How the audit is scored, check by check
The audit runs twenty-six deterministic checks on your public pages. Each one returns pass, partial or fail, with the evidence found on your page, one sentence on why it matters and one sentence with the fix. The score is a weighted total out of 100.
No model is called at any point. The same address gives the same score, every time. The score is a reading of your public pages, not a guarantee that any AI will cite you.
What each check reads
1. AI crawler access
Reads the site's robots.txt and reports, for GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot and Google-Extended, whether each is blocked, allowed or unspecified.
2. llms.txt
Reads /llms.txt at the site root and reports whether the site publishes one.
3. Server-rendered text
Counts the visible words in the raw HTML of the homepage, before any JavaScript runs.
4. schema.org JSON-LD
Reads the application/ld+json blocks, checks they are valid JSON, and matches the organisation type to the firm's trade.
5. FAQ or Q&A content
Looks for FAQPage structured data and for question headings that a machine can read.
6. Explicit niche statement
Looks for a we work with, we specialise in or for [audience] sentence on the homepage or about page.
7. Services list
Looks for a named list of services under a services heading, or schema.org Service data.
8. Exclusions
Looks for a we do not act for, not suitable for or out of scope statement.
9. Location and service area
Looks for an address, suburb, region or we serve statement, and whether the details agree across the pages read.
10. Credentials
Looks for a registration or licence number, a named professional body, or a named qualified person.
11. Title, meta description and single H1
Reads the title, the meta description and counts the H1s, and checks the trade is named in at least one.
12. Heading structure
Counts the H2 and H3 headings under the H1 and flags a page that is mostly images.
13. Contact path
Looks for a tel or mailto link so a machine can read a way to reach the firm.
14. Freshness
Looks for a visible date or a last-updated marker on the content.
15. Sitemap and robots.txt
Reads robots.txt and /sitemap.xml and checks the sitemap is present and referenced from robots.txt.
16. Served as HTML
Reads the content-type header of each page and flags a page not served text/html, which crawlers skip.
17. Canonical tag
Reads the rel=canonical link and checks it points at the page's own address.
18. Preview tags for shared links
Reads og:title, og:image and twitter:card, which shape the preview a shared link shows.
19. Direct answer at the top
Measures the first paragraph and checks it is a self-contained 40 to 60 word answer.
20. Question-shaped headings
Counts the H2 headings that are questions, the form buyers actually type.
21. About and entity page
Looks for a reachable about or entity page that carries the firm's address and phone.
22. Internal links
Counts the internal links on the homepage that build the topical cluster.
23. dateModified in structured data
Reads dateModified from the JSON-LD so a machine can see the page is current.
24. Image alt text
Counts the images on the homepage and the ones without alt text.
25. Mobile viewport
Reads the viewport meta tag, the mobile contract.
The score and the band
The score is the sum of the weight of each check you pass, plus half the weight of a partial, over the total weight of the checks that apply. The band is Citable at 80 or above, Partly citable from 50 to 79, and Not citable below 50. The three fixes at the top of a report are the failing and partial checks with the largest weight lost.
What the audit does not check
Backlinks, search rankings, paid advertising, analytics, and whether any specific AI will cite you. Those are out of scope and named here so no one is sold a coverage that does not exist. Run the audit.