An AI-visibility audit checks whether AI engines can reach, read, and quote your pages, not whether you rank a link. It has five parts: crawler access, extractability, product data, off-site corroboration, and quotability. Do them in that order, because a page an engine can't fetch can't be cited no matter how good it is, and a brand no third-party source mentions won't be recommended no matter how clean its schema.
What does an AI-visibility audit check?
Classic SEO audits grade a page's ability to rank a blue link. An AI-visibility audit grades a different chain: can an answer engine discover your content, extract a clean passage, understand your products, verify your brand against other sources, and lift a sentence into its answer. Each link can fail independently, so audit them separately.
| # | Check | The question it answers | Fails when |
|---|---|---|---|
| 1 | Crawler access | Can AI engines fetch your pages at all? | CDN/robots return 403 to AI bots |
| 2 | Extractability | Can an engine parse and cite a clean chunk? | JS-only content, no schema, no <h1> |
| 3 | Product data | Can shopping agents read your catalog? | No Product/Offer schema |
| 4 | Corroboration | Do off-site sources confirm your brand? | Thin/absent on Wikipedia, Reddit, reviews |
| 5 | Quotability | Would an AI lift an answer from your prose? | Buried answer, vague marketing fluff |
A page an AI can't fetch can't be cited; a brand no other source mentions won't be recommended. Structure and prose only matter once those two gates are open.
Step 1: Can AI crawlers reach your pages?
Fetch your key pages while identifying as the major AI crawlers and confirm you get HTTP 200, not 403. Check GPTBot (feeds ChatGPT), ClaudeBot (feeds Claude), OAI-SearchBot (ChatGPT Search), and PerplexityBot (Perplexity) against your homepage plus one example of each page type. A silent CDN block is the single most common reason a brand is never cited, and it never shows in normal analytics. While you're here, confirm robots.txt welcomes those user-agents and your sitemap.xml exists and is current. The crawler directory lists the exact tokens.
Step 2: Can an engine extract and cite your pages?
Most AI crawlers never execute JavaScript. They read the raw HTML. So measure how much real content exists before any script: if your content depends on JavaScript, an engine may see an empty shell. Then grade three structural signals per page: valid structured data (a real Organization block with sameAs links to your profiles, plus BlogPosting/FAQPage where relevant), one <h1> per page with labelled sections and image alt text so an engine can segment the page, and reasonable HTML weight so your schema isn't buried past a crawl budget. Pages that are readable to agents as clean, semantic HTML are the ones that get quoted.
Step 3: Can shopping agents read your products?
If you sell, check every product template for Product and Offer schema exposing name, price, and availability. The exact fields AI shopping agents read. A product page that looks perfect to a human but carries no machine-readable price or stock status is invisible to an agent composing a "best [category] under ₹X" answer. This check is skippable for a pure publisher, but decisive for D2C and ecommerce.
Step 4: Do off-site sources corroborate you?
AI engines lean on how well-established you are as an entity, often more than on your own pages. Audit where engines can verify you beyond your site: is there a Wikipedia or Wikidata entry, and does it list your official website; are you present and active on the platforms your customers cite (Reddit, LinkedIn, YouTube, Instagram); do independent reviews and roundups mention you. Gaps here are corroboration gaps. The slow, durable lever behind entity strength. "Not found" on a source your competitors dominate is an opportunity, not a verdict.
Step 5: Would an AI actually quote your writing?
Grade your best content pages. Products, articles, guides. Against three lenses an engine implicitly applies. Gets to the point: does the page state its core answer in the first visible text, before navigation and marketing preamble? Makes sense on its own: could a single paragraph be quoted alone and still parse. Named subject, no dangling "it" or "we"? Facts, not fluff: concrete numbers, specs, and comparisons versus adjectives and slogans. A page can pass every technical check and still never get lifted because the answer is buried under a promotional banner.
How often should you audit?
The structural checks (1-3) change only when you ship a release or a CDN rule, so re-audit monthly or quarterly. Corroboration (4) moves slowly. Review it quarterly. But whether you're actually named and cited shifts constantly, so pair the deep audit with the lighter one-hour weekly routine and read both as trends. An audit tells you whether the doors are open; the weekly snapshot tells you who's walking through. If you find a gap, the two-lens split, not named versus not cited. Tells you which fix to queue first.
Do the audit, then keep it fresh
Start with Step 1 today: it's ten minutes and the most common silent failure. Then work down the list once, fixing the highest-leverage gap each pass. Doing all five checks by hand across a whole site. Probing every AI crawler, grading schema and structure on each template, searching every off-site source, and scoring quotability page by page. Is exactly what Buffy Intel's Site Audit does automatically: it reads your pages the way AI crawlers do, scores each pillar 0-100, shows the specific fixes, and refreshes on a monthly cadence so you catch regressions early. Point it at your domain to see where the doors are already open and where they're quietly shut.