TL;DR
To audit your own AI search visibility, run five checks in about twenty minutes: (1) ask the major AI engines about your business by name and by category, (2) confirm your site declares one clean Organization entity in its structured data, (3) check whether you exist, correctly, in the Wikidata knowledge graph, (4) verify your schema, canonical tags and redirects, and (5) read your robots.txt to see which AI crawlers you let in. Every check is free. Together they tell you whether you are a candidate to be named in ChatGPT, Google AI Overviews, Perplexity and Claude, or invisible to them.
You do not need an agency to find out whether AI search engines can see your business. This is the check we run on our own site, written so you can run it on yours.
Why audit your own AI visibility at all?
Traditional search asked one question: does my page rank for this keyword? Generative search asks a different one: when an AI engine assembles an answer, is my business one of the entities it is confident enough to name?
The mechanics differ. You can rank on page one of Google and still be absent from the answer ChatGPT gives when someone asks it to recommend a business like yours. One system matches pages to queries; the other assembles entities into answers. If you have only ever optimised for the first, you have no idea how you look to the second. (If you want the background first, start with our complete guide to GEO optimisation or GEO vs SEO.)
The good news is that most of what decides your AI visibility is public and checkable. You can inspect your own site the way a competent GEO practitioner would. And if you are weighing up an agency, the companion piece, how to compare GEO agencies, applies the same checks to them.
The 20-minute GEO self-audit: five checks
Work through these in order. You need a browser, about twenty minutes and no paid tools. Keep a note open and mark each check "pass", "partial" or "fail". The pattern across all five tells you more than any single result.
Check 1: Ask the AI engines about yourself (5 minutes)
Start with the outcome, not the machinery. Open ChatGPT, Google (with AI Overviews), Perplexity and, if you have access, Claude. Ask each the same three questions using your real details:
- "What is [your exact business name]?"
- "Who are the [your category] businesses in [your location]?"
- "Tell me about [your business name] and what they offer."
What you are looking for. Does the engine know you exist? Does it describe you accurately: right services, right location, right founding facts? Does it confuse you with a similarly named company? Does it name you in the category question, or only when you name yourself first?
How to read the result. Being described accurately when asked by name is table stakes. Being named unprompted in the category question is the real prize: the engine holds you as a candidate for that answer. If the engines cannot describe you, or describe a different company with your name, you have an entity problem, and Checks 2 and 3 are where it lives.
Check 2: Inspect your entity declaration (4 minutes)
Open your homepage and view the page source (right-click, then View Page Source, or type view-source: before your URL). Search the code for application/ld+json.
What good looks like. A single, clean Organization (or ProfessionalService) node with your legal business name, a description, your location and a sameAs list linking to at least two authoritative external records: Companies House, LinkedIn, Wikidata, Crunchbase. sameAs is what lets an engine cross-confirm that you are a real, distinct entity.
Common failures. No ld+json at all (invisible entity), several conflicting Organization nodes (confused entity), or a name in the schema that does not match the name on the page or on your external records (weakened entity).
One caveat. If your site builds pages in the browser with JavaScript, the source view may not show everything a visitor's browser ends up with, and a check that only reads the source can report something missing that is actually there, or the reverse. If you find nothing, confirm with Google's Rich Results Test (Check 4), which renders the page before it reads it.
Check 3: Check the Wikidata knowledge graph (3 minutes)
Go to wikidata.org and search your business name. Wikidata is the open, machine-readable record that many AI systems lean on to work out who is who.
What you are looking for. An entry that is actually yours (correct founding year, sector and location), not a similarly named company, and not nothing. Then read what the record says you are: the "instance of" line. It is easy to get wrong. When we last ran this check on ourselves, our own record carried a stray classification alongside "business", the kind of slip that tells a machine something false about you while looking fine to a human.
How to read the result. No entry, or a thin one, is a partial at best. A correct, populated entry that your schema's sameAs also points to is a strong pass: it closes the loop between your site and the knowledge graph.
Check 4: Verify schema, canonicals and redirects (5 minutes)
Run your homepage and one key service page through Google's Rich Results Test (free) to confirm your structured data validates without errors.
Then check two things: that each important page has a single rel="canonical" tag pointing to itself, and that old URLs redirect cleanly with 301s rather than landing on soft-404 pages. Conflicting canonicals split the signals an engine uses to decide which version of a page to cite; broken redirects waste authority you have already earned.
What good looks like. Schema validates, one self-referencing canonical per page, and old links land on real content via 301. Unglamorous plumbing, and the difference between an engine citing your page and quietly skipping it.
Check 5: Read your robots.txt (3 minutes)
Type your domain followed by /robots.txt (for example, yourbusiness.co.uk/robots.txt) and read what is there.
