Every GEO engagement I take starts the same way, because every client asks the same first question: where do we actually stand?
A GEO audit answers that with evidence instead of vibes. It is not a 90-point technical checklist and it is not a paid-tool dashboard screenshot. It is six focused investigations that end in a prioritized fix list.
This is the process I run, in order, with the free prompts from my library that automate the repetitive parts. Run it yourself in a day at small scale, or use it to pressure-test whatever an agency is proposing to charge you for.
TL;DR
- Six investigations, one output: A GEO audit runs six focused checks that end in a prioritized fix list, not a score or a dashboard screenshot.
- The six parts: Establish a visibility baseline, audit the entity, review content extractability, check technical access, inventory the third-party record, and run the competitor gap.
- Baseline first: I start by running 30 to 50 real buyer prompts across ChatGPT, Gemini, Perplexity, and Google, because every later step gets prioritized by what those answers reveal.
- You can do it free: The core audit needs only the AI engines, a spreadsheet, and Google Search Console, since the thinking is the audit and the thinking is free.
- Prioritize by impact: Sort every finding by whether it changes an answer a buyer actually sees, so a robots.txt block that hides you outranks any content tweak.
What is a GEO audit?
A GEO audit is a structured assessment of how AI search engines see your brand: whether ChatGPT, Google AI Overviews, Gemini, and Perplexity mention you for the questions your buyers ask, whether they describe you accurately, and whether they cite your pages as sources.
The deliverable that matters is a prioritized fix list. If an audit ends in a score instead of a list of specific changes with owners, it was a sales document.
Step 1: Establish the visibility baseline
Start by measuring, because every later step gets prioritized by what you find here.
Write 30 to 50 prompts your real buyers would ask. Three types: category questions (“best [your category] for [your customer]”), problem questions (“how do I fix [the problem you solve]”), and brand questions (your name alone, your name plus your service). Run them across ChatGPT, Gemini, Perplexity, and Google, and log four things per answer: mentioned or not, cited or not, sentiment, competitors named.
My citation share audit prompt structures this whole exercise, and the brand-in-answer check is the quick single-query version. At scale, the monitoring tools I compared previously automate the loop, but do the first pass by hand. Reading the actual answers is where the insight lives.
Step 2: Audit the entity
AI engines can only recommend what they can describe. This step checks whether a machine reading your web presence can answer three questions without ambiguity: what is this business, who does it serve, what problems does it solve.
Read your homepage, about page, and service pages as if you knew nothing. Then check your LinkedIn, your Google Business Profile, and your top directory listings. You are hunting for two failure modes: vagueness (positioning that lives in mood language) and inconsistency (three surfaces describing three different businesses).
The entity clarity rewrite prompt scores a page and rewrites it in machine-literal language. A branded SERP audit shows you what currently surfaces when someone, or something, looks you up.
Step 3: Review content extractability
This is a passage-level read of your most important pages. For each, ask: which question does each section answer, is the answer in the first two sentences under the heading, and could the passage be lifted out and quoted on its own?
Most sites fail this not because the content is bad but because the answers are buried: three paragraphs of context before the claim, headings that are labels instead of questions, key facts living only in paragraphs that depend on earlier ones. The passage extraction rewrite prompt retrofits existing paragraphs into liftable blocks.
While you are in the content, check coverage against the question tree. AI Mode’s query fan-out retrieves at the sub-question level, so every buyer sub-question without a page is a branch you forfeit.
Step 4: Check technical access
Short, mechanical, and the source of the worst silent failures:
- robots.txt: review each token according to what it controls. OAI-SearchBot, GPTBot, ClaudeBot, Claude-SearchBot, and PerplexityBot are crawler or fetcher tokens. Google-Extended is different: it is a robots.txt product control, not a separate HTTP crawler, and governs certain Gemini training and grounding uses. A disallow rule can limit the affected system’s access or use, but it does not make a brand universally invisible. Worth checking even if you never added a block, because security plugins and CDN bot rules sometimes do it for you.
- Indexation: AI answer eligibility starts with being in the index. Spot-check money pages in Search Console.
- Structured data: valid, matching the visible content, on the page types that matter. Article, Organization, Person, FAQPage. The schema generator prompt builds JSON-LD from a page’s actual content.
- Speed and rendering: retrieval systems fetch pages like impatient users. Content that requires heavy client-side rendering to appear is content some retrievers never see.
Step 5: Inventory the third-party record
AI engines triangulate. What the rest of the web says about you weighs as much as what you say, and for trained model knowledge it often weighs more.
