How to check what ChatGPT, Gemini, Claude, Grok and Perplexity say about your brand

Updated

Last checked: Oct 31, 2026. The facts about each assistant below come from the vendors' own help and documentation pages on Oct 31, 2026 (sources at the end). The results come from our own daily tests, and we say exactly how we ran them.

Availability (Oct 31, 2026): MentionSweep's free score, $29 Full Report, $39/month Monitor and Agency plan are all live, with checkout through Stripe.

The short answer

To find out what an AI assistant says about your brand, ask it two kinds of questions, on more than one day, and write down what you get:

  1. "What is [your brand]?" This tells you whether the assistant knows you, and whether what it "knows" is true.
  2. Buyer questions without your name in them, such as "what's the best [what you sell] for [who you sell to]?". This tells you whether you're recommended, and who is named instead.

Then check one thing most people skip: did the answer come from a web search or from the model's memory? The same assistant can say very different things in each mode. In our own tests over 12 days, AI assistants answering from memory never once described our product correctly and made up a different product 24 times. Every correct answer came from an assistant that had searched the web and found a page about us.

Step by step

1. Write your questions before you start

  • Your brand question: "What is [brand]?" If your name is also a common word, add a hint: "What is [brand], the [category] tool?"
  • 5 to 10 buyer questions, written the way a customer would type them: "What are the best alternatives to [competitor]?", "Which [category] tool is best for a small team?", "Is there a free [category] tool?". Leave your brand name out, or you only measure whether the assistant can repeat it.
  • Keep the list fixed. If you change the questions every time, you can't compare one day with the next.

2. Ask each assistant, and note which mode it used

  • ChatGPT: OpenAI's help center says ChatGPT "may search the web automatically when your question would benefit from current information", and you can also start a search yourself from the tools menu. Answers that used search may include citations. If an answer has no sources, treat it as coming from what the model already knows. Your own account matters too: the same page says ChatGPT may use saved memories when it rewrites a search and an approximate location from your IP address, so your answer may differ from your customer's.
  • Gemini: Google documents "Grounding with Google Search", which "connects the Gemini model to real-time web content" and adds citations. In the API it's a tool the developer has to switch on, so an API answer without it comes from the model alone.
  • Claude: Anthropic documents a web search tool that "gives Claude direct access to real-time web content" with citations. In the Messages API it's a tool the developer adds to the request.
  • Grok: xAI documents a Web Search tool ("search the web in real-time and browse web pages") and an X Search tool that searches posts and users on X. Both are tools added to an API request.
  • Perplexity: Perplexity is built around web answers with citations. In its Agent API, the documentation says "Web search is opt-in": a request with only a model and a question "answers from the model's own knowledge and returns no citations".

In the apps you usually can't tell for sure which mode was used, other than by the sources shown. That's fine: write down whether the answer had sources and which ones.

3. Record four things per answer

What to record Why it matters
Were you named? (yes / no / position in the list) The basic visibility signal
Was what it said about you true? A wrong description is worse than none: buyers read it as fact
Who was named instead? Your real competitors in AI answers, which may not be the ones you expect
Which pages were cited? The pages you need to be on, or to fix

A spreadsheet with one row per question, assistant and day is enough.

4. Repeat on different days

AI answers change between runs. One answer is a snapshot; a week of answers is a pattern. Our results below show why.

What we found when we checked ourselves

Since Oct 19, 2026 we've asked the same 8 questions every day through the AI providers' official APIs: "What is MentionSweep?" plus 7 buyer questions about AI visibility tools. This page uses 12 run days: Oct 19 to 24 and Oct 26 to 31 (our Oct 25 run didn't happen because of an outage on our side). Engines were added as we got access: ChatGPT and Gemini from Oct 19, Perplexity from Oct 20, Claude from Oct 21, Grok from Oct 22.

How we asked: ChatGPT through the OpenAI API (gpt-5-mini) once without web search and once with OpenAI's web search tool; Gemini without Google Search grounding; Claude (claude-haiku-5-5) and Grok (grok-4.20-0309-non-reasoning) without web search; Perplexity through its Agent API, which returned web citations in our runs. These are API answers, not the ChatGPT app or any signed-in app, so what you see in an app can differ.

