Guide
AI visibility audit: what it is and how to run one
An AI visibility audit measures one thing: whether an AI assistant can fetch your site, understand it, and repeat your facts back to a buyer. It is not an SEO audit — ranking in Google and being quoted in an answer are different problems. Here is the audit, in the order to run it, and what a passing result looks like.
1. Retrievability — can the assistant fetch the page at all?
Start at the door. Fetch your own page with a plain HTTP client and no JavaScript, then read what comes back. If the body is an empty shell, assistants see an empty shell too.
robots.txtallows GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Google-Extended.- The page returns 200 quickly, with no redirect chain or bot challenge.
- Main content is in the server HTML, not injected after hydration.
- A sitemap exists and lists every page you want quoted.
2. Machine readability — is the meaning explicit?
Models do better with declared facts than inferred ones. Check for a single clear H1, a descriptive title and meta description, and JSON-LD structured data that matches the page: Organization or LocalBusiness on the homepage, Product or Service with price where you sell, Article on guides.
Also check the plain things: real headings instead of styled divs, alt text on images that carry information, and a consistent business name, address and description across every page.
3. Answer readiness — can a sentence be lifted straight out?
Assistants quote extractable statements. For each of your top five buying questions, your site should contain a self-contained answer in the first two sentences of a section — with the specifics included, not linked to.
- Prices, availability, coverage area and turnaround stated in text.
- Question-shaped headings that mirror how people ask.
- Short paragraphs and lists rather than long narrative blocks.
- An
llms.txtsummarising the business and pointing at your most useful pages.
4. Citations and trust — does anyone else confirm it?
Models weight third-party corroboration heavily. Audit how your business appears on the directories, review sites and reference pages that dominate your niche, and check that the name, address and description match your site exactly. Mismatched details are a common reason an assistant hedges or names a competitor instead.
5. Freshness — is the page still current?
Stale pages get dropped from answers. Keep visible last-updated signals, refresh prices and dates, and remove content describing products you no longer sell.
6. Test the answer itself
Finish by asking ChatGPT, Claude and Perplexity the questions your customers ask — "best X in Y", "how much does Z cost" — and record whether you appear, what is said about you, and which source is cited. That transcript is your baseline. Re-run it monthly; the answer shifts as models re-crawl.
What a good score looks like
GEOSCOPE X weights the audit the same way: retrievability 30%, machine readability 30%, answer readiness 28%, freshness and trust 12%. Under 40 means assistants largely cannot use your site. Between 40 and 70 you appear inconsistently. Above 70 you are quotable, and the remaining work is third-party citations.
Run the audit automatically
GEOSCOPE X runs every check above in about a minute, scores each category, shows how ChatGPT, Claude and Perplexity would answer about you, and hands back the exact fixes — including ready-to-paste structured data and an llms.txt. The first scan is free.
Run a free scan