Reading time: 13 minutes
To see if your business shows up in ChatGPT, ask it, Gemini and Google a customer’s questions without naming your brand. This free test needs no login and no AI visibility checker. However, it only gives a snapshot of your visibility in ChatGPT and other engines, because the answer changes every time.
In short
- To check AI visibility for free, ask like a customer, without naming the brand, logged out or in an unpersonalized temporary chat.
- When you’re logged in, a ChatGPT visibility test can mislead you, because ChatGPT may use memories and past chats, even in search. So test outside your account or without personalization.
- One round measures nothing. After all, in a SparkToro study, the chance of ChatGPT repeating the same list of brands was below 1 in 100.
- A free AI visibility checker can measure very different things, such as its own question database or what the model knows about the brand. A paid tool, by contrast, runs your questions and keeps the history.
How do you test AI visibility by hand in ChatGPT, Gemini and Google?
Open each engine logged out or in a chat without personalization, and ask the same questions. For a Brazilian brand, ask from Brazil, in Portuguese. That way, the answer is closer to what a new customer sees, not to what the AI knows about you. Then note whether the brand appears, who appears instead and which sources come up.
In ChatGPT: logged out or in a temporary chat
The simplest ChatGPT visibility test is to open chatgpt.com in an incognito window, without signing in. According to OpenAI, people who are not signed in can also use web search. In addition, ChatGPT estimates your location from your IP address. So run the test from your customers’ country and, for a local question, include the city.
When you’re logged in, the risk is different. In a regular chat with Memory on, ChatGPT may use past chats, saved memories and custom instructions. According to OpenAI, memory can even change the query it sends to search. In other words, if you’ve already asked ChatGPT to help promote your business, it may cite the business because of that memory.
To avoid this in your account, open a temporary chat and choose Unpersonalized before the first message. In this mode, ChatGPT doesn’t use memory, custom instructions or plugins. However, you can’t change the option once the conversation starts.
Run each question the default way first, as a customer would, since ChatGPT decides on its own when to search the web. If you want to see the sources, run it again with Search turned on, under View all tools.
In Gemini and Google Search
In some locations and on some devices, you can use Gemini without signing in, according to Google’s help. Without signing in, Gemini doesn’t tailor answers to your preferences or past chats. If you’re signed in with a personal Google Account, use a temporary chat, which doesn’t personalize answers.
In Google Search, results also change with time, location, language and account history, according to Search Help. To see the version without personalization, use an incognito window or the Try without link in the footer of the results page.
Then record two answers per question: the AI Overview, when there is one, and the AI Mode tab. The two features may use different models and techniques, according to Google’s documentation. In addition, AI Overviews only show up when they add something to classic Search. So also note when one doesn’t appear.
What to record in each answer
To compare visibility in ChatGPT, Gemini and Google, use a spreadsheet with one row per answer. In it, record the date, engine, mode and question. Then answer five questions about what the AI said:
- Did the brand appear? Also mark whether it came as a recommendation or just in passing.
- Did the site become a source? Check the citations and the Sources button, when there is one.
- Who appeared instead? These names are the competitors the AI sees.
- Which sites did the AI cite? Press, guides and directories show where the brand needs to be.
- Did the AI get anything wrong? Note a wrong price, address or service, with the cited source.
Which questions should an AI visibility test use?
Use the questions a customer asks before they know the brand: the category plus the city, their problem and a comparison between options. Then add two or three questions with the business name. They only serve to check whether the AI gets price, address and services right.
Take, for example, a language school in Recife, a city in northeastern Brazil. A quick test with ten questions could include these, in Portuguese, with a translation in parentheses:
- Category: “qual a melhor escola de inglês para adultos em Recife?” (“what’s the best English school for adults in Recife?”)
- Problem: “preciso de inglês para uma entrevista em dois meses, onde estudar em Recife?” (“I need English for a job interview in two months, where should I study in Recife?”)
- Comparison: “aula particular ou escola de idiomas, o que compensa em Recife?” (“private lessons or a language school, which is worth it in Recife?”)
