AI Visibility Audit: How to Measure Your Brand in ChatGPT, Gemini & Perplexity (2026)
Written by Martin J.
You can tell me your exact Google position for your top keyword, your monthly organic sessions, your click-through rate — and still have no idea what ChatGPT says about your brand when a buyer asks who to use. That transcript is being written hundreds of times a day, without your analytics in the room. It's either accurate, or outdated, or your competitor is the one being recommended in your place. The uncomfortable part: your Google dashboards can't tell you which, because rankings and AI citations stopped tracking each other — the overlap between AI Overview citations and organic top-10 results collapsed from 76% to 38%. Most teams find out what AI says about their brand the same way: a customer forwards them a screenshot. What would that screenshot say about you right now?
That question is getting more expensive to ignore every month. Roughly 68% of US Google searches now end without a click, and when an AI Overview is present, that number hits 83%. ChatGPT alone serves about 2.5 billion prompts a day. Meanwhile, across 1,700 businesses studied in early 2026, fewer than 12% showed up in any AI-generated answer about their category — and in the US, it was barely 7%.
Here's the part nobody warns you about: AI systems don't repeat themselves. Run the same question twice and you'll get different brands, in different order, every time. Which means the audits most guides teach — run a prompt once, count the mentions — aren't measuring anything. This guide walks you through an audit that survives that reality: a five-step protocol you can run this week, free, without buying a tool. No fluff, no vendor lock-in. If you've never checked what AI says about your brand, this is where you start — because right now, someone else is writing that answer.
What an AI Visibility Audit Actually Measures
An AI visibility audit is a structured process for measuring how often — and how accurately — AI search tools mention, cite, or recommend your brand when people ask questions in your space. It answers four distinct questions, and it's worth keeping them separate because they have different fixes:
- Mentions — does your brand appear in the answer at all? A mention can be direct (your brand named as a recommendation) or indirect (your product described, your content summarized, without your name attached). Most guides treat "mentioned" as binary. That's a mistake: an indirect mention means AI is using your material — without giving you credit or the click.
- Citations — when you appear, is there a source link back to your content? A mention without a citation means the model is drawing on training data or third-party chatter about you, not on your own pages.
- Accuracy and sentiment — when you're described, is the description current? Positive, negative, or neutral? Outdated pricing or a wrong use case does active damage before a buyer ever reaches your site. And sentiment has to be scored per platform — Google AI Overviews are measurably more likely to criticize brands than ChatGPT.
- Share of voice — relative to your top three to five competitors, are you mentioned more or less often? The gap between you and the category leader is your most actionable number.
If that list sounds like the beginning of a bigger practice, it is — the definition of AI visibility, and how the retrieval mechanics behind it work, is worth a read on its own. This guide is the operational half: how you actually run the audit.
The 5-Step AI Visibility Audit
The protocol has five stages. Step two is the one almost every guide gets wrong, and it's the difference between a snapshot and a measurement.
Step 1 — Build your question set
The quality of your audit depends entirely on the quality of your prompts. Generic queries return generic signals, so your set should mirror how real buyers talk about your category. Build three buckets:
- Category prompts — what your ICP types in research mode, before they're brand-aware. "Best tools for AI-powered marketing." "What should I use for keyword research in 2026." Run 8–10.
- Brand prompts — how AI describes you directly. "What is [your brand]?" "[Your brand] vs [top competitor]." "[Your brand] pricing." Run 5–7.
- Problem-first prompts — your buyer's thinking before they know the category exists. "How do I check if my content shows up in AI search?" "Why is my organic traffic dropping?" Run 5–8.
Twenty to twenty-five prompts total is the right starting point. Fewer and you'll miss category-level patterns; more and you won't run it consistently next month.
Step 2 — Run every prompt three times (the step everyone skips)
Here's the reality most published audits ignore: AI answers are non-deterministic. In a late-2025 volunteer study of 600 people running 2,961 brand prompts, there was less than a 1-in-100 chance that two runs returned the same list of brands — and roughly a 1-in-1,000 chance of the same list in the same order. Independent checks found similar instability: about 91% of repeated AI Mode answers differed run to run, and 40–60% of cited sources change month to month. A single run isn't a measurement; it's a sample of one.
So run each prompt at least three times, in fresh sessions, and log the result of every run:
- Run each prompt 3× per platform, in separate fresh sessions — never in one continuing thread, where the model's earlier answers contaminate the later ones.
- Log the model version and the date for every session. OpenAI shipped a model update in spring 2026 and the average number of unique domains cited per response dropped from 19 to 15 — visibility trend lines break on model releases, and you need the annotation to explain why yours did.
- Test ChatGPT's tiers separately: Free (training-data answers) and Plus with web search (live retrieval) can produce very different results for the same prompt. Track them as two separate rows.
- Run the same set across Google AI Overviews, Google AI Mode, ChatGPT, Gemini, and Perplexity — the platforms your buyers actually use. Skip Claude only if your audience isn't in larger organizations; it's where a meaningful share of enterprise research happens. Add Grok only if your category skews technical — platform fragmentation is real, and the citation gap between the most- and least-citing platforms measured 615×.
