How to Check If AI Mentions Your Brand (ChatGPT, Perplexity, Google AI)

These get conflated constantly, and they're not the same thing.

A mention is when the answer names your brand in the text — "For project management, tools like Asana, Linear, and YourBrand work well." A citation is when the engine links to your domain as a source, usually shown in a footnote, a source panel, or an inline link.

They come from different places. Mentions are pulled from the model's training data plus whatever it retrieves live — so a mention reflects your overall reputation across the web. Citations come almost entirely from real-time retrieval, so they reflect whether a specific page of yours ranks for the query the engine ran behind the scenes.

You want both, but they're fixed differently. Weak on mentions? You have a presence problem — not enough authoritative third-party sources describe you in your category. Weak on citations? You have a content problem — your pages aren't structured to be the source an engine grabs. When you check visibility, log them as separate columns. A brand can be mentioned constantly and never cited, or cited often yet rarely recommended by name.

You don't need a tool to get started. You need the right prompts and a logging habit.

Write 15 to 30 questions a real buyer would ask — never including your brand name. Cover the spread:

• Category: "best [product type] for [audience]"
• Comparison: "[competitor] alternatives" / "[competitor] vs what else"
• Use case: "how do I [job your product does]"
• Buyer-stage: "what should I look for when choosing a [product type]"

Run each one through ChatGPT, Perplexity, and Google (toggle AI Mode / watch for the AI Overview). For each, record four things: were you mentioned, were you cited, what position (first name listed vs. buried), and the sentiment (recommended, neutral, or mischaracterized). A plain spreadsheet beats memory here.

Two non-obvious moves. First, run ChatGPT prompts in both browsing and non-browsing mode — the answers diverge sharply, and most of your buyers are on the default. Second, ask each prompt two or three times in fresh chats. You'll watch yourself appear and vanish across identical questions. That flicker is the whole reason manual checking has a ceiling.

AI answers are non-deterministic. Same prompt, different session, different brands named. The model samples from a probability distribution every time, so a brand sitting at the edge of its "top tools" cluster shows up maybe 40% of the time. Ask once and you'll conclude you're in — or out — when the truth is a percentage.

That percentage is the only metric that matters, and you can't get it from one look. Mention rate — how often you appear across many runs of the same prompt — is the real KPI. A brand mentioned in 8 of 10 runs is in genuinely different shape from one mentioned in 2 of 10, even though a casual single check might catch either as "yes, we're there."

The answers also drift between engines and over time. Perplexity leans hard on live retrieval and citations; ChatGPT leans more on trained knowledge; Google AI Overviews track closely with what ranks. A model update or a competitor's new comparison page can move you five points in a week. Which is why visibility isn't a thing you check once — it's a thing you monitor.

Once you're past a handful of prompts across five engines, runs three deep, doing it by hand falls apart fast. That's the job a dedicated AI visibility checker exists for.

A good one does four things you can't practically do manually:

• Runs hundreds of prompts across ChatGPT, Perplexity, Google AI Overviews, Gemini and Claude on a schedule, not in your browser.
• Repeats each prompt enough times to compute a real mention rate instead of a coin-flip snapshot.
• Pulls competitors automatically, so you see share of voice — you're named in 30% of category answers, your top rival in 55%.
• Tracks the trend, alerting you when a number moves so you can tie it to a content change or a model update.

This is exactly what AEOeye is built to do — its free audit runs your real category prompts across every major engine and hands back your mention rate, citation rate, and where competitors are eating your share, in one pass. The point isn't the dashboard. It's converting a fuzzy "are we in AI results?" into a number you can move and defend in a meeting.

A check is only useful if it becomes a baseline. Once you have your numbers — mention rate per engine, citation rate, share of voice, sentiment — freeze them as your starting line, dated.

Then read the gaps. Three patterns and what each tells you:

• Mentioned but never cited → competitors own the source pages for your category. Fix: build the comparison and "best X" pages engines actually pull from.
• Cited but rarely recommended → engines find you but don't trust you enough to lead with you. Fix: earn more third-party mentions (reviews, listicles, expert roundups) that reinforce your category authority.
• Mischaracterized → the model has stale or wrong facts about you. Fix: tighten your own site's plain-language descriptions and get accurate third-party coverage; models correct slowly but they do correct.

Re-run the exact same prompt set monthly. Same questions, same engines, same number of repetitions — otherwise you can't compare. The trend line is the deliverable. One check tells you where you stand today; the same check repeated tells you whether anything you did actually worked.