20+ Prompts to Test if AI Assistants Mention Your Brand
You can't fix what you haven't measured. Below are 22 prompts to paste into ChatGPT, Claude, Gemini, Perplexity, and Copilot to see whether your brand shows up. Run each one twice (once with browsing on, once with it off), in a fresh incognito session with memory disabled, and swap in your real category and competitor names. Use the AI Mention Checker if you'd rather skip the copy-paste and get the same answers across all engines in 30 seconds.
You can't optimize what you don't measure.
Most founders have a vague sense that "AI is starting to matter for buyers" and have never actually run the prompts a real buyer would type. So they don't know if ChatGPT names them, if Claude knows they exist, or if Gemini hallucinates a wrong tagline. They guess, get nervous, and either do nothing or chase the wrong fix.
This post is the fix for that. Below are the exact prompts to paste into every major AI assistant. Copy them, swap in your category and competitor names, and you'll have a real picture of where you stand inside an hour.
How to use these prompts
A few rules before you start, because doing this wrong gives you garbage data.
- Run each prompt in every LLM you care about. ChatGPT, Claude, Gemini, Perplexity, Copilot. Most brands are visible in one or two and invisible in the rest, and you only see that pattern when you check side by side.
- Run each prompt at least twice. LLMs sample probabilistically, so the same prompt produces different answers in different sessions. One run is anecdote, two or three is signal.
- Run once with browsing on and once with browsing off. The two modes use different sources and can produce very different answers. Both reach real users.
- Use a fresh incognito session with memory disabled. If you've talked about your brand before and memory is on, the model biases its answers for you specifically. That's not a measurement, that's a mirror.
- Swap in your real category and competitor names. The prompts below use placeholders in brackets. Replace "[category]" with the term your buyer would actually use (not your internal positioning), and "[competitor]" with the brand they'd compare you to.
Category recommendation prompts
The most important set. These are the prompts buyers actually type when they're shopping. If you're not named here, none of the other categories matter much.
"What are the best [category] tools in 2026?"
The canonical buyer query. A healthy result names you in the top 5 and describes you accurately. If you're missing from a generic best-of list, the model has no shortlist text to pull from for your category.
"Recommend the top 3 [category] platforms for a small team."
Top-3 queries are tighter than top-10 lists. Being named in a top-3 answer is roughly 3x as commercially relevant. If you're in the top-10 but not the top-3, you have positioning, not dominance.
"What [category] tool do most [target user] use?"
Tests whether the model associates you with your buyer segment. A common failure mode: you're named in the generic answer, but not when the buyer narrows by their own role.
"Give me a shortlist of [category] tools that are good for [specific industry]."
Industry-pivot query. Reveals whether the model knows you serve a specific vertical or sees you as horizontal. Vertical positioning fails fast here.
"What's a [category] tool that's been getting attention lately?"
Tests recency and momentum. With browsing on, this is where new entrants and recently-launched brands tend to show up. With browsing off, it leans toward whatever the training data marked as "buzzy."
Alternative-to and competitor-comparison prompts
Higher-intent than generic category queries. A buyer running these has already heard of a competitor and is actively shopping for options.
"What's the best alternative to [main competitor]?"
The single highest-intent prompt for most categories. Being named in the alternative-to slot for the category leader is worth more than 10 generic best-of mentions.
"Cheaper alternatives to [main competitor]?"
Pricing-pivot variation. Surfaces brands the model associates with affordability. Be careful: if you're not actually cheaper, getting named here can pull in the wrong leads.
"Open-source alternatives to [main competitor]?"
Only relevant if open source is part of your category, but in those categories this is a top-3 buyer query. Strong open-source positioning shows up clearly.
"Compare [your brand] and [competitor]."
Forces the model to talk about you head-to-head. Watch the description, not just the verdict. A "they both do X" answer means the model can't tell you apart, which is its own problem.
"Is [your brand] or [competitor] better for [specific use case]?"
