How long does this take?

Typically one focused day: an hour for topics and phrasings, a few hours of running and recording, an hour of aggregation. The re-run is faster because the structure exists.

Should I use my normal accounts?

Fresh or logged-out sessions where possible. Your account history personalizes answers toward what you already looked at, and you are auditing what buyers see, not what you see.

What if we never appear anywhere?

That is a common and useful first result. It means the engines' picture of your category was formed without you, and the source pattern from step 6 tells you exactly which pages formed it. Start there.

The One-Day GEO Audit

Before strategy, before tooling, before budget: find out where you actually stand. This chapter is a first audit you can run this week with a spreadsheet, a browser and a few focused hours. It will not be laboratory-grade. It will be honest, structured and repeatable, which already puts it ahead of most of what passes for AI visibility checking.

What you are trying to learn

Four questions, in order of importance:

When buyers ask the questions that lead to your category, do you appear at all?
When you appear, what do the engines say about you, and is it accurate?
Who appears instead of you, and which sources make the engines say so?
How much does all of this move between runs, so you know what a real change will have to look bigger than?

Step 1: Pick five topics

From chapter 6: five clusters of buying intent, the questions that end in someone shortlisting a product like yours. Name each in plain language a leadership team would recognize. Resist starting with ten; a five-topic audit you finish beats a thirty-topic one you abandon.

Step 2: Write three phrasings per topic

For each topic, write the question the way three different real people would ask it: the blunt one, the detailed one with context, the skeptical one comparing options. Fifteen prompts total. Write them as full conversational questions, not keywords.

Step 3: Choose two or three engines

Start where your buyers are: typically ChatGPT plus Google’s AI answers, adding Perplexity, Gemini or Copilot if your audience leans that way. Use fresh sessions for each run, so earlier answers do not steer later ones.

Step 4: Run everything at least three times

Each prompt, each engine, three separate runs in fresh sessions, ideally spread across a day or two. Yes, that is 90 to 135 answers for five topics. This is the step that separates your audit from a screenshot safari, and it is exactly the step everyone skips. Three runs is the floor at which you can begin to see the difference between “we are absent” and “we caught a bad draw”.

Step 5: Record the same things every time

One spreadsheet row per answer:

Date, engine, topic, phrasing, run number
Mentioned: yes or no
If mentioned: roughly where (lead recommendation, mid-list, afterthought) and the sentence used
Sentiment: recommended, neutral, caveated, or wrong (quote the wrong claim exactly)
Competitors named, in order
Sources cited, with URLs

Twenty minutes of setup, and every future audit becomes comparable to this one.

Step 6: Read rates, not rows

Now aggregate, per topic:

Your mention rate: answers mentioning you divided by total answers.
Competitor mention rates: same arithmetic, for everyone named.
Spread: did the three runs of a prompt agree? Topics where they wildly disagree are unstable ground; note that instability rather than trusting whichever run flattered you.
Sentiment pattern: consistent story, or different claims in different answers?
Source pattern: which handful of pages keep being cited across answers in your category?

Resist reading any single answer as a finding. The rates are the findings.

Step 7: Turn the two worst gaps into work

Pick the two most commercially painful results, usually a money topic where you barely appear, or a factual error the engines keep repeating. For each, look at the sources the answers cited: those pages are what the engines currently believe about your category. Your options follow from chapter 4: make your own pages on that topic quotable (evidence, statistics, citations), correct the wrong fact everywhere it appears, or earn a presence on the third-party pages the engines keep pulling from.

Then, and this matters most: schedule the re-run. Same prompts, same engines, same method, four to six weeks later. The audit is the “before”; without the “after”, it was tourism.

What a first audit cannot tell you

Honesty about the limits: fifteen prompts and three runs is a coarse instrument. It will find large gaps and repeated errors reliably. It cannot detect small shifts, or prove that a fix caused an improvement. And there is a limit no amount of extra effort inside the spreadsheet removes: the audit measures the fifteen questions you wrote, and cannot tell you whether those are the questions your market actually asks, or what share of your industry’s real question space they cover. Coverage is measured by modeling the question space itself, at prompt volumes nobody runs by hand. When decisions start riding on finer readings, or on knowing your coverage rather than hoping about it, the work becomes a tool’s. Chapter 8 is about making that call deliberately.