The complete GEO audit: how does AI see your brand today?

You have read the theory, you understand how AI search engines pick content, and yet one question lingers: where do you actually start? My answer is always the same. You start with an audit. Before you can optimise anything, you need to know where you stand today. And in the world of AI search results, most brands are staring at a blank wall: they have no idea whether ChatGPT or Perplexity ever mentions them. Why a correct AI picture matters for every brand is something you can read in GEO for everyone: how we make sure AI shows your brand correctly.

In this article I walk you through a complete GEO audit. GEO (Generative Engine Optimization) is the discipline that makes AI assistants cite your brand in their answers. The audit I describe below is deliberately practical: you finish it in weeks, not months, and you need no exotic tooling for it.

Want the broader framework first? Then read the ultimate GEO guide, which brings all the building blocks together. This article zooms in on one thing: how to establish your baseline today.

Check your own page: start with a free GEO score via our GEO check.

What a GEO audit actually is

A GEO audit is a systematic photo of your current AI visibility. It answers three questions at once: where are you cited, where are you not, and why do competitors score in the places where you are missing? The result is not a vague gut feeling but a concrete starting point with numbers, gaps and a priority list.

I always break the audit down into seven components. Each component builds on the previous one, and together they give a complete picture:

1. Map the prompt landscape
2. Measure current visibility
3. Extraction analysis of your content
4. Determine your competitive position
5. Evaluate brand salience
6. Check technical readiness
7. Gap analysis and prioritisation

Those seven components are not a loose checklist but a rising sequence: every step builds on the previous one, until you arrive at a prioritised gap list.

Below I walk through each component as a step-by-step plan.

The step-by-step plan: seven components

Step 1: Map your prompt landscape

Before you can measure visibility, you need to know which questions actually matter for your business. I always start from three types of prompts:

• Brand prompts: “What is [your brand]”, “[your brand] reviews”, “[your brand] vs [competitor]”.
• Product or service prompts: “[product] features”, “[product] alternatives”.
• Category prompts: “best [category] software”, “how do I [task]”. These are the battles you really want to win.

Expand that core with intent variants (informational, commercial, transactional) and modifiers such as “2026”, “best”, “for SMEs” or “for enterprise”. Then give every prompt a priority score based on business value, volume and feasibility. Document 50 to 200 prompts in a simple overview. That set becomes the measurement foundation for the rest of the audit.

Step 2: Measure your current visibility

Now that the prompt set is ready, you start measuring. For every priority prompt you submit the same question to each major AI platform: Google AI, ChatGPT, Perplexity and Claude. Each time, note whether you are cited, in what position, and which competitors appear. Repeat every prompt three to five times, because AI answers vary from run to run.

From that data you calculate a few core metrics:

• Selection rate: in what percentage of prompts are you cited?
• Citation share: what share of all citations goes to you versus the rest?
• Primary source ratio: when you are cited, how often are you the main source?

Break those numbers down per platform. ChatGPT, Perplexity and Google AI behave differently, and you want to know where you are strong and where your gaps are. For the practical setup of this kind of measurement, citation mining is a useful deep dive.

Step 3: Analyse what AI extracts from your content

Being cited is one thing, but knowing which passages a model uses is worth gold. Take the pages that received citations in step 2 and compare the AI answer with your source text. Which paragraphs are copied verbatim or paraphrased? Which sections never come out?

For each page, also calculate a rough coverage rate: how much of your text is actually extracted? You often see the same pattern: your definition section scores high, while your case study or conclusion is barely touched. Those are exactly the insights that later steer where you need to fine-tune. Also look at your non-extracted content and ask yourself: is it findable, relevant, well structured and competitive enough?

Step 4: Determine your competitive position

A number in isolation says little. A selection rate of 30 percent is strong if the market leader sits at 35, and weak if they sit at 80. That is why I always map the competition.

First identify your real competitors, but do not forget the content competitors: trade media, review sites and knowledge platforms that pick up your prompts without being your commercial rival. Then map, per priority prompt, who won and why. Often the answer is painfully simple: the competitor has an up-to-date comparison page with a feature matrix, and you only have a product page. Finish with an overview of your advantages (where you win) and your disadvantages (where you structurally lose).

