Full site audit
The full report, for a one-off $9.99
The free audit tells you what is wrong. The full report tells you which pages, scores each one, names every section that needs rewriting, and sets you against three competitors you choose. Then you fix things and run it again.
A one-off payment, not a subscription. 5 re-runs are included over the following 30 days, and it costs nothing extra on any paid AI Visibility plan.
| Free | Full report · $9.99 | |
|---|---|---|
| Which pages each problem is on | Not included | Included |
| A score for every page | Not included | Included |
| Every weak section, named and quoted | Not included | Included |
| What to do about each finding | Not included | Included |
| Competitor comparison | Not included | 3 sites |
| Re-run after you fix things | Not included | 5 in 30 days |
| Which AI crawlers you let in, one by one | Not included | Included |
| Save as PDF | Not included | Included |
| Pages read | 10 | 1,000 |
| Every check we run | Included | Included |
| All findings, with severity | Included | Included |
| Account needed | Not included | Not included |
We check we can read your site before you pay.
What the audit checks, before you run it
Five stages, in the order an AI answer actually happens. A check only earns its place if real sites measurably fail it, and each one tells you which of the five you are failing rather than handing you a single number.
- 1 · Access
Can a crawler fetch your pages at all?
- We read your robots.txt against the crawlers that actually matter, and we sort them into three groups rather than one, because a block costs you completely different things depending on which one it is. Then we request your pages as an identified crawler. Bot protection sits in front of robots.txt, so a firewall can shut out AI crawlers while your robots.txt politely says come in. That one is invisible to every checker that only reads robots.txt.
- 2 · Rendering
Is anything there before JavaScript runs?
- GPTBot, ClaudeBot and PerplexityBot fetch your raw HTML and move on. If your pages are assembled in the browser, those crawlers receive an empty shell. We check what survives with the JavaScript switched off, which is the state most assistants see your site in.
- 3 · Discovery
Can they find everything worth reading?
- We walk your sitemap, including nested indexes, and sample real pages from it. Dead entries matter more than most people expect: measured on production logs, ChatGPT's crawler spends 34.8% of its fetches on URLs that no longer exist against Googlebot's 8.2%, so every stale line costs you a page that could have been read instead.
- 4 · Chunking
Does each section stand up on its own?
- Retrieval does not hand an assistant your page. It splits the document at its headings, matches the question against one section at a time, and returns the section that wins. We split your pages the same way and measure every section: how long it is, whether it owns any text at all, and whether it opens on a word like this or however whose meaning lives in a section the assistant never fetched.
- 5 · Evidence
Is there anything in there worth quoting?
- Assistants quote specifics. We count the sections carrying a statistic, a quotation or a link to an outside source, and report the share. This one comes with a caveat we state in the report rather than bury: the published measurement behind it found the effect concentrated on pages that ranked poorly to begin with, and slightly negative for pages already ranked first.
How AI actually reads a website
Every point below comes from an AI company's own documentation or from a published study, and where the evidence is thin we say so rather than round it up. Nothing here is our opinion about how AI search works.
It fetches your HTML. It does not run your site.
GPTBot, ClaudeBot and PerplexityBot request your raw HTML and move on. Measured across a large sample of production server logs, no major AI crawler executes JavaScript: ClaudeBot downloaded script files in roughly a quarter of its requests and ran none of them. Google's Gemini and Apple's AppleBot are the exceptions, because both render through their own search infrastructure. So a page can sit at position one on Google and be invisible to ChatGPT at the same moment.
One question becomes many searches.
Assistants rarely run your question once. Google describes AI Overviews and AI Mode as using a query fan-out technique, issuing multiple related searches across subtopics and data sources simultaneously, and describes Deep Search as taking the same technique to hundreds of searches. Your page is therefore not competing for the question a buyer typed. It is competing for whichever sub-question it happens to answer better than anyone else.
Retrieval takes sections, not pages.
A retrieval system splits a document into passages, matches the question against one passage at a time, and hands only the winning passage to the model. Retrieval systems commonly index passages of a few hundred words. That makes the section, not the page, the unit that competes, and it is why a beautifully written page built out of forty-word stubs loses to a worse page whose sections each stand up alone.
Evidence travels further than adjectives.
In the GEO study presented at KDD 2024, adding citations, quotations and statistics were the three best performing of nine tested content changes, beating the baseline by up to 41%, while keyword stuffing was one of two changes that failed to beat it at all. One detail the headline usually drops decides whether this is worth your time: the gains concentrated on pages that ranked poorly to begin with, where citing sources roughly doubled visibility, and the same edits measurably reduced visibility for pages already ranked first.
Not every crawler does the same job, and blocking one is not like blocking another.
Crawlers split into three jobs, and most audits treat them as one. OpenAI states that sites opted out of OAI-SearchBot "will not be shown in ChatGPT search answers", so blocking that one genuinely removes you. GPTBot, by contrast, only collects training data. Google-Extended governs Gemini app grounding, and Google states it "does not impact a site's inclusion in Google Search", so blocking it does not touch AI Overviews or AI Mode. Blocking Bytespider or CCBot is a deliberate choice thousands of publishers make on purpose. Babel42 reports the three groups separately and only calls the first one blocking.
