Guide

How to check whether ChatGPT, Perplexity and Gemini mention your brand

By Mandel Sherrod · Updated October 19, 2026

You can't improve what you don't measure, and no AI assistant gives you a dashboard. This is the method we use, including our own starting numbers.

Step 1: Build your question list

Your list is the ruler you'll measure with for months, so take care building it. Use the words customers use, not the words in your marketing.

• "Best" questions: "best [category] for [type of customer]", "best [service] near [place]"
• Problem questions: "how do I [job your product does]"
• Comparison questions: "[competitor] vs [competitor]", "alternatives to [competitor]"
• Decision questions: "is [category] worth it for [situation]", "how much does [service] cost"
• One branded question: "What is [your brand] ([your domain])?" This one tests accuracy, not visibility.

Good sources: recordings or notes from sales calls, support emails, the Queries report in Google Search Console, and the questions people ask in your industry's forums. Leave out questions only you would ask.

Step 2: Ask in clean conditions

Assistants personalize, and they vary. Make the test as fair and repeatable as you can:

1. Use a temporary or incognito chat, or switch memory off, so past conversations don't steer the answer.
2. Make sure web search is on. That's the mode that cites sources, and it's the one your fixes can influence fastest.
3. Paste the question exactly as written, every time. Don't add "include [your brand]".
4. Ask each question two or three times. If you're named once out of three, record it that way. That's real information.
5. Note the date, the engine, and the mode or model shown.

Step 3: Record the same fields every time

A spreadsheet works fine. One row per question, per engine, per run:

Step 4: Turn answers into three numbers

• Mention rate = answers that name you ÷ total answers (leave the branded question out).
• Citation rate = answers that link to your site ÷ total answers.
• Share of voice = your mentions ÷ (your mentions + mentions of the competitors you track).

Here's a real example from our own baseline. In October 2026 we ran our monitoring pipeline on our own brand. Gemini, with Google Search on, answered 12 of our 13 non-branded questions; one hit an API rate limit and wasn't counted. It named us in 0 of the 12, so our mention rate was 0%. Companies on our competitor list were named in several of the same answers. As a brand-new service that's the expected starting point. It's still worth writing down, because it's the line we measure against.

The "sources cited" column is often the most useful. It shows which sites the engines trust for your topic, and so where being listed, reviewed or quoted would help most.

Step 5: Check what AI says about you

Being named with the wrong details can be worse than not being named. Ask the branded question in every engine and compare the answer with your real facts. Look for:

• Old prices, discontinued products or a former business name
• Confusion with a different company that has a similar name
• Features you don't have, or missing ones you do

When we asked Gemini "What is The Companion Guide (thecompanionguide.com)?", it described a companion to a homeschool history curriculum, which is a different organization. The fix is on our side: say what we are clearly and consistently on our own site and profiles, so engines have something better to go on. Our guide to getting recommended by ChatGPT covers how.

Step 6: Automate it (and the caveat)

Doing this by hand for 20 questions across five engines every week gets tedious. Most of it can be scripted with the engines' developer APIs. Google's Gemini API can ground answers in Google Search and return the sources. OpenAI's and Anthropic's APIs offer web-search tools. Perplexity's Sonar API returns citations. A script can ask each question, then check the answer text for your brand and the citations for your domain.

The caveat: API answers aren't identical to what a consumer sees in the app. Models, settings and personalization differ. Treat automated results as a consistent instrument for spotting trends, and spot-check the real apps by hand now and then. Some engines, including Copilot and Google's AI Overviews, have no general answer API. For those you need manual checks or a third-party data provider.

Common mistakes

• Changing the questions every month. Then you can't compare. Add new questions, but keep the core set fixed.
• Judging on one answer. Assistants vary. A single bad result isn't a trend.
• Leading the witness. "Is [your brand] the best option for X?" measures nothing.
• Only tracking yourself. Without competitors on the list, you can't tell your progress apart from the whole category moving.

Questions