Site Audit measures readability only. It does not measure how often AI actually mentions or cites you — that lives in Visibility and Citation.
Running an audit
You don’t have to run the first one. When a workspace finishes onboarding an audit starts automatically, and it runs again on its own about once a month; the How this was measured card at the bottom tells you when the next one is due. To run it now, press Re-run audit at the top right. If nothing has ever run you’ll see No audit yet with a Run audit button instead. While it runs, the Audit in progress card walks through three stages.1
Discovering pages
Crawls your sitemap and internal links. It finishes as “[count] pages found”.
2
Measuring pages
Fetches each page as the AI search bots and as a browser, and reads the text inside large images. Sites over 500 pages get a representative sample per URL pattern instead of every page, and the card says so.
3
Analyzing issues
Groups what it found into findings with a fix attached.
What’s on the screen
The result is a column of cards, top to bottom.Can AI read this site?
The headline card. One sentence gives the answer, for example “On 3 of 23 sampled pages, AI reads only part of the key facts” or “AI can read the body text on all 23 sampled pages”. Under it, a bar splits the sampled pages into Readable, Partially readable, Unreadable, Unreachable and Not measured; click a segment to open the reader view on just those pages. Open reader view opens it for every page. Affected pages per stage lists how many pages drop out at each step of the way an AI reads a page: Fetch (the fetcher receives HTML), Index (indexing and snippets are allowed), Extract (facts are extracted from the body) and Select (whether AI actually picks the page — not measured, shown as —). By segment breaks the sample down by page type, with an example address for each.What to work on first
An editorial pass over a sample of your active customer questions: which of your pages should answer them, and whether they do. The line under the title says how many pages were linked to how many questions. Each item names a page, who does the work (Developer, Content, SEO or Infra) and how long it takes (Minutes, Hours or Days), then three blocks: Observed text (what the page says today), Edit to make, and Done when (how you’ll know it’s fixed). Expand Review customer questions and evidence to see every sampled question with its page and a verdict: Answer evidence found, Answer needs detail, Answer missing on this relevant page, Current applicability needs review, or No relevant evidence in this page sample.This card is an editorial reading of a sample; it does not measure citation lift. When nothing could be evidenced it says so rather than guessing. After editing, re-run the audit and compare the same questions in Visibility — the Open Visibility link takes you there.
Access gate
A table of the AI fetcher profiles (the ChatGPT, Perplexity and Google search bots, plus Browser (control)) against the sampled pages, with counts of OK, robots block, WAF block and Error. Policy check lists things that are facts, not faults, such as blocking AI training bots. Measurement status notes anything that couldn’t be measured. Show robots.txt prints the file as we read it; if there is none, it says every crawler is allowed.Missing facts
Key facts the bot could not read, one row per fact: Fact (Price, Specs, Ingredients, Shipping, Returns, Opening hours and so on), Pages (how many), and Where it lives (only after JavaScript runs, inside images, and so on). Click a row to open the reader view filtered to those pages. If everything is readable the card says so.Findings
Grouped by the stage where an AI reader loses the page: Fetch — access, Index — indexability, Extract — body text, Select — title and snippet, then Other technical checks. Each finding shows its severity (Error, Warning, Notice), how many pages it affects (an estimate is marked estimated), who does the work and how long it takes, and a status: Open, In progress, Done, Verified fixed, Dismissed or Snoozed. Show fix expands What it is, Why it matters, How to fix and the Evidence pages inline. Verify re-fetches the affected pages and reports Re-checked [n] URLs · fix confirmed or · [n] still affected. Open finding goes to the finding’s own page (below). The Sitemap button opens the URLs this run discovered, grouped into sample sets.How this was measured
Which fetcher profiles each page was fetched as, what “bot text” and “browser text” mean, how many images were read for text, and when the next automatic audit runs.The reader view
Opening the reader view from the headline, the missing-facts table or a finding slides in a panel with three tabs: Pages, Access and Facts. Pick a page and you get Text the bot received on the left and Text the browser shows plus Text inside images on the right, with a character count for each. The Facts tab lists each fact with its value and a verdict: readable, visible only after JavaScript, or unknown. This is the evidence behind every verdict on the screen.The finding page
Open finding takes you to a page for one finding. The top holds What it is, Why it matters and How to fix. Evidence lists each affected Page and What we saw. Proposed changes is the part you hand to whoever does the work. Each Change [n] gives Where (a CSS selector, a heading or an HTML block, with platform instructions), Current and Proposed with a Copy button, Why, Watch out for side effects, and Verify with … expect … for how to confirm it. Changes written by AI carry the chip AI-written — review before use. At the bottom, Verification history lists every Verify run with Resolved or Still affected, and three buttons track the work: Start working (→ In progress), Mark done and Dismiss. Mark done is your own note and stays as you set it; Verified fixed comes only from a successful Verify, and a verified finding reopens by itself if a later audit detects the problem again.Exporting
Once a run completes, an Export button appears in the header. Pressing it downloads a PDF; the caret next to it offers PDF, CSV, Markdown and Copy Markdown. The PDF is written in full sentences — the headline, the per-page verdicts, the findings with their proposed changes and how to confirm each one — so it works as-is for a leadership update or a developer handoff.FAQ
Do I have to run it myself?
Do I have to run it myself?
No. Onboarding starts the first one, and it re-runs about monthly. Re-run audit is there for right after you change the site.
Where is the score?
Where is the score?
There isn’t one any more. You get a sentence, per-page verdicts and evidence instead, because a number like “78” doesn’t tell anyone what to do. Older runs made before this change still open in the previous score layout.
It says ~1,200 pages but the table shows 30.
It says ~1,200 pages but the table shows 30.
Sites over 500 URLs are sampled, up to 30 pages per URL pattern. Counts are extrapolated from that sample, marked estimated, and can be larger than the number of rows you see.
Robots.txt allows GPTBot, but the Access gate says a bot is blocked.
Robots.txt allows GPTBot, but the Access gate says a bot is blocked.
Separately from robots.txt, each page was fetched using real AI search-bot identities and came back blocked (WAF block). That’s usually a rule in the firewall or security service in front of your site. A developer needs to adjust it.
Our robots.txt blocks GPTBot. Is that a finding?
Our robots.txt blocks GPTBot. Is that a finding?
No. Training bots are a policy choice and appear under Policy check only. The audit flags a block only when the search and answer bots are refused.
The page looks fine to me. What is the problem?
The page looks fine to me. What is the problem?
AI doesn’t look at the screen; it reads the raw document the server sends. Open the reader view and compare Text the bot received with Text the browser shows — the difference is what AI can’t see.
Can I mark a finding done or ignore it?
Can I mark a finding done or ignore it?
Yes. On the finding page, Start working, Mark done and Dismiss set the status. Verify is the one that actually re-checks the pages and can set Verified fixed; that verified status reopens if a later audit finds the problem again.
Will fixing everything raise my AI mentions?
Will fixing everything raise my AI mentions?
Being readable is a precondition, not a guarantee. Check the outcome in Visibility and Citation.
The audit failed.
The audit failed.
If you see Couldn’t complete the audit, press Re-run audit to retry. If it keeps failing, your site may be blocking crawlers — talk to your contact at Anymorph. A run stuck for over 30 minutes is stopped automatically.
Related
Site Audit report
What the audit delivers and what it leaves out.
Visibility
Check the outcome after your fixes land.
Citation
See which sources AI actually pulls from.
On-site
Create pages built to be cited.