At some point in the last quarter, someone in a meeting asked what ChatGPT says about your brand. And nobody in the room actually knew.
You have a rank tracker for Google and a brand tracker for awareness. But when a buyer asks an AI assistant "what's the best option for someone like me?" — where shortlists now get written — you have no dashboard, no report, no idea.
That's the gap an AI visibility audit closes. Not a tool purchase, not a six-week project — ten minutes with a spreadsheet, and by the end of this guide you'll have run one.
Why should I care, though? Fair question. Because if you're responsible for pipeline, you want to be present wherever buying decisions are forming, and they're increasingly forming inside AI answers. Adobe measured traffic to US retail sites from generative-AI sources growing 4,700% year over year in July 2025. The traffic isn't just growing, it converts: across 500+ high-value topics Semrush studied, the average AI search visitor is 4.4 times as valuable as the average visitor from traditional organic search, based on conversion rate. And your competitors have noticed — in Conductor's survey of 250+ enterprise digital leaders, 94% plan to increase AEO/GEO investment in 2026.
Before any of that spend makes sense, though, there's a first principle most teams skip straight past:
You can't fix how AI describes your brand until you've read, word for word, what it actually says.
That sounds obvious. It's also the step almost everyone jumps over — straight from "we should do GEO" to publishing "AI-optimized" content, with no idea what the engines currently say or which sources they trust. That's a strategy with no baseline. The audit is the baseline: read the answers, turn them into a score, and let the gaps set the fix list.
What an AI visibility audit measures
An AI visibility audit is a structured check of how answer engines — ChatGPT, Gemini, Perplexity, Google's AI Overviews and Copilot — present your brand when buyers ask buying questions. It measures four things:
- Presence — are you mentioned at all when your category comes up?
- Position — how early in the answer, and before or after which competitors?
- Sentiment and accuracy — how are you described, and is any of it wrong?
- Citations — which sources did the engine lean on, and are you one of them?
It's a different animal from an SEO audit, though the two get bundled together constantly:
| Dimension | SEO audit | AI visibility audit |
|---|---|---|
| Core question | Where do our pages rank? | What do the answers say about us? |
| Unit of measurement | Keyword positions, clicks | Mentions, position in answer, sentiment, citations |
| What it inspects | Your website | The engines' answers — plus the sources behind them |
| Output | A list of page and site fixes | A visibility score and a list of answer fixes |
You'll hear this discipline called AEO (answer engine optimization) or GEO (generative engine optimization) — same idea, different acronym. The audit is where either one starts, so open a blank spreadsheet and let's go.
Step 1: Write your 10 buyer questions
The audit is only as good as the prompts, and the most common mistake is auditing vanity queries. "Tell me about [your brand]" proves nothing — what matters is whether you show up when the buyer doesn't name you.
A good audit prompt has buyer intent baked in — phrased the way someone mid-purchase actually talks. Write about ten, mixing unbranded category questions (the ones that build pipeline) with a couple of branded ones (the ones that catch misinformation).
Here's a fill-in-the-blank pack — copy it, replace the brackets, done:
| # | Prompt pattern | What it tests |
|---|---|---|
| 1 | best [category] for [use case] | Your core unbranded shortlist moment |
| 2 | best [category] for [audience or company size] | Whether you surface for your actual ICP |
| 3 | top [category] brands in 2026 | Category roll-call — who makes the default list |
| 4 | [your brand] vs [main competitor] | How head-to-heads get framed, and who wins |
| 5 | [your brand] alternatives | Who the engines steer defectors toward |
| 6 | is [your brand] good for [use case]? | Accuracy and sentiment on a branded question |
| 7 | what should I look for when choosing a [category]? | Whether the buying criteria favor you — or a rival's spec sheet |
| 8 | [category] for [industry or vertical] | Visibility in the niches you claim to serve |
| 9 | is [your brand] worth the price? | How objections get answered when you're not in the room |
| 10 | [category] with [must-have feature or requirement] | Whether your differentiator is even attached to your name |
Ten is enough for a real signal. Save the exact wording — the same list, re-run monthly, is what makes month two comparable to month one.
Step 2: Run the prompts across the engines
Now put each prompt to four places: ChatGPT, Gemini, Perplexity, and a plain Google search so you capture the AI Overview (the AI answer box above the results) when one appears. That covers the answer engines most buyers touch; add Copilot if your buyers live in Microsoft-land.
Three bits of mechanics keep the data honest:
- Fresh chat per prompt. Earlier messages steer later answers — a new conversation each time keeps every answer clean.
