Seenford
Sign in

How to track brand mentions in ChatGPT

Typing the brand's name into ChatGPT and reading the answer feels like a check. It isn't one. The answer changes from one ask to the next, it changes with the country you ask from, and asking by name tells you nothing about whether ChatGPT would bring the brand up on its own.

This guide sets out a method that holds up: what to ask, what to record, and how to read the result.

Why a quick manual check misleads

  • Answers vary. Ask the same question twice and you can get different brands. One answer is a sample, not a verdict.
  • Asking by name is a leading question. "What do you know about the brand?" always gets the brand talked about. Buyers rarely ask that way. They ask "which … would you recommend?"
  • Location matters. ChatGPT's answer depends on where the question is asked from. A check from a London desk says little about a buyer in New York.
  • Being named isn't being recommended. A brand can appear in a list of ten with no endorsement, or appear with a warning attached. Both count as "mentioned". Neither is a win.

Ask two kinds of question

  1. A question that leaves the brand out. "Which folding bikes would you recommend for a city commute, and why?" This is how buyers ask. If the brand appears here, ChatGPT found it unprompted.
  2. A question that names the brand. "What do you know about the brand, and would you recommend it?" This shows what ChatGPT believes about the brand once it's asked directly, including any mix-ups with similar names.

Keep at least half of the questions free of the brand name, so the reading reflects how buyers really ask.

Record a verdict, not a count

For every answer, record which of these it is:

  • Top pick: the brand is named first, or highlighted as the best option available.
  • Recommended: ChatGPT suggests the brand as a strong, reliable choice.
  • Mentioned in passing: the brand's name appears, but ChatGPT doesn't back or endorse it.
  • Named with caution: ChatGPT names the brand but warns buyers about it, so the mention is a risk, not a win.
  • Not mentioned: the brand doesn't show up at all.

Then keep the quote: the exact sentence that earned the verdict. Without the quote, nobody can check the verdict later, and nobody can see what to fix.

Found vs Recommended

These two answer different questions, so keep them apart.

  • Found counts only the questions that leave the brand name out. It answers: does ChatGPT bring the brand up on its own?
  • Recommended can also come from questions that name the brand. It answers: once the brand is in the conversation, does ChatGPT back it?

A brand can be found in none of the unprompted answers, yet be recommended when asked by name. That gap is the most useful thing a check can show: ChatGPT thinks well of the brand but doesn't think of it.

A worked example: Seenford, checked on itself

On 16 Oct 2026 we ran a Free Report on Seenford across six engines. The unprompted question was "Which AI visibility tools for brand teams would you recommend, and why?"

  • Found: 0 of 6 engines named Seenford unprompted.
  • Asked by name: one engine described Seenford correctly and recommended it: "Yes, for brands and agencies actively investing in AI Search Optimization (AEO/GEO)."
  • Mix-ups: the other five confused Seenford with businesses that have similar names: a law firm, a car dealer, an online shop and a credit company. One answered: "I would not recommend Seenford", describing a different company.

That is a new brand's typical starting point: little unprompted presence, and a name that engines confuse with others. The fix is specific: make it clear, on the site and on other sites, who Seenford is and what it does. Then measure again.

Repeat on a schedule

A single day is one sample. Ask the same questions on a schedule, from the market the brand sells in, and look at the pattern over weeks, not one answer. Compare like with like: the same questions, the same market, the same engine.

Do it with Seenford

Seenford asks ChatGPT and five other engines these questions on a schedule, records a verdict for every answer, and keeps the quote behind it.