What AI says about your brand: how to check it

Run a hand check of AI answers this week, then track five numbers to see whether you are moving, and use the answers to decide what to write and fix.

Written by the BotSay team. Published 25 September 2026.

A buyer can now ask an AI assistant for a shortlist before visiting any vendor's site. The answer names a handful of brands, puts them in an order and gives each a sentence or two. A brand left out of it has missed the first cut. You can find out where you stand with a spreadsheet and an afternoon.

What an AI answer does for the buyer

Ask an assistant "What is the best payroll software for a 20 person company?" and it does the first round of research on the buyer's behalf. Many engines also link to the pages they drew on. A brand named early gets read first, and an accurate line that links to your own page gives the buyer an obvious next click.

Run a hand check this week

1. Write ten questions your buyers ask. Use their words, about the problems you solve, and leave your brand name out. Name yourself in the question and the engine will almost always repeat it back.
2. Ask each question on the six engines. Ask ChatGPT, Perplexity, Gemini and Claude. For Google AI Overviews, search the question on Google and read the overview when one appears; for Google AI Mode, ask it again in AI Mode. Open a fresh chat for every question, in a private window if you can, so your history stays out of it.
3. Log four things per answer: whether you were named, which rivals were named, where you came in the list, and which pages the answer linked to.
4. Copy the sentence that describes you. You will score it later.

One row of that log, before it becomes numbers, looks like this.

What one check tells you, and where it stops

After one pass you know which brands the engines put forward on your topics and which sites they lean on. Where a rival is the only brand in the answer, you have your starting point.

It is still one day and one wording, and answers shift from day to day even when nothing about you has changed. Ten questions on six engines give you 60 answers, enough to see where you stand and too few to tell a small move from ordinary variation. To see movement, ask the same questions in the same words every week. How many questions do you need to track AI visibility? shows how the count changes what you can say.

Five numbers to keep from week to week

Once you ask every week, five numbers track where you stand, all from the log you already keep.

• Mention rate. Of the answers you read, the share that named you.
• Share of voice. Of all the brand mentions in those answers, the share that was you.
• Average rank. When you are named, how early. Lower is better.
• Your pages cited. Of all the links the answers gave, the share that pointed at your sites.
• Sentiment. How favourably you are described. By hand, score the sentences you copied from 0 to 100 and take the average.

Watch Mention rate first, because Average rank and Sentiment only count answers that name you. Write the number of answers beside it every time. A rate built on a dozen answers can swing a long way on one reply, while one built on a few hundred holds steady enough to report, and to say so in the room with confidence.

Turn the answers into work

To move up, go back to the answers.

Where a rival is named and you are not, read what the answer says about that rival, then write the page that answers the same question for your product. When the same review sites, communities and publishers keep turning up in the links, get listed or discussed there. And if an engine describes you in a way you would change, find the page it drew on and correct it, whether it is yours or a listing you can edit.

Do first whatever could win you the most questions, and note the day each piece goes live, so the next check can be lined up against what you shipped.

When the weekly check gets heavy

The first check takes an afternoon. Keeping it going is harder, because 60 answers a week have to be read, counted and scored by hand, with the wording held exactly the same. That weekly workload is why we built BotSay.

BotSay asks your questions on all six engines every week, through each provider's official developer access or a licensed search-data provider, and works out the five measures for you. Mention rate comes with the number of answers behind it and a likely range, and BotSay calls a change real only once it is bigger than the uncertainty in both weeks' readings put together. The other four show the figure and how far it moved. For every answer, BotSay shows the searches the engine ran to write it, so you see what it went looking for before it named anyone.

Your reading arrives by email every Wednesday. It opens with the headline and the next thing to do.

If you stay with the spreadsheet, keep the questions and the wording fixed. Our full method is on How we measure if you want to copy it.