What you log per run: brand mentioned (direct / indirect / none), the response position, description accuracy, sentiment, and the cited URL if one appears. After three runs, report the mention rate — "mentioned in 2 of 3 runs" — never a single run's outcome.
Step 3 — Classify and score
With three runs per prompt behind you, score what you saw. Four scores:
- Mention rate — across your full prompt set, what percentage of responses mentioned your brand in at least one run? Below 15% signals a serious gap for an established brand. 15–35% is typical for mid-market brands. Above 50%, your content is well aligned with how AI retrieves and recommends.
- Direct vs indirect split — of your mentions, how many name you directly versus describe your product or summarize your content without the name? A high indirect share means AI is using your material without the credit — often a fixable content-structure problem.
- Citation rate — of responses where you appeared, how many linked back to your content? A low citation rate next to a solid mention rate means AI is referencing you from training data and third-party chatter, not from your live pages. That's a structural gap: your content isn't being treated as a primary source.
- Accuracy and sentiment — for every mention, is the description current (yes / partial / wrong)? And is the sentiment positive, neutral, or negative? BrightEdge found Google AI Overviews are 44% more likely to criticize brands than ChatGPT — so score per platform and don't average across them.
Then run the same set for your top three to five competitors and compare. If the category leader's mention rate is double yours or more, that gap is measurable, and it's the strongest argument you'll get for a focused content and authority sprint.
Step 4 — Map what drives your mentions
For every response that mentioned you, record which source the model drew from: your own site, a review platform, Reddit, YouTube, an analyst report, a news article. This is the step most audits skip, and it's where the fixes hide. The evidence is clear about what you'll find: across 6.8 million AI citations, 86% traced back to brand-managed sources — 44% to your own website, 42% to business listings you control. Owned content is the dominant citation driver; third-party sources are a real but long-tail lever.
That changes your priorities. A prompt where you're mentioned but the model cites a two-year-old forum thread about you isn't a content gap — it's an authority problem, and publishing more blog posts won't fix it. A prompt where the model cites your own page with outdated pricing is a content-freshness problem with a different fix. Map first, then decide.
Step 5 — Track over time
An audit you run once is an anecdote. The value compounds when you repeat the protocol monthly and compare. Monthly is the right cadence for most teams: AI search results change continuously as models update, new content gets indexed, and competitors shift strategy. Monthly snapshots give you trend lines without audit fatigue. If you're mid-content-sprint or in a fast-moving category, bi-weekly checks on your ten highest-priority prompts make sense.
Keep a model-version log in the same sheet. When your mention rate moves, you need to know whether your content changed or the model did. Content updated within 30 days earns a meaningful citation multiplier in AI answers, which is why the audit's final step always points back to publishing — but the log tells you which lever actually moved your number.
What Your Results Mean
The baselines have shifted, and they're brutal. Across 1,700 businesses studied in early 2026, only 11.9% had any AI visibility at all — 88.1% were completely absent from AI discovery, and US businesses fared worse at 7.1%. A separate study of 177 brands across eight AI platforms found 90% had zero AI search mentions. The category leader in your space is probably visible in only a third of relevant answers.
So read your numbers against those baselines, not against perfection:
- Under 15% mention rate — you're invisible, which puts you in the majority. The fastest path forward is content: publish factually dense pages that directly answer your category and problem-first prompts, because first-party content drives the largest share of AI citations.
- 15–35% — typical, with room to grow. Look at your direct-to-indirect split and your citation rate first. Raising citation rate usually means making your existing pages more authoritative and structurally clearer — not publishing more.
- Above 50% — your content is doing its job. The highest-leverage move now is fixing inaccurate descriptions and defending the position, because AI answers are volatile and the model-update annotation in your log will show you exactly when your number can move.
One honest caveat before you over-celebrate a high number: visibility isn't traffic. AI Overview citations perform at roughly the level of a position-six click, and the cited brand still loses the click to the answer itself. But cited brands do earn meaningfully higher organic and paid CTRs than uncited competitors. Audit what you can act on — and you can act on all four scores above.
Manual Audit vs Tools: When DIY Stops Scaling
The manual protocol above works. Run it once and you'll learn more about your AI presence than any dashboard you've seen. But let's be honest about the arithmetic: 20 prompts × 3 runs × 4–5 platforms is roughly 250 logged responses per cycle, and a careful manual audit runs three to four hours a month — before you do anything with the results. Non-determinism makes it worse, because your 3× protocol exists precisely because a single pass can't be trusted.
That's the line where DIY stops scaling. Three categories of tooling exist:
- Purpose-built AI visibility platforms — Allable, Profound, Otterly, Peec AI, and a growing handful of others automate the prompt-running, the run-to-run repetition, and the trend tracking. This is the right category when you need audits at scale, across multiple clients, or on a schedule you'll actually keep.
- Traditional SEO suites with AI visibility add-ons — Ahrefs Brand Radar, Semrush's AI toolkit, and SE Ranking's AI features bolt AI visibility onto existing SEO platforms. Methodology varies widely between them — check which engines they actually track and how they handle response variability before you commit.
