GEO audit: how to run one, step by step

Search has quietly split in two. Next to Google's familiar list of blue links sits a second surface: the written answer of an assistant like ChatGPT, which summarises a topic and then cites a handful of sources. The conceptual side of this, what a GEO audit is and the dimensions it scores, is covered in depth on our GEO audit pillar. This page is deliberately different. It is the operator's manual: a repeatable checklist you can run yourself, in an afternoon, to see whether the AIs cite you and to decide what to fix first.

What a GEO audit checks, in one minute

A GEO audit (Generative Engine Optimization audit) checks three things in order: can AI engines find your pages, can they pull a clean answer out of your content, and do they trust your sources enough to cite you. Failing any one of the three is enough to keep you out of the answer.

The term comes from a 2024 research paper, GEO: Generative Engine Optimization, from a team including researchers at Princeton and the Allen Institute for AI, presented at the ACM SIGKDD conference. It was the first rigorous framework for making content visible inside generative answers. A GEO audit is the practical front end of that discipline: before you optimise, you measure where you stand.

The three checks mirror how an assistant actually builds an answer. It first retrieves candidate pages through a web search, reads them, extracts passages, and then cites what it genuinely used. Each stage is a filter, and each filter is one axis of the audit: findability, extractability and credibility. The same paper found that adding cited statistics and credible sources could lift visibility in generative answers by roughly 40 percent, while old-school keyword stuffing did nothing. That is why credibility is a check, not a nicety.

Before you start: what you need

To run a GEO audit you need three free things: the assistants themselves with web search turned on, a way to view your robots.txt and rendered HTML, and a simple spreadsheet to record one row per query. No GEO-specific dashboard exists for the public in 2026, so the work is method, not software.

Set aside an afternoon for the first pass and a recurring slot afterwards. Open ChatGPT and Perplexity, and confirm web search is enabled in each, because without it you are testing the model's memory rather than what it cites today. Have a browser tab ready to check yoursite.com/robots.txt, and know how to view the rendered source of a page so you can confirm your answer text is actually in the HTML, not hidden behind JavaScript the crawler never runs. Finally, make a plain table with one row per query and columns for cited yes or no, who else was cited, and the form of the mention. That table is your whole audit.

The five-step GEO audit, in order

Run the audit in five steps: build your real query list, measure your baseline citation rate, check findability, check extractability and credibility, then prioritise and re-test. Do them in this order, because a beautifully written page that is blocked from retrieval will never be cited no matter how good it reads.

Each step has a clear output, so you always know whether you finished it. Work top to bottom; do not skip ahead to fixing content before you have confirmed the pages can even be retrieved.

1. Build your real query list
   Write down 10 to 20 questions your customers actually ask before buying or choosing, phrased the way a human types them into an assistant, not as generic keywords. This single step decides the value of everything that follows: audit the wrong questions and the result is worthless. Pull them from your sales calls, your support inbox and your search query reports.
2. Measure your baseline citation rate
   Submit each query to the assistants with web search on, and record, query by query, whether your brand is cited, who else is cited, and how (a link, a named mention, a paraphrase). This snapshot is your starting point, far more useful than a composite index. Because the same question can give different answers from one session to the next, test cleanly and note the variation rather than trusting a single run.
3. Check findability
   For every query where you are absent, ask first whether you were even retrievable. Confirm the relevant page is indexed, that useful AI crawlers are not blocked in robots.txt, and that your answer text lives in the rendered HTML. ChatGPT, for instance, leans on Bing's index plus its own OAI-SearchBot, so a page invisible to retrieval is never a candidate. Our guide on AI crawlers and invisible websites lists the most common traps here.
4. Check extractability and credibility
   Where you were retrievable but still not cited, look at the page itself. Does each section open with a direct answer that stands on its own, or is the point buried in the third paragraph? And are your key claims backed by named sources and figures, or left as assertions? Models prefer passages they can lift cleanly and claims they can trust, which is exactly what our piece on E-E-A-T and AI content digs into.
5. Prioritise and re-test
   Rank the fixes from highest return to heaviest: technical unblocks that exclude you outright come first, then quick direct-answer restructuring, then deeper content and authority work. Ship them in that order, and re-run the same query list on a regular cadence so you can compare like with like and see what actually moved.

The one-page GEO audit checklist

Use this checklist to run a GEO audit end to end. Each row maps to one of the three checks, with a concrete pass condition you can verify yourself, so the audit produces evidence rather than opinion.

Print it, or paste it into your spreadsheet. The point of a checklist is that it is boring and complete: it stops you from declaring victory after the easy steps and forgetting the one that actually blocks your citations.

How to read what you find

Read a GEO audit by looking at the decomposition, not the average. Being absent on queries where competitors are cited is a strong signal; being absent where nobody is cited is weak. Likewise for causes: a technical block is fixed in hours, topic authority is built over months.

The trap, when you have the spreadsheet in front of you, is to fixate on a single global number such as "I'm cited 30 percent of the time". The average is a useful marker, but the rows are where the decisions live. Sort by the queries where a competitor took your place and you have your priority list. Then attach a cause to each gap, because the cause sets the cost and timeline of the fix.

Read well, a GEO audit is not a report card. It is a map. It tells you where the holes are, which ones cost the most, and where to start filling them. The rest is execution.

Five mistakes that ruin a GEO audit

The five most common GEO audit mistakes are auditing generic keywords instead of real questions, running each query once and trusting the result, skipping the findability check, chasing the average score, and stopping at the audit without fixing anything. Each one quietly invalidates the work.

I see the same handful of errors across the sites I test by hand. They are easy to avoid once named.

• Auditing keywords, not questions. People type full questions into assistants. Testing "best CRM" instead of "what's the best CRM for a small accounting firm" produces a result that has nothing to do with how your customers actually ask.
• Trusting a single run. One query, one answer, one conclusion is unsafe, because the same prompt varies between sessions. Test each query cleanly and note the spread.
• Skipping findability. Rewriting content before confirming the page is even retrievable wastes the heaviest work on a page no engine can see.
• Chasing the average. A composite score hides the only thing that matters, which is where you lose to a named competitor.
• Auditing and stopping. An audit with no execution behind it is a tidy way to feel busy. The value is in the fixes that follow, shipped in order.

Avoiding these five is most of the difference between a GEO audit that changes your visibility and one that just produces a document. For the optimisation side that follows the audit, our practical method for appearing in ChatGPT and Google AI is the natural next read, and the broader GEO vs SEO comparison sets the two disciplines side by side.

Alexis Dollé

CEO & Founder of Cicero Studio

I audit AI visibility the slow way, by hand, query after query, across dozens of sites on their real business questions. This checklist is the method I actually use at Cicero Studio, stripped of jargon. My view: a GEO audit only earns its keep if it ends in prioritised fixes and a measurement you repeat, not in a score that flatters.

LinkedIn →

What this guide does not cover

For honesty, and because it is exactly the kind of transparency AI engines reward, here are the limits to keep in mind before you build a strategy on a GEO audit. Saying what a method does not do is usually worth more than overselling what it does.

Going further

We document our approach in the open, because that is our best proof. Each resource below digs into one angle of the audit and of visibility in AI engines, from the credibility evidence to the crawler mechanics. Pick whichever matches your next question.