How we score agencies

Every agency in the index gets a score out of 100. The score is computed from the record on every build, never typed in by hand, and this page describes the exact formula that runs. Last updated 12 Nov 2026.

The five components

ComponentMaxSource
Third-party reviews30Clutch, G2 and Google profiles with a URL and a date we read them.
Founder credibility25Years in search, prior roles at named companies, published work, speaking, prior exits. Each founder needs a LinkedIn URL.
Company track record20Years in business, named case studies with a stated result, certifications and awards, each with a URL.
Own AI visibility15The agency's own domain tested in ChatGPT, Perplexity, Gemini and Google AI Overviews for its category queries.
Transparency10Public pricing, a named team, a physical address and a contact route we have confirmed reaches a person.

Third-party reviews, 30 points

Each platform gives a signal between 0 and 1: the rating out of five, multiplied by how much of a full review base the agency has. The review base is on a log scale and saturates at 50 reviews, so the first ten reviews say the most and the five hundredth adds almost nothing. A rating of 4.9 from 60 reviews is stronger evidence than 5.0 from 3, and a plain average would say the opposite.

The signals are weighted by how hard the platform is to game: Clutch 1, G2 0.8, Google 0.6. Clutch interviews reviewers; anyone with a Google account can post a Google review. The weighted sum is divided by 1.8 and clamped, which means a full-strength Clutch profile alone earns about 17 of the 30 points, Clutch plus G2 at full strength earn all of them, and a third platform adds nothing once the component is full. No single source can carry the component.

Only reviews about search work count. A full-service agency with twenty website builds and one SEO project has twenty-one Clutch reviews and one that says anything about search. Clutch tags each review with the service it was for and shows a service focus chart, so for Clutch we count the reviews tagged SEO or GEO, or, where the reviews cannot be counted one by one, the review base multiplied by the search share of the focus chart. The profile shows both numbers ("12 search of 48"). G2 and Google carry no service tags and count in full, which is one reason they weigh less.

We cite the rating, the count, the URL and the date we read it. We never copy review text, never average platform scores into a rating of our own, and never publish review markup for reviews collected elsewhere.

Founder credibility, 25 points

For each founder with a LinkedIn URL: up to 10 points for years in search, capped at 15 years; up to 6 for prior roles at named companies or agencies, capped at 3; up to 9 for notable items such as published work, conference talks and prior exits, capped at 3. The strongest founder counts in full and the second strongest adds 20% of their own score. A founder without a public profile does not count.

Company track record, 20 points

Up to 6 points for years in business, capped at 10; up to 9 for named case studies with a URL and a stated result, capped at 3; up to 2.5 for certifications and 2.5 for awards, each capped at two and each needing a URL. A case study that does not name the client, or names a client but states no result, does not count.

Own AI visibility, 15 points

An agency that sells visibility in AI answers should have some. The questions are this index's own ranking pages: for every service, every service and industry, and every service and region page, we ask an AI engine which agencies it would recommend and to name their websites. Each agency is then checked against the answers for the pages it appears on, and its score is the share of those answers that name it or its domain. That share maps straight onto the component: named in 60% of its answers is 9 points. The test date and the question count are shown on the profile.

The current figures come from 203 questions asked on 8 Oct 2026 of ChatGPT (real answer via DataForSEO) and Perplexity (sonar via DataForSEO). Answers are cached, so a re-run only asks what is new.

Transparency, 10 points

4 points for pricing published on the agency's own site, with a link; 2 for a named team; 2 for a published address; 2 for a contact route we have confirmed reaches a person. These are the things a buyer needs before a first call and the things an agency can fix in an afternoon.

The number on the page

The five components add up to a raw figure out of 100, and that raw breakdown is what every profile shows. The headline score is relative: the strongest agency in the index is set to 95, and every other agency's score is its raw points as a share of the leader's. Nobody scores 100 because the formula's maximum asks for evidence no agency has produced yet: full review bases on three platforms, a tested AI visibility figure, every proof in place. Scaling keeps the order exactly as the raw points put it; it only makes the number readable as "how close to the best in the index".

Colour zones follow the headline score: green from 70, amber from 40 to 69, grey below 40. Green means the record is close to the strongest evidence in the index. Grey usually means the record is thin, often because the agency has not claimed its profile and the public sources say little, rather than because anything is wrong with the agency.

The rules

  • Every proof needs a URL. No URL, no points. A claim we cannot open is a claim we cannot score.
  • Every component is clamped. A huge review base, a founder with thirty years, ten awards: none of them can push a component past its maximum, so no single strength dominates the score.
  • Unverified claims score zero. A review profile without a date we read it, a case study without a result, a founder without a profile: each counts for nothing.
  • The score is recomputed on every build. Nothing is stored by hand. Change the record and the score follows.
  • Ties go to the most recently verified agency. Two agencies on the same score are ordered by the date their record was last checked.
  • Every update is a recorded run. When scores change, the positions on every page are recorded with the date, and the site shows each agency's movement since the run before. The date of the last update is printed on every ranking page.
  • The headline score is scaled to the index leader. Raw points are shown on every profile; the headline number is the share of the leader's raw points, with the leader at 95.
  • Nothing can be bought. No agency pays to be listed, to be ranked, or to be placed. We do not take reviews on this site.

Listed, claimed, verified

Listed means we wrote the profile from public sources and the agency has not been in touch. Claimed means someone with an email address at the agency's domain has confirmed the facts and we are re-checking each proof. Verified means that re-check is complete and every proof on the profile was opened and matched. The status is shown on every profile and every ranking row, together with the date the record was last verified. We re-verify every agency at least every six months.

Ranking pages

A ranking page exists for every service, every service and industry, every service and region, and every service, industry and region that three or more agencies match. Below three we neither generate the page nor link to it. If a narrower page would list exactly the same agencies in the same order as a broader one above it, we generate only the broader page and redirect the narrower address to it, so no two pages on this site say the same thing.

Corrections

We do not invent facts and we write every profile from sources anyone can check. If an agency's situation has changed, the place to change it is the source: update the review profile, publish the case study, put the pricing on the site, then claim the profile and send the links. The listing follows the record.

Conflicts of interest

Agencies connected to this site's operator are listed under exactly the same rules and score from the same formula. The formula is published here so that anyone can check the result.

The weights on this page are read from the same file the scorer runs on, so the published formula and the one that produced today's scores for all 110 agencies cannot drift apart.