What you are looking for. The AI companies run different crawlers for different jobs, and the distinction matters more than most robots.txt advice admits:
- Search and answer crawlers fetch pages so an assistant can cite them in a live answer, for example OAI-SearchBot (ChatGPT search), PerplexityBot and Claude-SearchBot. Block these and you are asking not to be cited.
- Training crawlers collect content to train future models, for example GPTBot (OpenAI), ClaudeBot (Anthropic) and CCBot (Common Crawl). Blocking these is a separate decision with separate trade-offs.
- Google is its own case. Google AI Overviews are part of Google Search and use the normal Googlebot. The Google-Extended token controls whether your content is used for Google's Gemini models; it does not remove you from Search or from AI Overviews.
Check each AI company's own crawler documentation before you edit, because names and roles change. See Google's list of common crawlers and OpenAI's bots overview.
The judgement call. Allowing AI crawlers is a decision, not an automatic yes; some businesses have good reasons to restrict training access. But if your goal is AI visibility, blocking the crawlers that fetch pages for answers is self-defeating. Know what your robots.txt says, and make each choice on purpose rather than inheriting it from a template.
Reading your results: the pattern matters more than the score
Lay your five notes side by side. A common and fixable pattern is a business that passes Check 5 and half of Check 4 but fails Checks 2 and 3: the plumbing is fine, but there is no clean entity declaration and no knowledge-graph record. That business is crawlable but not resolvable. The engines can read the site but cannot confidently say what it is. The fix is entity work, and for most small businesses it is the highest-leverage place to start.
If you failed Check 1 but passed 2 to 5, you may simply be early. The structural signals are in place and need time, and outside corroboration, to compound.
A warning about where that corroboration comes from, because we got this wrong ourselves. The obvious move is to chase inclusion in the "best [your category] agencies" lists that rank for your commercial terms. We spent nineteen weeks doing exactly that and got nothing, and when we finally read those pages properly rather than pitching them, the reason was structural. In many competitive categories, those lists are published by agencies competing in them. The compiler is not a neutral editor; including you demotes someone and introduces a rival into a document that generates their leads.
Where there is a way onto one of those lists, it is often a paid one. We know because we were offered a rate for a placement, not a hearing. That can be a defensible purchase, but it is advertising, and it should be bought and measured as advertising rather than mistaken for editorial endorsement.
The corroboration that compounds is the kind nobody controls: being cited because you published something worth citing. It is slower, and it is the one that is actually open to you.
If you failed everything, that is not a crisis. It is a clean starting point, and the order above is also the order to fix things in.
What to do once you have your results
This self-audit tells you where you stand. Turning it into a prioritised plan is the next step, and much of it is doable in-house if you have the time and technical confidence. Every check above can be re-run monthly to track movement.
If you would rather have the five checks run for you and returned as a prioritised, plain-English plan, including the entity and knowledge-graph work that is hardest to do alone, that is the starting point of our GEO optimisation work. Run the audit yourself first, then talk to us about what you found.
FAQ
How do I know if my business shows up in ChatGPT?
Ask ChatGPT directly: "What is [your business name]?" and "Who are the [your category] businesses in [your location]?" If it describes you accurately by name, you have basic visibility; if it names you unprompted in the category question, it holds you as a candidate entity, which is the stronger position. If it cannot describe you or confuses you with another company, start with your entity declaration.
Is checking my AI search visibility different from checking my Google ranking?
Yes. Google ranking measures whether your page matches a keyword query. AI visibility measures whether generative engines hold your business as an entity they are confident enough to name in an assembled answer. You can rank well and still be absent from AI answers, which is why a separate self-audit is worth running.
How often should I run a GEO self-audit?
Monthly is a sensible cadence. AI engines change how they retrieve and describe businesses without notice, and fixes to your schema or knowledge-graph record take time to show up. A monthly re-run is the simplest way to see whether your changes are working.
What is the single most important check?
Entity declaration (Check 2) combined with a correct Wikidata record (Check 3). Many businesses that are invisible to AI search are crawlable but not resolvable: the engines can read the site but cannot confidently say what the business is.
Do I need to allow AI crawlers in my robots.txt?
If you want to be cited, do not block the crawlers that fetch pages for live answers, such as OAI-SearchBot, PerplexityBot and Claude-SearchBot. Training crawlers such as GPTBot and ClaudeBot are a separate choice. Google AI Overviews use the normal Googlebot; blocking Google-Extended does not remove you from them.
Can I do all of this myself, or do I need an agency?
You can run the full audit yourself in about twenty minutes with free tools, and act on much of it in-house. The parts most businesses find hardest alone are entity declaration in structured data and getting a Wikidata record right. Run the self-audit first either way, so you know what you are buying.