rubric is printed on the page, each agency shows its points per criterion and the claims behind them, and the agencies without local-SEO evidence are listed unranked below the ranking.