How to Know If ChatGPT and Gemini Recommend Your Brand

To find out whether ChatGPT and Gemini recommend your brand (what GEO calls AI visibility), run a manual test: write 10 questions a real customer would ask, run each one 3 times in each tool, in new conversations, and note in a spreadsheet whether the brand appeared, whether there was a link and whether the description was correct. A single answer proves nothing, because AI changes its list of brands on every attempt.

Quick summary

• The manual test takes 2 to 3 hours and costs nothing. What it produces is a rough snapshot, not an exact number.
• Use questions that do not include your brand name (the way a customer asks) and repeat each one at least 3 times.
• Note three things per answer: was the brand mentioned, was it cited with a link, and was it described correctly.
• A 2026 SparkToro study found less than a 1 in 100 chance that AI repeats the same list of brands, so only frequency across many attempts counts.
• Automate when you pass a few dozen answers a month or when you need history.

This test is the first stage of the map described in SEO, AEO and GEO: what is the difference. If the result is poor, the next step is how to show up in ChatGPT.

How to read this post: if you have never tested, follow the "Step by step" section from start to finish. If you already measure and want rigor, go to the limits section and the reading method, then to the prompts and metrics posts.

Why isn't a single answer proof?

Because AI never gives the same answer twice. SparkToro and Gumshoe.ai asked 600 volunteers to run 12 prompts in ChatGPT, Claude and Google AI, for a total of 2,961 runs between November and December 2025. The result, published in January 2026: there is less than a 1 in 100 chance that the list of brands repeats and less than 1 in 1,000 that it repeats in the same order, according to the SparkToro study.

The same study has a useful side: the strongest brands in a category showed up often (in headphones, Bose, Sony, Sennheiser and Apple appeared in 55% to 77% of answers). The authors' conclusion is that the percentage of times a brand appears, across many runs, is a reasonable measure, and that "ranking position in AI" is not.

In plain English: asking AI once whether it recommends your brand is like flipping a coin once to find out if it is rigged. What matters is how many times it lands heads in 30 flips.

For SEO professionals: treat each run as a sample from a distribution, not as a fact. The right research question is "what is the probability the brand appears for this set of prompts?", estimated by frequency. In how to measure AI visibility I show how to calculate the margin of error.

What do I need before I start?

Three things: a list of questions, a spreadsheet and recording discipline. You do not need a paid tool.

• Access to ChatGPT and Gemini. Ideally, also run each attempt in a new conversation. If you use memory features or custom instructions, remember they can influence the answer, so turn them off or use a logged-out browser for the test, and note which choice you made.
• The list of competitors you expect to see in the answers (3 to 5 names).
• A spreadsheet with the columns in the table below.

Column	What to record
Date	Day of the test (answers change over time)
Tool	ChatGPT or Gemini, and whether web search was on
Question	The exact text used
Attempt	1, 2 or 3
Brand mentioned?	Yes or no
Link to your site?	Yes or no (ChatGPT and Gemini show sources in some answers)
Correct description?	Yes, no or partial; copy the passage if it is wrong
Who appeared	Brands cited, in the order they appeared

Gemini does not always show sources. According to the Gemini Apps Help Center, the app "may" provide links to sources and related content, which means the absence of a link is not an error in your test.

Step by step: how to run the test in one afternoon

1. Write 10 questions that do not mention your brand name. Use the ones a customer would ask ("what's the best specialty coffee roaster to subscribe to online?"), not the one you would like to hear.
2. Split them into three groups: 4 discovery questions ("what to look for in..."), 4 comparison questions ("best X for Y") and 2 decision questions ("is X worth it in [city]?"). The full bank, with ready-made templates, is in prompts to test your brand's visibility.
3. Open a new conversation for each attempt, in the chosen tool. Reusing the same conversation contaminates the answer with what was already said.
4. Run each question 3 times and paste the result into the spreadsheet. That is 30 answers per tool, 60 in total.
5. Ask 2 direct questions about the brand: "What do you know about [your brand]?" and "Is [your brand] reliable for [service]?". They measure whether AI knows the brand and whether what it knows is right. They are not recommendation questions.
6. Calculate a simple rate: answers with the brand divided by total answers. Example: 4 mentions in 30 answers gives 13%.
7. Also read what is wrong, not just what is missing. An incorrect description is more urgent than not appearing.

In plain English: step 5 is the "name test." If AI does not even know who you are when asked directly, the problem comes before any recommendation question.

How do I interpret the results?

Compare the mention rate with what you would expect and with your competitors. The table summarizes what each pattern usually indicates. They are hypotheses to investigate, not diagnoses.