Most of what an assistant quotes is not your website.
Analyses of AI citations disagree on the exact figure and agree on the direction. One study of 25 million cited links across ChatGPT, Claude and Gemini put earned media at 84% of citations and brand-owned sites at 13.7%; others put owned sites nearer 5 to 10%, and one found Gemini leaning on brand-owned sites far more heavily than ChatGPT. Whichever number is right, most of what gets quoted is somebody else writing about you. Fixing your own site is necessary and nowhere near sufficient.
This is the first of two questions.
Can AI assistants read you?
Technical, verifiable, and fixable in an afternoon. That is what this page answers, free, in a few seconds.
Do they actually bring you up?
A different question entirely, and the one that decides whether buyers hear your name. Babel42 runs real buyer questions across ChatGPT, Claude, Gemini and Perplexity, and tracks how often you come up against your competitors.
A perfect score here does not mean assistants recommend you. It means nothing technical is stopping them.
Every finding says how firm it is
A great deal of what gets said about AI search is guesswork dressed as fact, so every finding in your report carries a label, and only the first two are allowed to move your score.
- Measured on your siteCan affect your score
- We fetched it and this is what came back.
- Backed by published measurementCan affect your score
- A named study or an operator's own documentation says so, and the report cites it.
- Follows from how retrieval works, not measuredNever affects your score
- A reasonable inference we have not proved. Reported, never scored.
- Emerging convention, unconfirmedNever affects your score
- A convention no major crawler is confirmed to read, llms.txt among them. Reported, never scored.
Two claims we deliberately do not make. First, we do not treat every blocked AI crawler as a disaster. Blocking the ones that only collect training data is a deliberate choice, and a common one: an audit of 14,000 websites found that in a single year more than 28% of the most actively maintained sources in a major AI training set became fully blocked in robots.txt, using rules written specifically for AI companies and treating different companies differently. That is a policy decision, not a mistake, and we report it without scoring it. Second, we do not tell you that a section which never names your brand costs you the credit for it. Assistants credit you with a link, and that link travels with the section whether or not your name appears in the words. What an unnamed section can cost you is matching, on the questions that use your name, so we report it as something to consider and never as something you failed.
Questions people actually ask
What is an AI SEO audit?
An AI SEO audit checks whether AI assistants can fetch, read and quote your website, rather than whether it ranks on Google. Babel42's free audit requests your pages as an identified crawler, reads your robots.txt against the crawlers that feed AI answers, checks what survives without JavaScript, and measures whether each section of a page can be retrieved and understood on its own.
Why is AI not citing my website?
Three causes account for most of it. Either a crawler cannot reach you, because robots.txt or a firewall refuses it; or it reaches you and finds an empty page, because the content only appears after JavaScript runs; or it reads you fine and your sections have nothing specific enough to quote. Babel42's audit separates the three, because they are fixed by different people.
Can AI assistants read my site?
Most sites are readable and some are not, and the difference is rarely visible in a browser. The common failures are a firewall that challenges non-browser clients, a robots.txt rule that blocks the crawler feeding an assistant's search, and client-side rendering that leaves the served HTML empty. A free Babel42 audit fetches your pages the way a crawler does and reports exactly what came back.
Does ChatGPT read JavaScript?
No. Measured across a large sample of production server logs, no major AI crawler executes JavaScript, including OpenAI's, Anthropic's and Perplexity's. They download script files without running them. Google's Gemini and Apple's AppleBot are the exceptions, because both render through their own search infrastructure. If your pages are assembled in the browser, most assistants receive an empty shell.
Do I need an llms.txt file?
No, on the current evidence. Across 137,000 domains, 28% publish an llms.txt file and 97% of those files received no requests at all in a month, and Google states you do not need to create machine-readable files for its AI features because Search does not use them. Babel42 reports whether you have one and never counts its absence against your score.
What does it mean if my site is blocked to GPTBot?
Less than most tools imply. GPTBot collects training data, so blocking it does not remove you from ChatGPT's search answers; OAI-SearchBot is the crawler that does that, and OpenAI says opted-out sites will not be shown there. Blocking training crawlers is a legitimate publishing decision with a slow, long-run trade-off in what a model knows about you when it answers from memory.
How do I get my site ready for AI search?
Work in order of what it costs you. Let the answer-feeding crawlers in, make sure your pages carry their content in the served HTML, and clear dead URLs out of your sitemap. Only then edit: give every heading real text, open each section by naming its subject rather than saying this or it, and put a number, a quotation or a source link in the sections you most want quoted.
Is being readable enough to get recommended?
No, and treating it as enough is the most common mistake in AI search. Readability is the floor: it decides whether an assistant can use you, not whether it chooses to. Studies of AI citations put a brand's own site somewhere between 5% and 14% of the sources assistants draw on, with the rest being other people writing about you. Babel42's paid product measures the other question, which is whether assistants bring you up at all.
Readable is the floor, not the goal.
Fixing everything above puts you in the running. Babel42 tells you whether you are actually winning: which buyer questions bring up your name, which bring up a competitor instead, and what changes that.