Step 5: Evaluate your brand salience

Some models mention you even without your content being retrieved live, purely based on what they learned during training. I call that salience. Test this by asking category prompts without your brand name, with the search function switched off: “What are the leading platforms for [category]?” Are you mentioned spontaneously, and in what position?

Then test the associations: complete prompts such as “[your brand] is known for…” and see whether the model gets it right. Does the association match your positioning, or is a wrong or outdated picture attached to your brand? Compare this across several models, because salience differs strongly per model. If you want to map specifically where, outside your own site, the authority that feeds salience is missing, then look at a separate off-site authority audit.

Step 6: Check your technical readiness

The best content is useless if AI crawlers cannot reach it. So work through a technical checklist:

• Crawlability: does your robots.txt allow crawlers such as GPTBot and Google-Extended? Is content not stuck behind a login wall or heavy JavaScript?
• Indexation: are your key pages in Google and Bing? Are your sitemap and canonical tags correct?
• Structured data: do you use correct schema.org markup (Product, FAQ, HowTo) without errors?
• Content structure: do you have a clear heading hierarchy, and is your text actual text (not an image)?
• Performance: do your pages load fast enough and are they reliably available?

Give each element a score and calculate a technical readiness score. A single weak element can undermine all your content work, so this is not a detail.

Step 7: Run the gap analysis and prioritise

Now everything comes together. Collect the concrete gaps from every previous component: prompts with a low selection rate, missing content types, segments where competitors dominate, incorrect brand associations, technical blockers.

Score every gap on impact (how much does it deliver?) and effort (how hard is it?). That splits everything into four boxes: quick wins (high impact, little work), big projects, nice-to-haves and things you are better off dropping. You tackle the quick wins first. Pour the whole thing into a phased roadmap, so you know what has to happen in weeks 1 to 4 and what comes later.

Finish with a baseline document: record all your core numbers on the audit date. That starting point is what makes an audit valuable in the first place, because three months from now you want to be able to prove that your selection rate actually went up.

Checklist: is your GEO audit finished?

Use this checklist to make sure you skip nothing:

• Prompt set of 50 to 200 prompts built and prioritised
• Every priority prompt tested three to five times on every platform
• Selection rate, citation share and primary source ratio calculated
• Numbers broken down per platform (Google AI, ChatGPT, Perplexity, Claude)
• Extraction analysis done on your most cited pages
• Competitors and content competitors mapped
• Win/loss analysed per priority prompt
• Brand salience tested with brandless category prompts
• Brand associations checked for accuracy
• Technical checklist completed (crawl, index, schema, structure, speed)
• Gaps scored on impact and effort
• Phased roadmap drawn up
• Baseline with all core numbers recorded on the audit date

Work through this list and you will have no loose observations but a real foundation to build on. Want to turn the measurement into a repeatable report afterwards? Then read how to create an AI visibility report, and which AI visibility tools take the work off your hands.

Frequently asked questions

How long does a complete GEO audit take?

For a first full audit I count on roughly five weeks: one week for the prompt landscape, one for the visibility measurement, one for content and competitor analysis, and the final weeks for salience, technical checks and synthesis. If you only want a quick baseline, you can run a slimmed-down version in a few days by limiting yourself to the top prompts and the core metrics.

Do I need expensive tools to do this?

No. You can run the basics entirely by hand by submitting prompts to ChatGPT, Perplexity, Claude and Google AI, and noting the citations in a spreadsheet. Tools mainly speed up the repetitive work and the repeated measuring. Start manually, so you get a feel for what the data looks like, and only then automate the parts you redo every month anyway.

How often should I repeat the audit?

The full audit is a snapshot you rerun once or twice a year. Core metrics such as selection rate and citation share are better measured monthly or quarterly, because that is how you see trends emerge. Citation data is noisy by nature, so do not let one dip steer you: only a trend across two to three measurements is a real signal.

What do I do first with the results?

Always start with the quick wins from your gap analysis: technical fixes that unblock visibility and missing content you can fill in quickly, such as a comparison page. Then align your priorities with the stakeholders and set up a fixed measurement system. That way the audit becomes the starting point of an ongoing optimisation process instead of a report that disappears in a drawer.

Need help?

Want to translate this into execution? See how we approach this with AI visibility.