What you saw	Hypothesis to investigate	Where to act
Never appears, not even in the name test	AI has little information about the brand	External mentions, profiles, "About" page
Appears in the name test but not in general questions	Known, but not associated with the category	Content that links the brand to the problem; comparisons
Appears without a link	Cited from memory, or source not identified	Clear, indexable pages; robots.txt and crawlers
Wrong or outdated description	Old or ambiguous sources	Update "About," profiles and service pages
Competitors always appear, you almost never	They have more presence on the web	See Reddit, YouTube and Quora in AI citations

An Ahrefs study of 75,000 brands found a strong correlation between a brand's presence in AI answers and mentions of it on the web and on YouTube, according to Ahrefs' analysis. It is correlation, not proof of cause, but it indicates where to start.

Hypothetical example: an online coffee roaster

Hypothetical example: "Blue Ridge Roasters," a fictional roaster that sells a specialty coffee subscription, runs the 10 questions, 3 times, in ChatGPT and Gemini.

In ChatGPT, the brand appears in 4 of 30 answers (13%). In Gemini, in 9 of 30 (30%). In the name test, ChatGPT describes Blue Ridge as "a coffee shop in Seattle," which is wrong, since it is an e-commerce business.

Honest reading: with 30 answers, ChatGPT's 13% rate is compatible with something between about 5% and 30%, and Gemini's 30% with something between about 17% and 48% (95% confidence interval, Wilson method). The ranges partly overlap, so you cannot say with confidence that Gemini "likes" the brand more. What you can state: the wrong description is a concrete finding, and the action is to fix "About" and the profiles to say, unambiguously, that the company is a subscription e-commerce business.

What are the limits of the manual test?

The manual test shows a small sample, at one moment, with your way of asking. That limits what it can claim:

• Small sample: 30 answers give a wide range. Two measurements taken weeks apart can differ by chance alone.
• Asker bias: your vocabulary is not your customer's. Ask two or three people outside the company to write questions.
• Personalization: memory, location and history can change the answer. Record the conditions.
• Product changes: models and search features get updated, so a rise or drop may come from the tool, not from your work.
• Appearances are not visits: appearing does not mean generating clicks. Check traffic sources in GA4 and Search Console.

For SEO professionals: record the date, the selected model, whether web search was on and the country/language. Without those variables, a time series is not comparable. Avoid "position" metrics within the answer: SparkToro itself considers them invalid.

When is it worth automating?

When you need history, many questions or alerts, manual work stops paying off. As a rule of thumb, if you pass about 100 answers per round or need to repeat every week, automate. There are several tools on the market; see the criteria in tools to monitor AI citations. Contentor's AI Visibility tracks, for the questions you choose, whether ChatGPT cites the brand. It covers ChatGPT only, so Gemini stays in the manual test.

Frequently asked questions

What is the minimum number of questions and repeats?

For a first reading, 10 questions with 3 repeats (30 answers per tool). That is enough to find gross problems, such as a wrong description or total absence, not to measure small variations. To compare periods with confidence, you need far more answers. The margin-of-error calculation is left for the metrics post.

Can I use my brand name in the questions?

Only in control questions. If you write the brand in the prompt, AI will cite it because you asked, and that measures knowledge, not recommendation. To find out whether you appear when the customer does not know the brand yet, ask without the name, the way they would.

Why do ChatGPT and Gemini give different answers?

They are different systems, with their own models, search sources and rules. Ahrefs observed that only 7 of the 50 most-cited domains appeared in ChatGPT, Perplexity and Google AI Overviews at the same time, according to Ahrefs' evidence review. So measure each tool separately and do not mix the numbers.

What if AI says something wrong about my company?

Note the passage and the date, and fix the source: "About" page, social profiles and directories, service pages. There is no way to "edit" the AI directly. Repeat the test in a few weeks, since sources are re-read, but with no guaranteed timeline. If the wrong information is serious (health, finance), also record the context, so you can report it to the platform.

How often should I redo the test?

Monthly is enough for most small businesses, plus after every major change to the site or content strategy. Keep the same questions and the same method. One-off tests that differ from each other do not form a comparable series.

Next step

Now that you have a first measurement, refine the questions with the prompt bank for testing your brand in AI. When you want more rigorous indicators, move on to how to measure AI visibility. If you plan to build a full routine, the GEO, AEO and SEO checklist organizes actions by priority.

Read also:

• How to show up in ChatGPT
• Tools to monitor AI citations
• SEO, AEO and GEO glossary