- Brand: “a Escola X é boa?” (“is Escola X good?”) and “quanto custa o curso da Escola X?” (“how much does the Escola X course cost?”)
Asking only “você conhece a Escola X?” (“do you know Escola X?”) tests whether the model knows the brand exists. A question without the brand, on the other hand, shows whether it makes the recommendation, which is what brings in customers. That’s why most of the list doesn’t name the business.
Write the way customers write, in their language, and take phrases from WhatsApp, phone calls and the sales team. Also include the city or neighborhood. After all, OpenAI itself suggests adding your city, neighborhood or postal code to the question when local results show the wrong area.
Why doesn’t one test measure AI visibility?
Because the same question gets different answers on every run, and each person gets their own version. After all, the model, the location, the account’s memory and the search sources all change. So one round can show the brand by luck or hide it by chance, without saying anything about how often it really appears.
In a SparkToro study, 600 volunteers ran 12 prompts 2,961 times in ChatGPT, Claude and Google’s AI. The chance of ChatGPT repeating the same list of brands in two answers was below 1 in 100. The study involved Gumshoe.ai, a startup that sells AI tracking.
The plan matters too. According to OpenAI’s release notes, Free and Go use a different model from Plus and Pro. Therefore, if you measure ChatGPT visibility only on a paid account, you may see a different answer from the one a customer on the free plan gets.
Finally, count the sample size. With 10 questions run 3 times in one engine, you get 30 answers. With that sample, the margin of error for the presence rate ranges from 18 to 31 points, up or down. The math, at 95% confidence, is in our guide to measuring brand mentions and citations. That’s why the free test works as a snapshot, not as a measurement.
How often should you repeat the test?
In the quick test, run each question three times, in new chats and on different days. For a firmer diagnosis, the free GEO audit calls for five runs per engine. Then repeat the list every month and after each major change, such as a new page, a press story or a fix on the site. That way, you compare snapshots taken the same way.
Keep the questions, engines, logged-out mode and city the same, so that one month’s ChatGPT visibility compares with the next. After all, if the test changes, the difference may come from the test, not the brand. The short interval also makes sense, because sources rotate fast. In a SISTRIX study in six countries, not including Brazil, AI Mode replaced 56% of cited sources every week.
To measure a trend, however, the sample needs to grow. The protocol in the measurement guide uses 50 questions, 5 runs and 4 engines per month, or 1,000 answers per brand. At that volume, manual collection no longer works, and that’s when a tool comes in.
What does a free AI visibility checker measure?
Each free AI visibility checker measures something different, even when the names sound alike. Some query their own question database, others ask the model what it knows about the brand, and others score a page. So the scores can’t be compared with each other, and they don’t replace your customers’ questions.
| Checker | What it measures | Limit |
|---|---|---|
| Ahrefs | Brand mentions in Ahrefs’ question database | Shows only the top 5 items in each section |
| Semrush | 0 to 100 score and mentions, based on the domain | Uses Semrush’s database, not your questions |
| HubSpot | What ChatGPT, Perplexity and Gemini say about the brand from training | Full report requires a form |
| RD Station | Presence in the segment and location, in ChatGPT and Gemini | Doesn’t say how many questions it runs |
| Frase | 0 to 100 score of how citable a page is | Doesn’t look at AI answers |
How each free AI visibility checker works
Ahrefs and Semrush start from their own question databases. The free AI visibility checker from Ahrefs requires no signup and uses questions drawn from real searches. It covers ChatGPT, Gemini, Perplexity, Copilot and Google’s AI. The Semrush checker, on the other hand, starts from your domain and draws on more than 26 million prompts and responses. In both, however, you don’t choose the questions.
HubSpot’s AI Search Grader asks GPT-5.4 mini, Perplexity and Gemini how they describe the brand based on their training. In its 0 to 100 score, sentiment is worth up to 40 points. So it measures the brand’s reputation in the model, not whether the brand makes it into a recommendation with search.