- Watch for personalization. If you're logged in, memory and chat history can flatter you — the assistant may know where you work. Use a logged-out session where you can, or note that your results are personalized.
- Run each prompt twice. Here's the part nobody tells you: AI answers vary run to run. The same prompt can name you first at 9am and skip you at 2pm — you're not auditing a billboard, you're auditing a very confident improviser. Two runs show which mentions are stable; if you only have time for one pass, treat the result as a snapshot, not a verdict.
Paste or screenshot every answer as you go. You want the raw wording on file — Step 3 scores it, and future-you will want the receipts.
Step 3: Score presence, position and sentiment
Reading forty answers gives you vibes. Scoring them gives you a number you can defend in a meeting and compare next month. For each answer, score your brand:
| Signal | Points | Why it counts |
|---|---|---|
| Mentioned anywhere in the answer | +1 | Baseline presence — you exist in the engine's mental model of the category |
| Named before any competitor | +1 | First mention is the AI answer's version of ranking #1 |
| Actively recommended, not just listed | +1 | "Consider X for this" converts; appearing tenth in a list of twelve doesn't |
| Your site cited as a source | +1 | The engine trusts your content enough to link it — mentions are talked about, citations are listened to |
| Factual error or negative caveat | −1 | "Good but overpriced" and wrong pricing both cost you deals — flag the exact wording for Step 6 |
Set up the spreadsheet with one row per prompt, one column per engine, and the rubric total in each cell. With 10 prompts across 4 engines the maximum is 160; your total as a percentage of that is your AI visibility score. Here's what filled-in rows look like, using well-known running-shoe brands:
| Prompt | ChatGPT | Gemini | Perplexity | AI Overview | Row total |
|---|---|---|---|---|---|
| best running shoes for marathon training | 2 — mentioned and recommended, after Nike and Adidas | 1 — mentioned once, mid-list | 3 — recommended, on.com cited | 0 — Nike, Adidas, Hoka only | 6 / 16 |
| Hoka vs On for everyday training | 3 — named first, recommended | 2 — mentioned and recommended, no clear winner | 4 — first, recommended, cited | 2 — mentioned, on.com cited | 11 / 16 |
| best running shoes for flat feet | 0 — absent | 1 — one passing mention | 0 — absent | 0 — absent | 1 / 16 |
Even three made-up rows show how the audit turns into strategy: this illustrative "On" is strong when the buyer already knows its name, buried on the big generic prompt, and invisible on a use-case prompt it should own. That last row is the finding — not "our AI visibility is 38%," but "we don't exist for flat-footed runners, and here are four answers proving it."
See what AI says about your brand
Run the free audit and see what AI assistants tell buyers about businesses like yours.
Step 4: Check who gets recommended instead
Your zeros are only half the story. Every answer that skipped you recommended somebody — so go back through the saved answers and log which competitors appear, how often, and the exact language used to describe them.
Add a "competitors named" column and tally. A pattern emerges fast: one or two rivals appear in most answers while everyone else, you included, rotates through the leftover slots. That high-frequency brand is the category's default recommendation, and its share of your prompt set is the benchmark your score is really competing against.
The descriptions matter as much as the counts. If a rival is consistently "the premium option for larger teams" while you're "a cheaper alternative," the engines have assigned you a positioning you never signed off on. Write those phrases down verbatim — they're either a problem to fix or free market research.
Step 5: Check which sources the AIs cited
Now find out where the answers came from. Perplexity shows numbered citations on every answer; AI Overviews show link cards; ChatGPT and Gemini link sources when they browse. Click through and log every domain in a "sources" tab of your sheet.
After ten prompts you'll usually find the answers lean on a surprisingly short list: a couple of industry review sites, a comparison post or two, Reddit threads, maybe a news piece — and your own site, present or conspicuously absent. This is the most actionable artifact the audit produces, because engines don't invent opinions; they synthesize them from sources they trust. If the three sources they lean on don't mention you, nothing on your own site changes the answer. The fixes live in that list — being reviewed where the engines read, getting into the comparison posts you're missing from. We've dug into which domains each engine leans on in our guide to AI citation sources.
Step 6: Read your score and set fix priorities
Time to read the result. As a rough guide:
- Low (under ~20%) — the engines barely know you exist on unbranded prompts. Your problem is presence; start with the sources from Step 5.
- Mixed (~20–45%) — you appear, but late, unevenly across engines, or with caveats. The most common result and the most fixable: specific prompts and engines to win.
- Strong (over ~45%) — you're a default recommendation for much of your prompt set. The job flips to defense: accuracy, sentiment, and watching for a rival taking your slots.