Use-case-conditioned comparison. The best test of whether the model has nuanced understanding of your differentiation. Most brands fail this one even when they pass the generic comparison.
Use case and problem prompts
How buyers actually phrase queries when they don't yet know the category. The model has to reason from problem to product, which is where positioning either pays off or breaks.
"I'm a [role] and I need to [job-to-be-done]. What tool should I use?"
The classic job-to-be-done query. Tests whether the model can route a buyer from their actual workflow to your product.
"What's the easiest way to [outcome the buyer wants]?"
Outcome-first query. The model picks the tool it thinks is easiest, which usually means most heavily covered in beginner-friendly content. Being named here correlates with strong tutorial-level footprint.
"I have [specific problem]. What software solves this?"
Problem-first query. Often surfaces a different shortlist than the category query, because the model maps problems to tools differently than tools to categories.
"My team is struggling with [pain point]. What's a good fix?"
Team-context variation. Tests whether you're associated with team-level problems versus individual ones. B2B SaaS brands often fail this if their content over-indexes on solo users.
"What do [target persona] use to handle [specific task]?"
Social-proof framing. The model answers based on what it has seen attributed to that persona in its training data. Reveals whether your social proof has actually made it into the model's worldview.
Brand awareness prompts
The basics. Does the model know you exist at all, and can it describe you without making things up?
"What is [your brand]?"
The single most important brand awareness prompt. A clean, accurate one-paragraph description is the baseline. "I don't have information on that" means you're invisible. A hallucinated description is worse than invisible.
"Who founded [your brand]?"
Tests whether the model has correct biographical facts. Common failure mode: it names the wrong founder, conflates you with another company, or confidently invents a name.
"What does [your brand] do?"
Functional description test. The answer here is the elevator pitch the model will give to every buyer who asks. If it's vague or wrong, every downstream recommendation is too.
"When did [your brand] launch?"
Tests training recency and factual grounding. The model's answer reveals whether you exist in the base training data, and if so, whether the dates line up with reality.
Trust and risk prompts
How buyers vet you before they buy. These prompts surface the negative framing the model has absorbed, which is often more revealing than the positive one.
"Is [your brand] legit?"
Direct trust query. A confident "yes, here's why" with real citations is the gold standard. Hedging, "I can't verify that," or warnings about being unable to confirm legitimacy all indicate a thin or low-trust footprint.
"What are the risks or downsides of using [your brand]?"
Forces the model to summarize criticism. The answer is a mirror of what's been written about you negatively, which is useful even when it's painful. If the model invents downsides, you have a hallucination problem.
"What do reviewers say about [your brand]?"
Review-aggregation test. Models lean heavily on G2, Reddit, Trustpilot, and review-style listicles for this. The answer tells you which review surfaces the model has indexed and which ones it considers credible.
Skip the copy-paste
The free AI Mention Checker runs the canonical buyer-intent prompts across ChatGPT, Claude, Gemini, and Perplexity in one click, and shows the exact sources each engine pulled from. 30 seconds instead of an hour of tab-switching.
Run the AI Mention CheckerYou run the same prompt three times and get three different shortlists. What does that mean?
Right. LLMs sample from a probability distribution, so list-style answers vary run to run. Run each prompt two or three times and look at which brands show up most often, not which appeared in one run.
This is normal nondeterminism, not a bug. Treat the pattern across multiple runs as the real signal. A brand that appears in 3 of 3 runs is a confident answer. A brand that appears in 1 of 3 is borderline.