Radar GEO, a Brazilian tool from RD Station, asks for your segment, location and business email. In return, it shows presence in ChatGPT and Gemini, sentiment and competitors. Frase’s GEO Score, on the other hand, rates the text of a page without looking at what AI answers. All five companies sell paid marketing or SEO software.
What changes with a paid tool?
Unlike a free AI visibility checker, a paid tool runs your questions every day or every week. It also keeps the history of your visibility in ChatGPT and other engines. In general, it compares the brand with competitors, shows the cited sources and, in some tools, exports the answers. That way, the snapshot becomes a series you can follow.
The vendors themselves draw this line. Ahrefs, for example, calls its free AI visibility checker a one-time snapshot. History, share of voice and your own questions are in Brand Radar, which is paid. Custom prompt packages start at $50 per month, and Ahrefs’ question database at $199. Similarly, HubSpot sells HubSpot AEO for $50 per month, with specific questions and competitors week over week.
In our comparison of GEO tools, entry plans range from $29 to $99 per month. In addition, Peec AI, Otterly and Profound offer free trials. A tool, however, only measures. The work of getting the brand into the answers is GEO, which our guide to what GEO is covers. NotHuman, which publishes this blog, also works in this market.
When is a free AI visibility test no longer enough?
A free test stops being enough when the question changes from “does the brand show up?” to “is the brand growing?”. In practice, this happens in four situations:
- Trend: you need to compare months, with a smaller margin of error.
- Competition: you want to know the brand’s share against three to five rivals.
- Volume: collection goes beyond a few hundred answers a month.
- Reporting: the results go to leadership or to a client, with history and exported answers.
If the brand doesn’t show up in any answer, on the other hand, a tool will just record its absence every day. In that case, start with the free GEO audit, which checks crawlers, indexing and sources. Then see how to show up in ChatGPT.
What we found when we analyzed 8 guides on AI visibility
We analyzed 8 Portuguese-language pages that teach how to check visibility in ChatGPT or other AIs. All of them tell you to ask category questions, without the brand name. However, only 3 tell you to test logged out or in an incognito window. In addition, only 2 warn that the account’s memory and history change the answer. None mentions ChatGPT’s temporary chat.
- Repetition: 5 warn that the answer changes with every run, but only 2 say how many times to repeat, from 3 to 5.
- Sources: 5 tell you to note the sources AI cites.
- Checkers: only 2 recommend a free AI visibility checker, and only 1 says what it leaves out.
- Commercial interest: 7 of the 8 come from agencies or companies that sell digital marketing services or tools.
Frequently asked questions
No. According to OpenAI, web search works on the free plan and even for people who aren’t signed in. The plan, however, changes the model: Free and Go use a different model from Plus and Pro. So if you test ChatGPT visibility only on a paid account, you may see a different answer from the one a customer on the free plan gets.
It only checks what the model knows about the brand, such as services, price and address. The question doesn’t show whether the business makes the recommendation, because customers ask about the category without naming it. So an AI visibility test starts with questions without the brand, and the questions with the name help you find errors.
In GA4, look at the AI Assistant channel, which groups visits from sources such as ChatGPT, Gemini and Copilot, according to Google Analytics Help. Clicks from AI Overviews and AI Mode, however, fall outside this channel. The step-by-step process is in our guide to AI traffic in GA4.
Want to know how your brand shows up in AI answers today?
NotHuman is opening spots for agencies and brands to track how they appear in AI answers. Join the waitlist and we’ll let you know when a spot opens up for you.
Main sources
- OpenAI, Searching the web with ChatGPT (2026).
- OpenAI, Temporary chat in ChatGPT (2026).
- Google, Use Gemini Apps, including without signing in (2026).
- Google, Why your Google Search results differ from others (2026).
- SparkToro, research on how inconsistent AI brand recommendations are (2026).
- Ahrefs, free AI visibility checker (2026).
- HubSpot, AI Search Grader (2026).
- RD Station, Radar GEO, in Portuguese (2026).
