The FOUND Score: how it is worked out.
Score version found-score-v1.0. Five letters from the FOUND Standard, each 0 to 100, and the overall score is their mean. Public evidence counts fully. Same evidence, same score, every time. No model decides a number.
LiveA published, deterministic method: 0 to 100 for each letter and overall, with evidence levels that cap what a score can claim.FOUND Score
The calculation, in six lines
- Each letter has three or four parts, worth up to 100 points between them (the table below). Being named in AI answers about your category counts in Findable, Opinions and Nods, because it is the outcome the score exists to measure.
- Public evidence counts fully, whether our crawler read it on the page or live search found it: your site, review sites, analysts, awards, press, social following and answers from AI engines.
- Volumes (reviews, followers, stories, pages) are log-scaled, so the first 10 reviews are worth about as much as the next 90 and a small company sees each step move the number.
- Your own answers are marked self-reported and count for at most 8 points per letter.
- Evidence a run could not gather (a site that refused our reader, a search that failed) is estimated from the strongest live evidence, labelled, never counted as zero.
- A confidence label (High, Medium or Low) says how much of the score rests on evidence we read ourselves. It never lowers the score.
Calibrated so that a company that owns its category, such as Google, scores about 90 or more, a category leader such as CrowdStrike 75 to 88, and a tiny unknown site 5 to 20. The score is a diagnosis, not a forecast of revenue. Approved by the owner on 19 October 2026 as version 1.0; weights may be fine-tuned in a later version, and every result keeps the version it was scored under.
The seventeen parts
| Part | Points | What is measured |
|---|---|---|
| Open to AI and search crawlers | 10 | robots.txt rules for AI crawlers, a sitemap and llms.txt, read on the site. |
| Named in AI answers about your category | 50 | Share of buyer questions about your category (never naming you), asked live to answer engines with web search, whose answer names or cites you, counted as named ÷ (asked + 1) for a small sample. With no sample, estimated from web presence, at most 60%. |
| Third-party pages and stories about you | 40 | Third-party pages and earned stories found by live search, log-scaled, full at 60. |
| Part | Points | What is measured |
|---|---|---|
| Ratings and reviews on review sites | 50 | Per review family, rating quality (3.0 of 5 is nothing, 4.75 is full) times log-scaled volume (from 2, full at 20,000). Business-software sites count 1, Trustpilot 0.4; a figure from search counts 0.9 of one read on the listing page. When the listings found are weaker than AI answers suggest, estimated from the AI-answer share, at most 85%, labelled. |
| Recommended in AI answers | 15 | The same AI-answer share: the opinion buyers now read first. |
| Following of the brand and its leaders | 25 | Brand followers (full at 20 million, 15 points) and the chief executive’s (full at 10 million, 10 points), log-scaled from 50. |
| People talking about you | 10 | Posts by others naming you in 7 days, log-scaled, full at 300,000. |
| Part | Points | What is measured |
|---|---|---|
| Pages machines can read and quote | 15 | Structured data, title and description, and text that loads without scripts, read on the site. |
| Content worth citing | 25 | How many of 5 kinds the site has: blog or newsroom, customer stories, partners, academy or docs, events or community. |
| Picked up by others | 60 | Earned stories (full at 40, 35 points) and third-party pages (full at 30, 25 points), log-scaled. |
| Part | Points | What is measured |
|---|---|---|
| Analyst reports that name you | 35 | Distinct analyst reports (the analyst’s own page 1, another third party 0.8, your own page 0.6), log-scaled, full at 4. |
| Awards and lists | 20 | Distinct awards or lists, weighted the same way, log-scaled, full at 5. |
| Tier-1 press | 25 | Stories in tier-1 outlets, log-scaled, full at 10. |
| Treated as an authority by AI answers | 20 | The same AI-answer share: a nod from the source buyers consult. |
| Part | Points | What is measured |
|---|---|---|
| Security and trust published | 25 | security.txt, a trust page and named certifications, read on the site. |
| Evidence reachable | 10 | HTTPS and a quick response. |
| Independent proof | 65 | Customer stories on the site (15) and units of independent proof: review families, analyst and award recognition, public filings (2), independent sources (1 per 4, up to 5); 50 points, log-scaled, full at 10. |
Review sites: business buyers first
Business-software review sites (G2, Gartner Peer Insights, Capterra, GetApp, Software Advice, TrustRadius, PeerSpot, SourceForge, SoftwareReviews) count fully. Consumer sites such as Trustpilot count 0.4 as much, because their reviewers are rarely the buyer of business software. Capterra, GetApp and Software Advice share one pool of reviews, so they count once. A rating found by search counts 0.9 of the same rating read on the listing page; the newest page read wins, so the number does not drift between searches. The strongest site counts in full and each further site adds less, so a weak extra listing never lowers the score.
Confidence label
High when at least 70% of the points rest on evidence we read ourselves (your site, a listing page, an awarder’s own list, an answer engine’s reply), Medium from 45%, otherwise Low. Evidence from live search and estimates still count; the label says how much you can lean on the number.
The site checks underneath
Fourteen checks read your public home page, a few standard files and any analyst or award page your home page links to. None needs a third-party key. Each result lists what it saw and the address it read. Two more, the award-list checks, run where we know the company’s name: today, the companies in our peers store.
- HTTPS and page speed hints feed Accessibility. Speed is one request from one place, a hint and not a Lighthouse result.
