AI Visibility Check: Run This 10-Minute Test Before You Buy Anything

Before you buy an AI-monitoring subscription, a reputation tool, or even our own $29 report, run this test yourself first. It costs nothing and takes about 10 minutes: five real buyer questions, asked cold, in at least two AI engines. Most brand teams skip straight to typing their own company name into ChatGPT and feel good about whatever comes back — but that isn't an AI visibility check, it's a vanity check. It tells you nothing about whether the AI would ever volunteer your name to a stranger. Here's the version that does, plus exactly where it breaks down.

What Is an AI Visibility Check, Exactly?

An AI visibility check tests whether answer engines — ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews — name your brand when someone asks a buying question in your category, without being prompted with your name first. It's the AI-era version of a mystery shopper, except the shopper is a language model and the store is a chat window.

Searching your own brand name only checks whether a model has heard of you. It says nothing about whether the model recommends you unprompted, which is the signal that actually predicts AI-influenced traffic and purchases. We turn mention rate, position, and sentiment into a single score in our AI visibility score methodology; this article is the manual, no-tooling version of that same idea.

Scale is why this is worth 10 minutes of your time: 34% of U.S. adults have now used ChatGPT, roughly double the share who had in 2023, according to Pew Research Center. That's not a niche channel anymore — it's a growing share of the people who will eventually go looking for what you sell.

The Uncomfortable Number: How Often Brands Actually Get Named

Across 90 completed AEOeye audits and 660 buyer-question rows, the audited brand was named in only 23% of AI answers — and that rate swings wildly by engine, from 31% down to 2%. If you've only tested one engine, you've measured almost nothing.

Source: AEOeye internal data, 90 completed audits, 660 buyer-question rows.

Read that table with its limits in mind. Claude's sample is large — 464 question-rows, because every free and paid AEOeye audit queries it — and reasonably stable. The other four engines each rest on just 49 rows, since only full, paid multi-engine reports query them — small enough that a few queries can swing the percentage several points. Treat the size of the cross-engine gap as directional, not exact. Treat the underlying pattern — mention rates are low overall, and vary a lot by engine — as solid, since it holds at the level we can actually measure with confidence.

This isn't a marginal shift, either. Gartner predicted back in 2024 that traditional search engine volume would fall 25% by 2026 as generative AI tools substitute for queries people used to type into Google. Whether or not that exact number lands, the direction is obviously right — more of the "which brand should I buy" moment is moving inside a chat window, and a 23% mention rate inside that window is a real leak, not a rounding error.

The 10-Minute Manual AI Visibility Test (No Sign-Up Required)

You can get a directional read on your AI visibility in 10 minutes, for free, with nothing but two browser tabs: write five buyer-style questions, ask them logged out across at least two engines, and record exactly who gets named and in what order. Here's the exact sequence we'd run.

1. Write five buyer questions — never your brand name. Phrase them the way an actual customer types, like "best [category] for [specific use case]" or "[category] that doesn't [common complaint]." Typing your own brand name tests recall, not recommendation — it's the single most common mistake we see.
2. Ask each question in a fresh, logged-out session, across at least two engines. Use a private/incognito window and confirm you're actually signed out. This isn't paranoia: OpenAI's Memory FAQ documents that ChatGPT keeps track of details from past chats and references that history to personalize future answers, and Perplexity's help center confirms the same for signed-in accounts. A logged-in test measures your history with the tool, not what a stranger sees.
3. Record three things per answer: were you named, at what position, and who was named instead. Position matters almost as much as presence — third on a list of three reads very differently from one mention buried in a paragraph of ten competitors. Write the competitor names down too; that list is often more useful to sales than the fact that you were missing.
4. Repeat the exact same five questions a day or two later. Answer engines sample from a probability distribution rather than looking up a fixed fact, so the same prompt can return a different answer on a different run — that's documented, expected model behavior, not a glitch. A brand that appears on run one and vanishes on run two is borderline, not absent.
5. Sort every miss into one of two buckets — they need opposite fixes. "Not found" means the model has little reliable information to draw on: a content and structured-data problem. "Found, but not recommended" means the model knows you exist and picked someone else anyway: a positioning and evidence problem. Conflating the two is why so many AI-visibility fixes fail.

Photo by Jakub Zerdzicki on Pexels

What Your Results Actually Mean (And What to Fix First)

Your five-question test almost always produces one of four recognizable patterns, and each one calls for a completely different first move — find your symptom in the table below before you brief an agency or rebuild anything.

The fix follows from the bucket in step 5 above, not the other way around — that's the entire reason to separate the two failure modes before you touch your website.

Where the Manual Test Breaks Down

Hand-testing tells you something is wrong in about 10 minutes, but it can't tell you how wrong, how often, or why. It doesn't scale past a handful of questions, isn't reproducible enough to trend over time, and can't distinguish a live, web-grounded answer from one pulled out of a model's static training memory.

That last distinction matters more than it sounds. Google's own documentation describes AI Overviews as generated by retrieving and synthesizing current web pages, not just recalling training data — so a miss there might be a live indexing or content problem you can fix this week. A miss in a model with browsing turned off might mean it was simply never trained on strong signals about you, which is a slower, different fix. Five manual questions can't tell you which situation you're actually in.

The stakes are rising, not flat. SparkToro's research puts US Google zero-click searches at roughly 68% in early 2026 — more of the buying decision is happening inside the answer itself, before anyone reaches a results page you could rank on. This is the specific gap AEOeye's free audit is built to close: it runs a structured set of buyer questions across all five engines automatically, logs the answer text, and repeats runs so you get a mention rate instead of a single anecdote. It won't tell you anything the manual test couldn't reveal in principle, just at a scale no human tab-switching can sustain. For a closer look at what it checks, see our guide to the free AI visibility checker.

Manual Spot-Check vs. Ongoing Monitoring

A once-off manual test is good for a gut check; it's the wrong tool if you need to know whether a fix actually worked or a competitor is pulling ahead over time, because that needs repeatable, logged monitoring rather than a fresh set of tabs every time.

If you're weighing doing this by hand every month against adopting a tool, we compare the realistic options in our roundup of AI brand monitoring tools. And if you want to see exactly how a mention-rate number like the 23% above becomes a single visibility score, our methodology breakdown doesn't hide the scoring logic behind a black box.

The Bottom Line

Run the five-question, two-engine, logged-out test today — it costs nothing, takes 10 minutes, and tells you more about your real AI visibility than typing your own name into ChatGPT ever will.

When you want the automated, five-engine version — score, saved answer text, and a repeat run for stability all included — start a free audit at AEOeye, see how the full process works, or upgrade to the complete report for a one-time $29. No subscription, because we don't think you should pay monthly to find out whether a chatbot likes you.