How to read the results
Four categories of outcome. Each one points to a different fix.
| Result type | What it looks like | What it means |
|---|---|---|
| Named correctly | You appear in the shortlist, the description matches your real product, the cited sources are ones you'd want to be cited from. | You're inside the recommendation flywheel. Keep the work going so it compounds across model updates. |
| Named wrong | You appear, but with a competitor's feature description, the wrong category, an incorrect founder, or a tagline that isn't yours. | Urgent. The model is teaching buyers something false about you before they ever visit your site. The fix is overwriting the bad source with a clean accurate one. |
| Not named | You're missing from category answers entirely, even with browsing on, and the model can't describe you when asked directly. | Footprint problem. You need more mentions in sources the model already trusts, not better phrasing on your own site. |
| Hallucinated | The model confidently invents a description, a founder, a feature, or a launch year that doesn't match reality. | You have just enough presence to register but not enough clean canonical content for the model to quote. Fix: publish (or earn) one or two authoritative descriptions the model can latch onto. |
What to do after each result type
Match the action to the gap. Doing the wrong work on the wrong problem is the most common mistake.
- Named correctly. Keep doing what's working. Track quarterly so you spot regressions early. Pour fuel on the source types that are getting cited most.
- Named wrong. Place an accurate description of your product on a high-trust source the model already pulls from. One clean Wikipedia paragraph or one accurate trade-press write-up tends to overwrite a year of misattribution.
- Not named. Build editorial mentions in the sources the model trusts for your category. The classic link building motion, with the bar raised. Agentic outreach is the execution layer if you want volume without a full-time PR hire.
- Hallucinated. Publish or earn a clear factual "about" surface the model can quote. A solid Wikipedia citation, a clean trade-press profile, or a well-structured About page with FAQPage schema all work. Repetition across high-trust sources locks in the correct version.
Now that you know the gap, close it
MentionAgent finds the niche blogs and trade publications AI engines trust for your category, writes the pitch, and follows up until you get the mention. $99/mo flat.
Start Getting Mentioned For $99/moFrequently asked questions
Why do answers change each time I run the same prompt?
Model nondeterminism. LLMs sample from a probability distribution at each token, so list-style questions where several brands are plausible answers tend to vary the most. Run each prompt two or three times in a fresh session and treat the pattern as the signal, not any single answer.
Does my account history affect what the model says about my brand?
For your own sessions, yes. Memory features in ChatGPT, Claude, and Gemini store facts from prior chats and bias future responses for you specifically. They don't change what other users see. Always test in an incognito window with memory disabled, otherwise you're measuring your own conversation history, not real brand visibility.
Can I automate running these prompts across multiple LLMs?
Yes. The free AI Mention Checker runs the canonical buyer-intent prompts against ChatGPT, Claude, Gemini, and Perplexity in parallel and shows which engines named you, plus the sources each one cited. Doing the same manually is four browser tabs, a fresh session per prompt, and a spreadsheet for the comparison.
Which LLM matters most for my buyer?
Depends on your buyer. Developers and technical founders skew Claude and ChatGPT. Mainstream B2B buyers skew ChatGPT and Gemini. Research-heavy buyers skew Perplexity. Microsoft-shop buyers skew Copilot. Test across all of them, then prioritize the engine your top-of-funnel is most likely to use.
Why does ChatGPT with browsing differ from ChatGPT without browsing?
They use different sources. With browsing off, ChatGPT pulls only from its training data, which has a cutoff months in the past. With browsing on, it does a live Bing search and can pull fresh sources, including brands launched after the training cutoff. A brand can be invisible in one mode and visible in the other, so check both.
Can I A/B test prompts to see which surfaces my brand?
You can, but it's misleading. The point of testing prompts is to learn whether real buyer queries surface you, not to find the one phrasing that does. If your brand only appears when you swap in a hyper-specific feature or your own product name, you don't have visibility, you have a confirmation bias check. Run the natural buyer phrasings even when they hurt.
How many prompts do I really need to run?
Ten to fifteen covers most categories. Mix three or four broad category prompts, three or four alternative-to and competitor prompts, three or four use case prompts, and two or three brand awareness prompts. Adding more rarely changes the verdict, the pattern stabilizes fast.
What if I sell to a non-English audience?
Translate the prompts and run them in your buyer's language. LLMs return different shortlists per language because the training corpus and browsing surfaces differ. A brand can be visible in English and invisible in German, Spanish, or Polish. Test the languages your buyers actually use.