Two gaps, said plainly: Perplexity couldn't be checked on Oct 30 and Oct 31, because our Perplexity API requests were refused for billing reasons, and ChatGPT without web search couldn't be checked for "What is MentionSweep?" on Oct 31 because of network errors on our side (twice). Those 3 are left out, not counted as "no".

"What is MentionSweep?": 63 answers over 12 days

Assistant and mode Answers Correct Made up a different product Said it didn't know, or couldn't find us
ChatGPT, no web search 11 0 3 8
Gemini, no web search 12 0 12 0
Claude, no web search 10 0 0 10
Grok, no web search 9 0 9 0
ChatGPT with web search 12 1 0 10 (+1 mixed)
Perplexity (web search) 9 5 0 4
All 63 6 24 32 (+1 mixed)

What the table says:

  • Without web search, 0 correct in 42 answers, and 24 made-up products. Gemini described MentionSweep as a social-listening or brand-monitoring tool all 12 days. Grok invented something different on different days, most often an OSINT tool for finding accounts that mention a username. ChatGPT without search did the same on 3 days and said it didn't know on the others. None of it is true: MentionSweep checks whether AI assistants name your brand.
  • Claude never made anything up. In all 10 of its answers it said it didn't have reliable information about MentionSweep. "I don't know" is the honest answer for a brand a model hasn't seen.
  • All 6 correct answers came from engines that searched the web, and every one cited a page about us. Perplexity was right on Oct 23, 24, 27, 28 and 29; ChatGPT with web search on Oct 29. Five of the six cited a directory listing about us on geodeck.io; on Oct 29 Perplexity cited our own About page and home page. Before that listing went live, the search engines answered "not found".
  • It doesn't stick. ChatGPT with web search was right on Oct 29 and back to "couldn't find a clear match" on Oct 30. On Oct 31 it found the same listing but mixed us up with a differently named "brand mention sweep" tool; we count that one as mixed, not correct. Perplexity was right on Oct 23 and 24, then not on Oct 26, then right again from Oct 27.
  • Listings get repeated, mistakes included. When Perplexity first described us from that listing, it repeated an error in it (that the free score includes Perplexity). Keep what third parties say about you accurate.

The buyer questions: 0 of 440

Across the same 12 days, the assistants gave 440 answers to our 7 buyer questions (API errors excluded). None of the 440 answers named MentionSweep. The tools they did name were mostly the established ones: Otterly.AI (not Otter.ai, the meeting-notes app), Peec AI, Profound, Semrush and Ahrefs. That's normal for a product whose website went live on Oct 19, 2026, and it's the point of checking: it shows where you stand, and who you're up against, before you spend anything on fixing it.

Correction (Nov 2, 2026): an earlier version of this page counted one answer fewer. Our count had missed one Oct 22 answer that a same-day retry recovered; it doesn't name MentionSweep.

What to do with what you find

  • An assistant makes you up: you can't edit a model's memory, but you can make the truth easy to find for the assistants that search. Publish a plain About page that says what you are and what you aren't (ours is here), and get listed accurately in your category's directories. In our tests, a listing on another site reached search-backed answers before our own pages did.
  • You're not named for buyer questions: look at who is named and which pages are cited, then get onto those pages or write the comparison pages buyers' questions lead to. Our guide on how to get recommended by ChatGPT goes through this.
  • Only some assistants know you: that's normal. Engines differ a lot in what they know. In our data Claude named the established AI visibility vendors far more often than Grok did, so check every assistant your buyers use rather than just one.

Doing it with MentionSweep

Checking 5 assistants by hand, on several days, takes time. MentionSweep's free score does the buyer-question part for you: it writes 10 questions your buyers might ask from what you sell, with your brand name left out, asks ChatGPT, Gemini, Claude and Grok each one through their APIs without web search (40 answers), and shows whether you're named, your position, and who is named instead, as a 0 to 100 score. It needs an email to send you a copy, and no card.

The $29 Full Report is a one-time check: 50 buyer questions on ChatGPT, Gemini, Claude, Grok and Perplexity, with Perplexity using its own web search, share of voice against up to 3 competitors, the sites AI cites in your category, and a fix list. Monitor ($39/month) repeats the check every week, so you see the trend instead of a snapshot. If you buy a Full Report and then start Monitor within 30 days, the $29 counts toward it.

