Journal  ›  Monitoring and prevention

How to track what AI says about your brand

There is no feed to subscribe to. You have to ask on a schedule and keep the record yourself.

Tomas Lindgren· AI Visibility Lead· · 3 min read
A notebook with a pen beside a screen
Short answer

Build a fixed prompt set, run it across several assistants on a schedule, and record the answers and cited sources verbatim. There is no alert mechanism for AI output, so the only way to see change is to ask the same questions the same way over time and keep your own log.

Build the prompt set once

Fifteen to twenty questions, written once and never changed, because the value is in comparability over time. Changing the wording resets your baseline.

  • What is [business]?
  • Is [business] reputable?
  • What are the complaints about [business]?
  • Who is [executive name]?
  • Should I use [business]?
  • What are the best [category] in [city]?
  • [Business] vs [competitor]
  • Has [business] been sued or investigated?

Record what matters

Log the date, the assistant, the exact question, the answer verbatim, and every source cited. The sources are the actionable part: they tell you which pages to address.

A spreadsheet is entirely adequate. The discipline of running it on the same schedule matters far more than the tooling.

Run it across at least three assistants, since they draw on overlapping but different sources and a problem visible in one is often invisible in another.

The answer tells you the symptom. The citation tells you the page to fix.

Reading the results

Consistent citation of one negative source across assistants means a source problem, and it is addressable. See how to remove false information from ChatGPT.

Assistants contradicting each other usually means thin ground truth, which is a publishing problem rather than a removal one. Details from another company appearing means entity confusion, covered in why AI says something wrong about your business.

Not appearing at all in category questions is its own finding, and increasingly a commercial one.

Common questions

How often should I run it?

Monthly for most businesses, weekly during or after an incident, and immediately after any source page is removed so you can see the change.

Do answers differ between users?

Yes, they vary by phrasing, session and personalisation, which is why a fixed prompt set and a consistent method matter more than any single answer.

Is there a tool for this?

Several vendors now offer prompt tracking. Check whether they genuinely query the assistants on a schedule or simply model what they think the answer would be.

TL
Tomas Lindgren
AI Visibility Lead, ReputationHound
Tomas tracks what assistants say about our clients. He works on the gap between what is published about you and what a model repeats when someone asks.

Find out if it can come down.

The scan is free and confidential. You get a viability answer on each item, a fixed price, and a timeline. Anything accepted and still published at 14 days is refunded in full.

Run a free scan