The AI-Visibility Audit: Check If AI Can Find and Quote You in 15 Minutes

You can assume AI has read your website. You can also be wrong, and the only way to know is to check.

TL;DR AI cannot quote a page it cannot read, and plenty of good sites are quietly unreadable to machines. The gap is real and measurable: a Vercel and MERJ study of over 500 million GPTBot fetches found the major AI crawlers do not execute JavaScript, so any content that only loads after a script runs is invisible to them. This is an 8-point audit you can run in about 15 minutes to see whether AI can find, understand, and quote you. Each step tells you how to read the result, then we cover what to fix first, then the shortcut.

Why you cannot assume AI can read you

Here is the uncomfortable part. Your website can look flawless to a human and still be a blank page to a machine. A crawler does not see your design. It sees your markup, your rules, and your structure, and it makes fast decisions about whether your page is readable, relevant, and safe to quote.

If any of those decisions go the wrong way, you do not get a warning. You just quietly do not appear when someone asks an AI assistant a question you should own. And being readable is only half the battle. Citation rates vary sharply by engine: Conductor's benchmarks put ChatGPT's rate of answers that include a source link at about 0.7 percent, against 13.8 percent for Perplexity and 9.5 percent for Google AI Mode. The point is not the exact number, it is that every engine is stingy with citations, so the pages that do get quoted are the ones that make themselves impossible to overlook. So instead of assuming, run the checks below. Pick one page that matters, ideally a product or high-intent page, and work through the eight points.

The 8-point AI-visibility audit

Each step has a thing to check, how to check it, and how to read the result.

1. Crawlability and robots rules

How to check: Visit yoursite.com/robots.txt in a browser. Read it for Disallow lines that block important paths, and look for rules aimed at AI crawlers by name. Then confirm the page you care about is reachable without a login and returns a normal page, not an error or a redirect loop.

How to read it: Good means AI crawlers can reach your key pages and nothing important is disallowed. Warning means you see broad disallow rules you do not remember setting. Fail means you are blocking the crawlers you want, or the page is gated behind something a bot cannot pass. This is the first check because if machines cannot reach the page, every other point is moot.

2. Structured data and JSON-LD

How to check: View the page source and search for application/ld+json, or paste the URL into a public structured-data testing tool. Look for schema types that match the page, such as Article, Product, FAQ, or Organization.

How to read it: Good means valid structured data that describes what the page actually is. Warning means structured data exists but throws errors or describes the wrong thing. Fail means there is none at all, so the machine has to guess your page's purpose from raw text. Structured data is how you tell a crawler "this is a product, this is the price, this is the answer" in a language it trusts. One caveat worth knowing: Google states there is no special schema you must add to appear in AI Overviews, and that supplying fewer but complete and accurate properties beats stuffing in every possible one. Treat schema as a clarity aid, not a magic switch.

3. Answer-ready copy

How to check: Read your page as if you were a machine looking for a single, liftable fact. Does a clear question have a clear, direct answer near it? Are your headings phrased the way a person would actually ask? Or is the real answer buried three paragraphs into a wall of text?

How to read it: Good means direct answers sit close to plain-language headings, easy to lift. Warning means the answer exists but is wrapped in fluff. Fail means your key claims are implied, scattered, or written only to sound impressive. AI assistants quote content they can extract cleanly. Vague, meandering copy gives them nothing to grab.

4. Entity and brand consistency

How to check: Search your exact brand name across a few AI tools and search engines. Check that your name, what you do, and your core facts are described the same way everywhere: your own site, your social profiles, and any listings. Look for conflicting descriptions or old names.

How to read it: Good means machines get one clear, consistent picture of who you are. Warning means small inconsistencies, like a tagline that changed but lives on in three old profiles. Fail means the model cannot tell if you are one company or three. Consistency builds the confidence a model needs before it will name you in an answer.

5. Freshness signals

How to check: Look at your key pages for visible, honest dates and recent updates. Check whether your most important content has been touched in the last year, and whether your sitemap reports accurate last-modified dates.

How to read it: Good means important pages carry real, recent update signals. Warning means content is fine but undated, so a model cannot tell if it is current. Fail means everything looks years old. AI systems favor content they believe is current, especially for anything that changes over time. Stale-looking pages get passed over for fresher sources.

6. Internal linking

How to check: Pick your most important page and ask how a crawler would find it. How many clicks from the homepage? Do related pages link to it with clear, descriptive text? Or is it an orphan that nothing points to?

How to read it: Good means important pages are well linked from relevant places with meaningful anchor text. Warning means links exist but use vague text like "click here." Fail means key pages are buried or orphaned. Internal links are how machines learn which pages matter and how your topics connect. Poor linking hides your best work.

7. AI-readable files like llms.txt

How to check: Visit yoursite.com/llms.txt and yoursite.com/sitemap.xml. The sitemap should list your real pages. The llms.txt file, if present, should give AI systems a clean, plain-text map of your most important content and what your site is about.

