GEO Audit Methodology: Every Check and How the Score Is Calculated

How the MaslonLabs GEO Audit scores a website: up to 81 checks in 8 categories, their weights, the score formula and the 35-point cap.

In short: the MaslonLabs GEO Audit runs up to 81 checks in 8 categories on a homepage and up to 8 inner pages. 72 checks carry a weight from 1 to 3 and count towards a score from 0 to 100. This page lists every check, its weight and the rules behind the score, so you can verify any result yourself.

Version 1.2, last updated November 2026. Version 1.2 made the scoring stricter: warnings earn a quarter instead of half, share-based checks earn partial credit, and clear gaps such as no Organization markup or no trust pages now fail.

How the Overall Score Is Calculated

  • Every scored check has a weight: 3 for critical checks, 2 for important ones, 1 for the rest.
  • A check that passes earns its full weight. A warning earns a quarter of it. A fail earns nothing.
  • Checks measured as a share, such as “6 of 8 articles show a byline” or “74% of images have alt text”, earn partial credit in proportion to the share, up to three quarters of the weight below the pass mark. When none qualify, the check fails.
  • The score is the points earned divided by the points available, multiplied by 100 and rounded.
  • Each failed check with weight 3 takes off 5 more points, up to 25 in total.
  • The score measures AI readiness: whether AI search engines can reach, read and quote the site. It does not measure whether assistants already mention or recommend you. That depends a lot on signals outside your site, such as reviews, mentions and links, so the report gives you prompts to test it yourself.
  • Checks marked Not scored are shown for information and never change the score.
  • Checks that do not apply are left out. A site with no articles is not marked down on author bylines.

Category scores use the same formula, limited to the checks in that category and without the extra deduction.

Score Bands

ScoreBandMeaning
80 to 100GoodAI crawlers can reach and read the site. Only small fixes are left.
50 to 79Needs workThere are real gaps. Fix the failed checks first, then the warnings.
0 to 49PoorCrawlers or visitors are likely running into real problems.

The 35-Point Cap

Some problems stop AI engines from using a site at all. When the audit finds one, the score is held at 35 or lower, whatever else is right, and the report names the reason. The cap is triggered by any of these:

  • The homepage does not answer HTTP 200
  • The homepage carries noindex, or “none”, in the robots meta tag or the X-Robots-Tag header
  • A noindex aimed only at Googlebot or Bingbot
  • robots.txt blocks the whole site, or blocks Googlebot or Bingbot
  • robots.txt blocks half or more of the AI search crawlers
  • robots.txt answers with a server error (429 or 5xx), which crawlers treat as “everything is blocked”
  • A firewall or CDN refuses half or more of the AI user agents we test
  • The raw HTML holds fewer than 50 words, so the content is missing or only appears after JavaScript runs

Every Check, by Category

AI readiness

Can AI crawlers get in, and do they receive a page they can read and trust? The article checks only run when the sampled pages carry Article markup.

CheckWeight
Homepage reachable3 (critical)
robots.txt answers without a server error3 (critical)
Googlebot and Bingbot allowed in robots.txt3 (critical)
AI search crawlers allowed in robots.txt3 (critical)
User-triggered AI fetchers allowed in robots.txt2 (important)
AI training crawlersNot scored
Firewall or CDN does not turn AI crawlers away3 (critical)
Content visible without JavaScript3 (critical)
Freshness signals1 (standard)
Structure that is easy to quote1 (standard)
Trust pages linked1 (standard)
Rendering: server-side or client-side?3 (critical)
Headings phrased as questions1 (standard)
FAQ or HowTo markupNot scored
Publish dates visible on articles1 (standard)
Articles declare an update date1 (standard)
A named author is visible on articles1 (standard)
Articles link to sources1 (standard)
Page is indexable and snippet-eligible3 (critical)
Articles show author and publish date1 (standard)

Content & citability

Is the text shaped so an assistant can lift a passage and quote it? These checks run on the homepage and the sampled pages that have at least 300 words. They are best practice for quotable writing; no AI engine publishes them as rules.

CheckWeight
Main content is marked up1 (standard)
Summary or key takeaways near the top1 (standard)
Opening paragraph answers directly2 (important)
Sections open with a short answer2 (important)
Section openings stand on their own1 (standard)
Key terms are defined in plain sentences1 (standard)
Lists and tables inside the content1 (standard)
Statistics name their source1 (standard)
Sentences are short enough to quote1 (standard)
Paragraphs are short1 (standard)
Claims are specific, not hype1 (standard)
No hidden instructions aimed at AI2 (important)

Indexing & crawling

Can crawlers find one clear, indexable version of every page?

CheckWeight
robots.txt does not block the whole site3 (critical)
Site served over HTTPS2 (important)
SSL certificate validity2 (important)
One version of the site (http/https, www)2 (important)
Canonical URL2 (important)
Missing pages return a real 4041 (standard)
XML sitemap2 (important)
No crawler-specific noindex3 (critical)
Snippets are not limited2 (important)
Short redirect chain1 (standard)
hreflang tags1 (standard)

On-page content

The basic tags every crawler reads first, checked on the homepage.