Inventory the sources: directory listings (accurate? current? duplicated?), review platforms (present? responded to?), press and citations, and community discussion. Reddit and niche forums punch far above their weight in AI answers, and the Reddit and forum citation finder maps where your category gets discussed.
Flag every source that is wrong, stale, or missing. This list feeds the correct-the-record process and it is usually the highest-effort, highest-payoff section of the whole audit.
Step 6: Run the competitor gap
Finish by turning the baseline into strategy. For every prompt where a competitor was named and you were not, ask why. Usually it is one of three answers: they have a page that directly answers the question and you do not, they have a stronger third-party record on that topic, or they get cited by a source the engines lean on (a comparison site, a community thread, a directory).
The competitor citation comparison prompt structures this head-to-head. The output slots straight into a content plan: which pages to create, which to restructure, which external sources to pursue.
Turning the audit into a plan
Prioritize findings by the customer decision affected, the evidence available, and the work needed to address them. A verified access problem on an important page may deserve attention before a cosmetic edit. Record uncertainty instead of treating every observation as a proven cause.
Here is an illustrative handoff, not a client case study:
| Finding and evidence | Next action | Owner and effort | How to verify |
|---|---|---|---|
| A target service page returns an error to the tested crawler; preserve the status and request date | Investigate the CDN rule or route before changing content | Developer; estimate after reproducing the cause | Repeat the same request and confirm the page body is accessible |
| The offer omits its intended customer and deliverable; quote the missing context alongside the buyer question | Revise the service page using approved facts and a clear inquiry path | Marketing and sales; small content change | Review comprehension with sales and monitor qualified inquiries |
| A sampled answer cites an outdated business description on a named source | Check the cited page and request a correction where you control or can verify the record | Marketing or account owner; depends on source access | Confirm the source changed, then rerun the same question over multiple dates |
In your own sheet, add the affected URL, observation date, baseline, due date, and dependencies. Avoid an impact score that implies more precision than the evidence supports. If the data is incomplete, make verification the next action.
Set the review date before making changes. Reuse the same prompt panel, platforms, markets, and recording method, and keep a separate change log. A later citation change is an observation; it does not by itself prove your edit caused it. See how to measure GEO for the measurement questions.
If your team needs help defining the scope or sequencing the work, my SEO and AI search strategy engagement covers the baseline, findings, and prioritized handoff.
Frequently asked questions
What is a GEO audit?
A GEO audit is a structured assessment of how visible, accurate, and citable a brand is across AI search surfaces like ChatGPT, Google AI Overviews, Gemini, and Perplexity. It covers six areas: a visibility baseline from a fixed prompt panel, entity clarity, content extractability, technical access for AI crawlers, the third-party record the engines learn from, and a competitor gap analysis. The output is a prioritized fix list, not a score.
How long does a GEO audit take?
A focused GEO audit on a small to mid-size site takes one to two weeks of calendar time. The visibility baseline takes a few hours of running and logging prompts. Entity, content, and technical reviews take a few days depending on page count. The third-party record review is the most variable part, since directory and review-site sprawl differs wildly by business. A solo operator can run a useful first pass in a single day.
Can I do a GEO audit myself for free?
Yes. The core of a GEO audit requires no paid tools: run a fixed panel of buyer prompts across the AI engines manually, check your robots.txt for AI crawler access, review your key pages for direct answers under question-shaped headings, and search your brand name to inventory the third-party record. Paid AI visibility tools speed up the baseline at scale, but the thinking is the audit, and the thinking is free.
What should a GEO audit include?
Six components: a brand-in-answer baseline across ChatGPT, Gemini, Perplexity, and Google AI Overviews; an entity clarity review of how consistently your positioning reads across your site and profiles; a content extractability review at the passage level; a technical access check covering robots.txt, crawlability, and structured data; a third-party record inventory of directories, reviews, and community mentions; and a competitor citation comparison showing who wins the answers you are losing.
How often should I repeat a GEO audit?
Run the full audit annually or after major site changes, and rerun the visibility baseline monthly. The baseline is the moving part: AI engine answers shift with model updates and fresh retrieval, while entity, technical, and third-party findings change slowly. A monthly prompt panel plus an annual deep audit catches both speeds of change.
What tools do I need for a GEO audit?
Minimum kit: access to ChatGPT, Gemini, and Perplexity, a spreadsheet, and Google Search Console. Optional additions: an AI visibility platform for baseline automation, a rank tracker for the traditional side, and a schema validator. The audit’s value comes from the fix list it produces, and most of the fixes are content and consistency work no tool does for you.
This six-step process is the first deliverable of my AI Search Visibility & SEO Strategy service, run for you with the findings explained in plain English. Prefer to talk it through first? Book a free 30-minute call.
