No AI, no model bias
No language model reads or judges your pages. Each criterion is a written rule: what it checks, why, and how much it weighs. No one, human or machine, adds personal taste.
Scoring grid 2026.10.9
The method
Cairn scores each site out of 100 by applying 62 verifiable rules every morning, split into three pillars: SEO for search engines, GEO for AI answer engines, AIO for brand and trust. With no AI, so no model bias, and with very few resources.
Three principles
Reproducible
The same site on the same day gets the same score. Anyone can redo a measurement by hand. The scoring grid has a version number, and its changes are counted separately.
Lean
One check a day, one crawl a week, the HTML and a few text files, never more than six connections at once. No GPU, no browser launched on our side.
The calculation, in brief
- Each criterion returns a score from 0 to 100. If it does not apply, or could not be measured, it is left out of the calculation instead of counting as zero.
- Each pillar is the average of its criteria, weighted by their importance. The overall score combines the pillars: SEO 40%, GEO 30%, AIO 30%.
- Page criteria are scored on the whole site, not on a sample: once a week, Cairn crawls the sitemap and the internal links, up to 500 pages, and says for each criterion how many pages comply. Since scoring grid 2026.10.9; some sites' scores moved in the following week, when a defect was hiding outside the eight pages read until then.
- The homepage counts double in page averages: it is the page everyone reads.
- Everything is measured from the outside, with no account or owner access (neither Search Console nor Bing Webmaster): a competitor is scored exactly like your sites. And Cairn respects the robots.txt files it reads, including the search engines' own: Bing and Brave disallow their search pages to bots, so Cairn does not query them.
- A third-party service that goes silent for a day does not sink the curve: its last measurement is carried over, for seven days at most.
- One score per answer engine, outside the overall score: can ChatGPT, Claude, Perplexity, Copilot and Google reach, read and find the site? It is not a citation forecast. What cannot be measured is marked “not measured”: Claude's index (Brave cannot be queried without an account or against its robots.txt) and Perplexity's. A partial score is read on the measured signals alone: next to the figure, Cairn shows how many signals the score counts (“3 of 4 signals”).
- Pages by type, against competitors, also outside the overall score: the crawl sorts each page into product page, category, article or other, by written rules (the declared schema.org type, then
og:type, then the URL), and judges each type on its own grid: structured price, reviews, described images and text of its own for a product; introduction, pagination and list for a category; author, date, markup and sources for an article. A site is an online store above 20% of product pages. A type a competitor does not have is stated, and does not count as a failure. - A change of less than 3 points from one day to the next is noise. Read the trend over a week.
The caps
An average would let good details hide a locked door. So a few blocking issues cap the score, whatever happens elsewhere:
- Homepage excluded from the index, returning an error, or blocked to Google and Bingoverall ≤ 40
- Most AI search crawlers blocked, or site unreadable without JavaScriptGEO ≤ 30
- Snippets disallowed site-wideAIO ≤ 30
The score measures verifiable best practices. It predicts neither a ranking nor a citation: no one knows those formulas.
The 62 criteria
Each one carries its evidence level: established (documented by the search engine or by a standard), reported (observed in published studies) or uncertain effect (plausible, with no public evidence, and weighing little). In the tool, each criterion also states what it found and the fix that resolves it.
SEO search engines
40% of the score, 33 criteria
Can Google and Bing crawl the site, index it and present it well? This is the foundation of the other two pillars: the answers Google and Bing generate draw on their index.
- Indexable pages established
- Homepage response established
- Google and Bing allowed established
- Rendering resources allowed established
- robots.txt file established
- HTTPS enforced established
- HTTPS remembered (HSTS) established
- A single URL for the homepage established
- Canonical URL established
- Sitemap established
- Clean sitemap URLs established
- Title tag reported
- Meta description established
- Main heading (h1) reported
- Heading hierarchy reported
- Image alt text established
- Crawlable links established
- Descriptive link text established
- Broken internal links established
- Direct internal links established
- Distinct pages established
- Unique titles and descriptions established
- Unknown URL returns 404 established
- Language versions established
- Mobile display established
- Declared language established
- Mixed content established
- Link preview when shared reported
- Mobile performance (Lighthouse) reported
- Core Web Vitals (field) established
- Server response time established
- Text compression established
- Document size established
GEO AI answer engines
30% of the score, 12 criteria
Can answer engines (ChatGPT, Claude, Perplexity, Copilot) read the site and quote passages from it? Access for their crawlers, text readable without JavaScript, content that is structured, dated and sourced.
- AI search crawlers allowed established
- Text readable without JavaScript reported
- No bot blocking reported
- Content freshness reported
- Citable passages uncertain effect
- Figures and sources reported
- Identifiable main content uncertain effect
- Answer first uncertain effect
- On-demand reading allowed established
- Present in DuckDuckGo (Bing) reported
- llms.txt file uncertain effect
- Training crawlers info
AIO brand and trust
30% of the score, 17 criteria
Is the brand recognized, and do the pages carry it? A unique, declared name, present in titles, on the homepage and in the text, so that every quoted passage stays attributed; then what builds trust: snippets allowed, structured data, trust pages, presence in open databases.
- Consistent brand name reported
- Brand in titles reported
- Brand presented on the homepage reported
- Brand in content reported
- Found on its name reported
- Snippets allowed established
- Valid structured data established
- Declared identity reported
- Article author and dates reported
- Trust pages reported
- Wikidata entity reported
- Linked profiles reachable uncertain effect
- Breadcrumb established
- Present in Common Crawl reported
- Domain prominence info
- Readable contact details reported
- Site name declared established
What Cairn reads
- The site itself, like a search crawler: the homepage and a sample of pages from the sitemap every day, nearly the whole site once a week, robots.txt, the sitemap, a non-existent address to check the error page, and the homepage as answer engines' robots receive it. Cairn obeys the rules it scores there: robots.txt and Crawl-delay, two requests at a time at most, and a stop after a run of refusals or server errors.
- PageSpeed Insights, for speed as measured by Google, and real visitor data when it has some.
- Wikidata and Common Crawl, for the site's presence in the open databases that search engines and models draw on.
- DuckDuckGo, whose results come largely from Bing, which feeds ChatGPT and Copilot: the site's pages in the index, and whether the site is found by its name. Once a week per site, through its lightweight HTML page, which its robots.txt allows; with no account, with spaced-out requests, and backing off when refused. An absence is only treated as certain after being seen twice in a row.
What the rules are based on
Each criterion states why it matters, and where that comes from. The main sources:
- Creating helpful, reliable, people-first content, Google Search Central: trust pages, author, date.
- Introduction to structured data markup, Google Search Central: the JSON-LD criteria.
- RFC 9309, Robots Exclusion Protocol, IETF, 2022: how robots.txt is read.
- GEO: Generative Engine Optimization, Aggarwal et al., KDD 2024: the statistics, sources and quotations that make a page citable.
See what the method says about your sites.
Contact us