CheckWeight
Title tag3 (critical)
Meta description2 (important)
Exactly one H12 (important)
Heading hierarchy1 (standard)
Image alt text1 (standard)
Language declared1 (standard)
Mobile viewport2 (important)
Open Graph and Twitter cards1 (standard)
Favicon1 (standard)

Structured data

JSON-LD markup that tells engines who is behind the site and what each page is.

CheckWeight
Structured data has its key fields1 (standard)
JSON-LD structured data2 (important)
Organization or person entity2 (important)
Linked profiles (sameAs)1 (standard)
Breadcrumb markupNot scored
WebSite markup1 (standard)
Organization markup has a logo1 (standard)
One brand name everywhere1 (standard)

Links & pages

Links on the homepage and up to 8 pages sampled evenly from the XML sitemap.

CheckWeight
Internal links on the homepage1 (standard)
Descriptive link text1 (standard)
Broken internal links2 (important)
Pages listed in the sitemap2 (important)
Title, description and H1 on inner pages2 (important)
Unique titles and descriptions1 (standard)
Page depth1 (standard)
Inner pages point their canonical to themselves1 (standard)
Sampled pages are indexable2 (important)

Rendering & speed

AI fetchers that open a page live work with short timeouts and read the raw HTML, so weight and response time matter.

CheckWeight
Server response time2 (important)
HTML size1 (standard)
Compression1 (standard)
Text-to-code ratio1 (standard)
Page size in elements1 (standard)
Render-blocking scripts1 (standard)

AI standards (optional)

Newer conventions that no large engine requires. They are reported for information and never change the score.

CheckWeight
llms.txt fileNot scored
llms-full.txt fileNot scored
Content-Signal lines in robots.txtNot scored
noai and noimageai tagsNot scored
Markdown version of the pageNot scored
RSS or Atom feedNot scored

Which Crawlers Are Checked

The audit reads robots.txt for 25 crawlers: 2 search engine crawlers, 8 AI search crawlers, 6 user-triggered fetchers and 9 training crawlers.

TypeCrawlersEffect on the score
Search engineGooglebot, BingbotWeight 3. Blocking either one triggers the 35-point cap
AI search crawlerOAI-SearchBot, Claude-SearchBot, PerplexityBot, Applebot, DuckAssistBot, MistralAI-Index, Meta-WebIndexer, Amzn-SearchBotWeight 3. Blocking half or more triggers the cap
User-triggered fetcherChatGPT-User, Claude-User, Perplexity-User, MistralAI-User, Meta-ExternalFetcher, Amzn-UserWeight 2. A block is a warning
Training crawlerGPTBot, ClaudeBot, Google-Extended, Applebot-Extended, CCBot, Bytespider, meta-externalagent, Amazonbot, cohere-aiNot scored. Blocking them is your choice

Perplexity-User, Meta-ExternalFetcher, Amzn-User are known to ignore robots.txt, so a rule against them is reported but not counted.

Live firewall test. The homepage is requested again with the user agents of 10 crawlers: OAI-SearchBot, Claude-SearchBot, PerplexityBot, Googlebot, Bingbot, Applebot, DuckAssistBot, ChatGPT-User, Claude-User, Perplexity-User. An answer of 401, 403, 406, 429 or 503, when a browser gets 200, counts as turned away. Googlebot, Bingbot and Applebot are the exception: many sites accept those names only from the crawler’s real IP addresses, so a refusal of our request is reported as normal and does not count against you.

How Pages Are Sampled

  • The audit always starts from the homepage of the domain you enter.
  • It finds the XML sitemap from robots.txt, or at /sitemap.xml, /sitemap_index.xml, /wp-sitemap.xml and /sitemap-index.xml.
  • It picks up to 8 URLs spread evenly across the sitemap, so the sample is not just the newest posts.
  • Up to 10 internal links from the homepage are tested for broken targets.
  • The whole run has a budget of 40 seconds. On a slow site some later checks are skipped rather than left hanging.

Where the Rules Come From

The checks fall into two groups, and the difference matters when you decide what to fix.

  • Documented behaviour. Crawler access, robots.txt handling, noindex, snippet limits, canonical tags, sitemaps and structured data follow what Google, Bing, OpenAI, Anthropic and the robots.txt standard publish.
  • Best practice. The Content & citability checks, such as answer length, summaries and sentence length, reflect how quotable text is usually written. No AI engine publishes them as ranking rules, so treat them as useful signals.

What the Audit Cannot See

  • It does not run JavaScript. Rendering is inferred from the raw HTML.
  • It runs from one server location, so it cannot see IP allowlists or rules that differ by country.
  • It does not ask ChatGPT, Claude or Perplexity anything, so it cannot tell you whether they mention your brand. The report gives you six prompts to test that by hand.
  • It does not measure backlinks, rankings, brand mentions on other sites or Core Web Vitals.
  • Response time is measured from our server once, so treat it as an estimate.

Version History

VersionDateChange
1.1November 2026Up to 81 checks. Added the Content & citability and Rendering & speed categories, crawler-specific noindex, snippet limits, redirect chain, hreflang, entity checks, the 35-point cap and four more crawlers. Crawlers are now reported by type.
1.0October 2026First public version with about 45 checks in five categories.

Ready to test a site? Run the free GEO audit.