How to read it: Good means a valid sitemap plus, ideally, an llms.txt that guides AI to your key pages. Warning means a sitemap exists but is outdated or partial. Fail means no sitemap and no AI-readable files, leaving machines to work it all out from scratch. Be honest about what llms.txt buys you today: adoption is still low, and no major AI company has formally committed to reading it in production, with Google Search Relations stating it has no plans to support the format. Your sitemap is the load-bearing file; llms.txt is a cheap, low-risk bet on where things are heading, worth adding but not worth agonizing over.

8. Mobile and core rendering

How to check: Load the page on a phone, or use your browser's device view. Then, and this is the real test, disable JavaScript and reload. See how much of your content still appears.

How to read it: Good means your core content and answers are present even without JavaScript, and the mobile layout is clean. Warning means some pieces need scripting to appear. Fail means the page is nearly blank without JavaScript, which means many crawlers see almost nothing. This is not a hypothetical: the Vercel and MERJ analysis found GPTBot fetched JavaScript files about 11.5 percent of the time and ClaudeBot about 23.8 percent, but neither crawler executed them. Gemini, which rides on Google's rendering infrastructure, is the notable exception. If your content only exists after heavy scripting, you are betting your visibility on every machine running that script perfectly, and most of the AI ones do not.

What to fix first

Eight findings can feel like eight projects. It is not. Fix them in order of what unblocks the rest.

• Fix access first. Points 1 and 8, crawl rules and rendering. If a machine cannot reach or read the page, no other fix counts. Start here every time.
• Fix understanding next. Points 2 and 3, structured data and answer-ready copy. Once a crawler can read the page, help it understand and quote you.
• Fix trust and polish last. Points 4, 5, 6, and 7, consistency, freshness, linking, and AI-readable files. These raise how often you get chosen once the basics work.

Always fix a fail before a warning, and always fix access before anything else. A perfect structured-data setup on a page no crawler can reach helps no one.

The shortcut

Running this by hand on one page takes 15 to 30 minutes if you know where to look. Doing it across a whole site, page by page, is a long afternoon that you will not repeat often enough to catch new leaks.

That is the exact job we built Clarvia to do. You give us a URL and we run all eight checks at once, across the five channels of AI search, then hand you one score and a prioritized list of the specific fixes that move it. The manual audit above is the honest version of what a scan does in seconds. If you would rather see the results than chase them, the shortcut is there.

Key takeaways

• AI cannot quote a page it cannot read, and good-looking sites are often unreadable to machines. Check instead of assuming.
• The 8 points, in order: crawlability, structured data, answer-ready copy, entity consistency, freshness, internal linking, AI-readable files, and rendering.
• Read each result as good, warning, or fail, and always fix a fail before a warning.
• Fix in this order: access first, then understanding, then trust and polish. Access unblocks everything else.
• A manual audit takes 15 to 30 minutes per page. A scan runs all eight at once and hands you the priority list.

Sources

• Vercel and MERJ: The rise of the AI crawler - 500M+ fetches showing GPTBot and ClaudeBot do not execute JavaScript.
• Google Search Central: AI Features and Your Website - Google's official stance on schema, indexing, and AI Overviews.
• Conductor AEO/GEO benchmarks, via Rank.ai - citation rates by engine: ChatGPT ~0.7%, Perplexity 13.8%, Google AI Mode 9.5%.
• IndexLab: Does llms.txt actually work? - adoption status and platform statements on llms.txt as of late 2025.

Frequently asked questions

How do I check if my site shows up in AI search?

Ask the AI tools directly. Open ChatGPT, Perplexity, or Gemini and ask a question a buyer would ask in your category, then see whether your brand or a competitor gets named. Pair that with the technical checks in this audit: robots rules, structured data, answer-ready copy, and AI-readable files. Together they tell you whether AI can find you and whether it chooses you.

Why would AI not be able to read my website?

Usually because something blocks or confuses the machine reading it. Common causes are robots rules that disallow AI crawlers, content that only loads through heavy JavaScript, missing or broken structured data, and copy written for skimming humans rather than direct questions. The page can look perfect to you and still be unreadable or untrustworthy to a crawler.

What is an AI SEO audit?

An AI SEO audit checks whether AI systems can crawl your site, understand it, and quote it as an answer. It goes beyond classic ranking factors to cover structured data, answer-ready copy, entity consistency, freshness, internal linking, AI-readable files like llms.txt, and mobile rendering. The goal is to find where you are invisible to AI, not just where you rank on a results page.

What should I fix first after an AI-visibility audit?

Fix anything that blocks access first, because it makes every other fix pointless. That means robots rules and rendering: if crawlers cannot reach or read the page, nothing else matters. Then fix understanding, meaning structured data and clear answers. Then fix polish, meaning freshness, linking, and experience.

How long does an AI-visibility audit take?

The manual version in this article takes about 15 to 30 minutes for one important page if you know where to look. Auditing a full site by hand takes much longer, which is why a scan that runs every check at once and hands you a prioritized list saves the most time.

Want the eight checks run for you? Run a free AI-search scan of your site and get one score, the eight signals broken out, and the specific fixes to make first.