What MentionSweep doesn't do: it asks through the official APIs, so it isn't the ChatGPT app with your account, memories and location, and the free score doesn't turn web search on. If you want to see the search-backed answer for a single question, ask it in the app yourself and note the sources, as in step 2.

Comparing tools first? See free AI visibility checkers, Grok visibility checkers, and our comparisons with Otterly, Peec AI, Profound, AthenaHQ, Scrunch, Semrush's AI Visibility Toolkit and Ahrefs Brand Radar.

A note on accuracy

AI assistants don't give the same answer twice, and every way of checking (app or API, signed in or not, with search or without) can give a different result. Our numbers are from 12 days of API answers to a fixed set of 8 questions about one new brand, so they show a pattern, not a law. MentionSweep isn't affiliated with OpenAI, Google, Anthropic, xAI or Perplexity.

Sources

  • OpenAI Help Center: Searching the web with ChatGPT (checked Oct 31, 2026). ChatGPT may search the web automatically; manual Search in the tools menu; citations; saved memories and approximate IP location can shape searches; "Search results and citations can be incomplete, outdated, or incorrect."
  • OpenAI API docs: Web search (checked Oct 31, 2026). Web search is a tool you enable in the Responses API; answers include sourced citations.
  • Google Gemini API docs: Grounding with Google Search (checked Oct 31, 2026). Connects Gemini to real-time web content and cites sources when the google_search tool is enabled.
  • Anthropic docs: Web search tool (checked Oct 31, 2026). Gives Claude real-time web content with citations.
  • xAI docs: Web Search and X Search (checked Oct 31, 2026).
  • Perplexity docs: Agent API quickstart (checked Oct 31, 2026). "Web search is opt-in"; without it, answers come from the model's own knowledge with no citations.
  • MentionSweep: our daily API checks, Oct 19 to 31, 2026 (no Oct 25 run); free score and plan details from our own product.

Want to see what ChatGPT, Gemini, Claude and Grok say when your buyers ask? Get your free AI visibility score.

FAQ

Questions, answered

How do I check what ChatGPT says about my brand?

Ask two kinds of questions: "What is <your brand>?" and the questions a buyer would ask without your name in them (for example "what's the best invoicing tool for freelancers?"). Note whether ChatGPT names you, whether what it says is true, who it names instead, and which sources it cites. Ask on more than one day, because answers change. OpenAI's help center says ChatGPT may search the web automatically when a question would benefit from current information, so check whether an answer shows sources: an answer with sources came from a web search, one without may come from what the model already knows.

Why does an AI assistant describe my brand as something it isn't?

Without web search, a model answers from its training data. If your brand is new or small, it may not be in that data, and some models then guess from the name. In our own API tests over 12 days (Oct 19 to 31, 2026), the engines without web search never described MentionSweep correctly in 42 answers, and made up a product in 24 of them. Claude was the exception: it said it didn't know in all 10 of its answers.

Which AI assistants search the web?

All five can, depending on how you ask. ChatGPT may search the web automatically, per OpenAI's help center. Through their APIs, OpenAI, Google (Gemini), Anthropic (Claude) and xAI (Grok) each document an optional web search tool that a developer has to turn on, and Perplexity's Agent API documentation says web search is opt-in there too. So the same assistant can give two different answers about your brand: one from memory and one from the web.

Does MentionSweep check all five assistants?

The free score asks ChatGPT, Gemini, Claude and Grok 10 buyer-style questions written from what you sell, with your brand name left out, through their APIs without web search: 40 answers and a 0 to 100 score. The $29 Full Report is a one-time check of 50 questions that adds Perplexity, which answers with web search. Monitor ($39/month) re-checks every week.

How often should I check?

More than once. Answers about a small brand change from day to day. In our own tests, ChatGPT with web search described MentionSweep correctly on one day (Oct 29) and couldn't find it the next. Look at the trend over several days or weeks, not a single answer.

Is ChatGPT recommending your competitor?

Check in about 30 seconds. Free, no card.

Get my free AI score