The fix list, in priority order
- Correct factual errors first. Wrong pricing or a dead product claim costs deals today. Trace each error to its stale source — often your own outdated page — and fix it there.
- Close the source gap. Get present in the third-party sites your Step 5 list showed the engines trusting: review platforms, comparison articles, community threads. This moves answers more than anything on your own domain.
- Fix the technical layer. Check robots.txt isn't blocking AI crawlers (GPTBot, PerplexityBot, Google-Extended — the user-agents the engines crawl with), add schema markup (structured labels that tell machines what your pages say), structure key pages with clear FAQ-style questions and answers, and keep them fresh — engines skip stale pages in fast-moving categories.
This fix work is the discipline of generative engine optimization — the audit is how you aim it. One audit is a photograph, not a film: re-run the same prompt list monthly and track the trend. The number to watch is your share of voice in AI answers — how often you appear versus competitors — the metric a CMO will actually track.
Why AI visibility audits get confusing
The confusion comes from one assumption: that AI visibility is downstream of Google rankings, so a good SEO audit covers it. It doesn't — the edge cases prove it:
- Ranking #1, invisible in answers. A brand can top Google for its money keywords while ChatGPT never mentions it — engines synthesize from sources they trust, not from rank order, and a review site's opinion can outweigh your position-one page.
- Cited from deep in the results. AI Overviews and assistants often cite pages that rank nowhere near the top — structured, quotable content can out-perform higher-ranking pages in AI answers.
- A clean SEO audit, a losing answer. Your technical SEO can be flawless while Perplexity describes you with three-year-old pricing pulled from a forgotten comparison post. Nothing in a crawl report catches that.
When you're unsure which audit a problem belongs to, apply one test: would fixing it change what an AI answer says, or where a blue link ranks? "We're missing meta descriptions" changes rankings — SEO audit. "Gemini recommends our competitor for our flagship use case" changes answers — AI visibility audit. Plenty of fixes (schema, structure, freshness) help both — which is why the audits get conflated — but the measurements never overlap, so run both and report them separately.
When to graduate from the spreadsheet
Everything above works with zero software, and for a first baseline the spreadsheet is the right tool. The pain arrives with repetition: month three, four engines, ten prompts, two runs each, plus competitor tallies and citation logs — per brand, if you're an agency. The ten-minute audit becomes an afternoon of copy-paste, and that's usually where the monthly cadence quietly stops.
That repetition is exactly what our free evaluation automates. It generates buyer questions for your category, puts them to AI assistants live, checks your site's GEO readiness — schema, FAQ structure, AI-crawler access, freshness — alongside SEO basics, and combines it all into one score out of 100. It's free, you don't need to sign up to see the score, and you can run up to three audits a day. It won't replace the reading — nothing does — but it takes the clerical work off the monthly re-run and gives you a fast second opinion on the baseline you just built by hand.
AI visibility audit FAQs
How do I check my brand's AI visibility?
Ask ChatGPT, Gemini and Perplexity the questions your buyers ask — around ten buyer-intent prompts like "best [category] for [use case]" — plus a Google search to capture the AI Overview. For each answer, record whether your brand is mentioned, how early it appears, how it is described, and whether your site is cited as a source. Score each answer on a simple rubric and total it. The whole check takes about ten minutes with a spreadsheet.
Is there a free AI visibility checker?
Yes. Several vendors offer free AI visibility checkers, and Apex's free evaluation is one of them: it generates buyer questions for your category, puts them to AI assistants live, checks your site's GEO readiness (schema, FAQ structure, AI-crawler access, freshness) alongside SEO basics, and combines everything into one score out of 100. It's free, you don't need to sign up to see the score, and you can run up to three audits a day. The manual spreadsheet method in this guide is also completely free.
How often should I run an AI visibility audit?
Monthly is the right cadence for most brands. AI answers vary from run to run, so a single audit is a snapshot — the trend across repeated audits is what tells you whether your visibility is improving. Re-run the same prompt list each time so the results are comparable, and add an extra run after any major content push, PR moment or product launch.
What to do in the next 30 minutes
- Run Steps 1–3 today, before the calendar eats the intention. A rough baseline this week beats a perfect one next quarter.
- Open yoursite.com/robots.txt and search for GPTBot, PerplexityBot and Google-Extended. An accidental block erases you from AI answers no matter what else you fix — a 60-second check.
- Put the score next to your organic rankings in this month's report. The gap between the two numbers is the argument that gets AI visibility budgeted.
- Read the share of voice guide next. It turns your one-off audit number into the trend metric you'll actually manage against.