We run these audits for a living, so every step below comes with the failure we most often find at that step. Some of them will sound implausible until you run the checks on your own site.
What is an AI search audit?
An AI search audit is a structured review of how your website performs across AI-driven search: ChatGPT, Perplexity, Gemini, Claude and the AI layers of Google. It examines whether AI crawlers and agents can reach your pages, what content each one is actually served, whether that content survives without JavaScript, and whether it is structured so answer engines can lift and cite it.
It differs from a traditional technical SEO audit in what it tests against. An SEO audit measures your site against Google's ranking systems; an AI search audit measures it against a set of agents that behave differently from Googlebot and from each other, and whose visits never appear in your analytics. A site can be technically excellent by SEO standards and still fail an AI search audit, usually at rendering or at access.
How to audit your AI search visibility
Six steps, roughly an afternoon, no special access needed beyond your own CDN login.
1. Check robots.txt and your CDN rules
Open yourdomain.com/robots.txt and look for disallow rules naming AI agents: GPTBot, ClaudeBot, CCBot, Google-Extended, PerplexityBot. Then check your CDN dashboard for bot-protection features and AI crawler policies. The most common finding at this step: blocking rules nobody remembers creating, because CDN providers now apply AI-blocking defaults that most site owners have never seen.
2. Fetch your pages as the agents
Rules say what should happen; only a fetch shows what does. Fetch your key pages as each AI agent and compare what comes back. Our free Agent Parity Check does this for any public URL in about 30 seconds, across seven agents. The most common finding: divergence, the same URL returning the full page to one agent, a thin shell to another and an error to a third, on the same afternoon.
3. Measure what survives without JavaScript
A 200 response is not the same as readable content. View the raw source of your key pages and look for your actual words. The most common finding: a healthy-looking response of 10 to 70 KB whose visible text is a few hundred characters, an application frame waiting for scripts that AI agents do not reliably run. Everything your buyers would read is, to the machine, missing.
4. Check extractability and structure
Answer engines lift sentences out of context, so your important facts need to survive alone. Prices, credentials, comparisons and claims should exist as clear standalone sentences under question-shaped headings, with the entity named rather than pronouned. The most common finding: the key facts exist, but only inside tabs, accordions, images or marketing prose an engine cannot safely quote.
5. Check llms.txt, sitemap and schema
These are the files that help engines understand your site deliberately rather than by inference: an accurate sitemap, structured data that matches the visible content, and an llms.txt file mapping your highest-signal pages. The most common finding here is simple absence, which is not fatal but is a cheap advantage left unclaimed.
6. Test the actual answers
Ask ChatGPT and Perplexity, with search enabled, the questions your buyers ask: about your brand by name, and about your category without it. Note what is said, what is cited, and what is wrong. The most common finding: the brand is described from stale or third-party sources because the site itself gave the agents too little to work with.
What a technical audit for AI search adds
The afternoon version above finds the loud failures. A professional audit exists for the quiet ones, and for proof.
The full Wellknown Audit runs your site through a purpose-built engine that crawls as every relevant agent family, indexing crawlers and live-fetch agents, compares rendered against raw content page by page, and validates findings against genuine bot infrastructure rather than user-agent strings alone. It returns a scored verdict split across indexing visibility and live-citation visibility, request-level evidence for every finding, and a fix list ranked by impact, written in plain English with a technical report available for your developers. It costs £3,500, requires only your URL, and takes about a week. If your DIY pass found anything amber, this is the difference between knowing something is wrong and knowing exactly what to fix first. See what the audit includes.
How often should you audit AI search visibility?
Twice-yearly as a baseline, and immediately after anything that touches your infrastructure: a replatform, a CDN change, a redesign, new bot-protection settings. The reason is that AI search failures are regressions with no alarm attached. A firewall rule added on a Tuesday can lock every AI engine out of your site for a quarter, and nothing you monitor will mention it. That silence is also why we offer ongoing monitoring as a service: re-audits that catch a regression the week it happens rather than the quarter after.
For the broader review covering Google's AI surfaces alongside the assistants, see our GEO audit page, and for the concept underneath all of this, start with AI visibility.
Frequently asked questions
What is an AI search audit?
An AI search audit reviews whether AI systems can access, read and cite your website: crawler access, the content each agent is served, rendering survival, and the structure answer engines need in order to quote you.
What is an AI audit?
Outside search, "AI audit" usually means a review of an organisation's own AI systems for risk, bias and compliance. An AI search audit is narrower: it examines your website's visibility to AI-driven search, not your use of AI internally.
What is AI search and how does it work?
AI search is answering through systems like ChatGPT, Perplexity and Google's AI features, which combine trained knowledge with live retrieval: crawlers index the web in advance, and agents fetch specific pages in real time when a question needs them. Your visibility depends on both layers working.
How do I do an AI SEO audit?
Follow the six steps above: check robots.txt and CDN rules, fetch your pages as the agents, measure what survives without JavaScript, check extractability, check llms.txt and schema, then test real answers in the assistants. For scoring, rendering analysis and validated evidence, use a professional audit.
Start with the free check
The fastest way into your own audit is step two: run a page through the free Agent Parity Check and see what seven AI agents are served, side by side, in 30 seconds.
Run the free check