- Your spreadsheet — viable for a small team running monthly audits on a tight prompt set. No cost, full control, and it will teach you the territory. Just budget the three hours, and accept that trend tracking is on you.
For a full comparison of what each tool measures — and where their methodologies differ in ways that matter — our best AI visibility tools guide covers the major options in depth. If you want the definitional background first, start with what LLM visibility is and the toolkit comparison.
How to Act on the Audit
Knowing your gap is only half the work — the audit pays off when it changes what you publish. The fix categories, in order of impact:
Content fixes. Publish pages that directly answer the prompts where you're missing — not thin "answer the question" pages, but substantive, factually dense content AI can pull from. This is the highest-leverage category because first-party content drives 44% of AI citations, and it's the one you fully control.
Authority fixes. When the model cites third-party sources about you — old reviews, outdated forum threads — your own content can't fix that. Getting listed and reviewed on platforms AI treats as authoritative, and earning coverage from sources it cites, moves the needle faster than ten more blog posts. Our guide to AI search monitoring covers how to watch these signals without burning your month on it, and the brand gap analysis walkthrough shows how to turn the visibility gap into a content plan.
Accuracy fixes. Outdated or wrong descriptions of your brand are a different problem from absence. Fix your own positioning pages first — models lean heavily on what your site says about you — then work the third-party sources that describe you incorrectly.
Then automate the loop. The manual audit gets you the first honest number. Keeping it honest every month is a system problem, and that's where Allable comes in. You define your question set once; Allable runs it across Google AI Overviews, AI Mode, ChatGPT, Gemini, and Perplexity, applies the run-to-run repetition so the number means something, and returns structured results — mention rate, direct versus indirect, citation rate, per-platform sentiment, and the actual AI-generated text describing your brand. It tracks changes between runs, and it annotates model-update dates so a trend break doesn't masquerade as a content win or loss.
When the system finds a gap — a prompt category where competitors are cited and you're not — it surfaces that as a content opportunity with keyword data attached. You go from "we should probably audit our AI visibility" to "here are six topics to brief this month, ranked by opportunity." Allable includes AI-visibility monitoring on every paid plan, starting at €83/month billed annually (€99 month-to-month) — a plan that also covers keyword research, content, and publishing, so the audit connects directly to the fixes it recommends. (Pricing verified September 2026.)
Frequently Asked Questions
- What is an AI visibility audit?
- An AI visibility audit measures how often and how accurately AI search tools — ChatGPT, Gemini, Perplexity, Google AI Overviews and AI Mode — mention, cite, or recommend your brand when users ask questions in your category. It tracks four things: mentions (direct and indirect), citations, description accuracy and sentiment, and share of voice against competitors. Unlike a Google SEO audit, it measures answers, not rankings.
- How do I check if ChatGPT mentions my brand?
- Build a set of 20–25 prompts in three buckets — category, brand, and problem-first — and run each one in a fresh ChatGPT session at least three times. Log whether your brand is mentioned directly, indirectly, or not at all, whether a source link appears, and how you're described. Report the mention rate across runs, not a single run's outcome, because ChatGPT answers vary run to run — two runs of the same prompt have less than a 1-in-100 chance of returning the same brand list.
- How long does an AI visibility audit take?
- A full manual audit — 20–25 prompts, run three times across four to five platforms — takes three to four hours the first time, including building your prompt set and spreadsheet. Subsequent monthly runs are faster, roughly two to three hours, once your setup is stable. Tools that automate the prompt-running cut that to a data review of about 20 minutes.
- What tools can automate an AI visibility audit?
- Three categories: purpose-built AI visibility platforms like Allable, Profound, Otterly, and Peec AI, which automate prompt-running and trend tracking; AI add-ons to traditional SEO suites like Ahrefs Brand Radar, Semrush, and SE Ranking; and your own spreadsheet for small, manual audits. Our best AI visibility tools guide compares the major options and their methodologies in detail.
- How often should I audit my brand's AI visibility?
- Monthly is the right default cadence. AI answers change continuously — models update, new content gets indexed, competitors shift strategy — and monthly snapshots give you trend lines without audit fatigue. If you're mid-way through a content sprint or in a fast-moving category, bi-weekly checks on your highest-priority prompts are worth the extra time. Always log the model version with each run, because visibility numbers move when models update, not just when your content does.
The Bottom Line
AI search doesn't send you a Search Console report. It describes your brand hundreds of times a day, and until you audit it, you're betting your content budget on a number you can't see. The protocol here — build your question set, run it three times, classify and score, map the sources, track monthly — takes a few hours and costs nothing. Run it once and you'll know whether you're one of the 88% of brands invisible to AI discovery, or one of the few being recommended. Run it every month and you'll know exactly what moved your number — your content, or the model. The audit is a start, not a system. But it's the only honest start you've got.
Martin Janeček is the CEO of Allable.ai, an AI-powered marketing platform that replaces Semrush, Jasper, and Surfer SEO with a single chat-first interface.
Run the audit once by hand — then let Allable run it every week.
See every AI answer that mentions your brand, and the ones that don't.