- robots.txt for 11 AI crawlers (GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-SearchBot, Claude-User, PerplexityBot, Perplexity-User, Google-Extended, Applebot-Extended, CCBot), the sitemap and llms.txt feed Open to AI and search crawlers. Crawlers that fetch pages to answer a buyer count double; training crawlers count once, since blocking them can be a choice. llms.txt is a proposal and counts least.
- Structured data, title, description, headings and server-rendered text feed Pages machines can read and quote.
- Review-site listings (G2, Gartner Peer Insights, Capterra, TrustRadius, PeerSpot, Trustpilot and others): found by live search, then read on the listing page by our crawler where the site allows it (robots.txt obeyed, no login, no bot check worked around). A figure read on the page wins; otherwise the search figure counts, labelled “from search, page not read”.
- security.txt, a trust page and named certifications feed Security and trust published. A name on a page is a mention, not a certificate.
- Analyst reports and awards: the awarder’s or analyst’s own published page counts most, another third party’s report of it 0.8, the company’s own page 0.6. Awards for people and programmes with no current public list are not counted.
Not finding a company on a list never lowers its score: absence from a list is weak evidence, since most programmes need an entry. Answer-engine questions are asked with live search (Gemini and Perplexity grounded); without a key the part is estimated and labelled.
A worked example
Example Company, a funded startup with a decent site, 40 G2 reviews and a little press, scores 45 of 100 (Medium confidence). Tiny Example, a one-person site with no outside evidence, scores 9 of 100.
| Part | Points | Why |
|---|---|---|
| F: Open to AI and search crawlers | 7 of 10 | 70% of the robots rules, sitemap and llms.txt checks pass on the site. |
| F: Named in AI answers about your category | 8 of 50 | Named or cited in 1 of 5 category answers (test); counted as 1 of 6 for a small sample. |
| F: Third-party pages and stories about you | 29 of 40 | 12 third-party pages and 6 earned stories found (full at 60). |
| O: Ratings and reviews on review sites | 7 of 50 | Only 1 review family found on this run, so estimated from AI answers and web presence (at most 85% of the part). |
| O: Recommended in AI answers | 3 of 15 | AI answers are the opinion buyers now read first. Named or cited in 1 of 5 category answers (test); counted as 1 of 6 for a small sample. |
| O: Following of the brand and its leaders | 9 of 25 | 4,000 followers on the brand account, 6,000 on its leader (from 50, full at 20 million and 10 million). |
| O: People talking about you | 3 of 10 | 40 posts by others in 7 days (full at 300,000). |
| U: Pages machines can read and quote | 11 of 15 | 70% of the structured data, meta and readable text checks pass on the site. |
| U: Content worth citing | 15 of 25 | 3 of 5 kinds found: blog or newsroom, customer stories, partners, academy or docs, events or community. |
| U: Picked up by others | 37 of 60 | 6 earned stories (full at 40) and 12 third-party pages (full at 30). |
| N: Analyst reports that name you | 0 of 35 | 0 analyst mentions (the publisher’s own page 1, another third party 0.8, the company’s own page 0.6; full at 4). |
| N: Awards and lists | 8 of 20 | 1 awards or lists (the publisher’s own page 1, another third party 0.8, the company’s own page 0.6; full at 5). |
| N: Tier-1 press | 7 of 25 | 1 stories in tier-1 outlets (full at 10). |
| N: Treated as an authority by AI answers | 3 of 20 | Answer engines naming you when buyers ask about the category is a nod from the source buyers consult. Named or cited in 1 of 5 category answers (test); counted as 1 of 6 for a small sample. |
| D: Security and trust published | 18 of 25 | 70% of the security.txt, trust page and named certifications checks pass on the site. |
| D: Evidence reachable | 7 of 10 | 70% of the HTTPS and speed checks pass on the site. |
| D: Independent proof | 52 of 65 | Customer stories found; 5 units of independent proof (review families, analyst and award recognition, no public filings, independent sources at 1 per 4; full at 10). |
| Overall (mean of the five letters) | 45 of 100 | F 44, O 21, U 63, N 18, D 77 |
What would move Example Company up most
- Get named when buyers ask AI about your category (about +4): Publish a plain comparison page and a category guide, and get them cited on third-party sites the engines read. Then re-check.
- Get named in one analyst report (about +4): Brief Gartner, Forrester or IDC for a Market Guide, Now Tech or vendor landscape; inclusion is often free for qualifying vendors.
- Earn 5 stories in the trade press, one in a tier-1 outlet (about +2): Pitch one piece of original data or a customer result to trade editors; a single tier-1 story lifts Nods most.
How a score is shown
Every score is shown as a number of 100 with its confidence label, the five letters, every part with the evidence behind it and the three steps that would move it most, with estimated points. Company size is a comparison group, never a score input. Calibration anchors (Google, Microsoft, OpenAI) are scored the same way and shown on the peers page for scale, never ranked with the peers. Until 19 October 2026 the score counted machine checks of the site only, which put a ceiling of 45 on every company; that ceiling is gone in version 1.0.
What the score does not say
- It does not forecast revenue, rankings or what any AI will say.
- The weights are a calibrated judgement, not a finding. We change them only in a new version, and old results keep the version they were scored under.
- Reading public evidence at one time is a sample. Run it again and the numbers may move.
- If you think a result is wrong, tell us: how corrections work.
Read our methodology for how we collect and interpret public evidence.
Same evidence, same score. See yours.
Free, open and no login.