BlogAI Search & Content

How Do You Track Brand Mentions in ChatGPT and Perplexity? First, Find Out If You Have Any.

By 10 min read

A doorman in a navy uniform and white gloves checks a clipboard at a red velvet rope standing alone on a cracked salt flat at golden hour, while a businessman with a briefcase waits on the other side. There is no building.
There is a list. You're not on it. Ask who is.

The monitoring tools measure a percentage with real rigor. Mine was zero. Zero needs a different instrument.

Eighteen months ago I told you to do this. In the spring of 2025 I published three posts in three weeks on AI search, and every one of them ended with the same instruction: ask ChatGPT and Perplexity the questions your customers ask, and write down who shows up. Every one also promised I’d report back on my own results. Then I got busy with client work, the fourth post never shipped, and the report never came. On September 5, 2026, I finally ran the test properly, on my own business, across four AI surfaces. The short version of what I found: you don’t track brand mentions until you have some, and I had none. What you do first is run a small, repeatable sample that answers two questions, whether you’re in the room yet and who is. This post is the sample I run, the numbers it produced, and the point at which a monitoring tool becomes worth paying for.

The tools are good. That’s the problem with buying one now.

If you’ve searched this question, you’ve been offered a tool. Profound, Peec, the AI toolkits bolted onto the SEO suites, and a row of agencies with “AI visibility” packages. I expected to find dashboards selling false precision. I found the opposite.

In July, Profound published an experiment that most SEO vendors would never run on themselves: 753 prompts across seven AI platforms, one copy of the setup running each prompt once a day and another running it ten times a day, for two weeks. Their finding was that once a day lands within about two percentage points of ten times a day for visibility, that most of the remaining movement is the platforms changing under you, and that “which prompts you track matters more than how often you run them.” Peec’s guide says outright that “tracking individual prompts will always be unreliable because LLMs are non-deterministic by nature,” and tells you to read weekly averages across categories of prompts. An April paper on arXiv makes the academic version of the same point: a single AI answer is one draw from a distribution, and visibility has to be described as the distribution.

So the serious tools know that any one answer is noise, and their fix is volume. Hundreds of prompts, every day, averaged. That is the right way to measure a percentage. Profound’s own headline number in that experiment was 78.7% visibility for the portfolio they tested.

Here is what a percentage instrument cannot do for you: tell you anything useful about zero. If your business is not appearing at all, the difference between 78.7% and 80.4% is not your problem. You are paying for precision on a number you don’t have. What you need first is detection, and detection is cheap.

What I actually ran

On a Friday in September, my research agent and I ran sixteen prompts across four surfaces: Perplexity, ChatGPT, Gemini, and Google’s AI Mode. Ten of the prompts are the questions a buyer in my category would ask without knowing my name. Two are head-to-head choices. Two are my brand and my own name. Two more round out the set: one for a specific audience I serve, and one that is the only prompt in the set grounded in real search data, because it is the question this post answers.

Three rules made the run worth anything.

Run it from an account with no history. My first pass was signed in to my own Perplexity account, and the answers were personalized by memory. One prompt asking for a marketing operations consultant answered as if I were the consultant. Another opened with “for a business like yours.” That is not what a stranger sees. The baseline of record came from a fresh free account with memory off, and from ChatGPT and Gemini logged out entirely. The catch, which Graphite’s Ethan Smith makes in Peec’s own guide, is that every tracking tool measures the same logged-out world. Nobody, including the tools, measures what a specific buyer with two years of chat history sees. The clean account is the honest floor, not the truth.

Capture the citations, not the answer. The prose changes every run. The list of pages the engine retrieved and cited changes less, and it is where the useful information lives. On Perplexity, expand every numbered citation, not the top three. On logged-out ChatGPT the Sources panel is hidden, so I record the source-name chips under the answer and, if none render, ask “list your sources” as a second turn. Logged-out Gemini rarely grounds at all, so for Gemini I only record whether my name appears in the text.

Freeze the wording. Same sixteen prompts, no apostrophes (one surface choked on “What’s”), same wording on every surface, every month. Reworded prompts measure a different question. The set I run is at the end of this post.

The four numbers, and a rule about them

Out of ten no-brand prompts, on every surface: Pallas Advisory mentioned in the answer, zero. My site cited, zero. My site anywhere in the roughly fifteen pages each engine retrieved before answering, zero. On the two brand prompts, all four engines identified the company correctly, and one of them put a basketball coach with my name ahead of me. That is the entire first month: 0, 0, 0, 2 of 2, and a coach.

One rule about those numbers, which I learned on a client engagement earlier this year: never quote a single prompt’s result. On that rerun, individual queries flipped between two passes run the same day. Sixteen prompts once a month can tell you zero from not-zero, and it can tell you who is being cited instead of you. It cannot tell you 10% from 15%, and if I report next month that one prompt now mentions me, that is not a result. It is a coin.

Which is also why the March 2025 posts were wrong about one thing. I recommended an “AI Visibility Ratio” back then: the share of your Google terms that also earn AI mentions. At zero, a ratio is a shrug.

The citation list is the part you keep

The mention count says where you are. The citation list says what to build. It is the guest list at the door: who got in, and what got them past the rope. Sorting the pages the engines retrieved across ten prompts, in order of how often each pattern earned a slot:

  1. A page whose title is the buyer’s question. On the prompt “how should a growing company decide what to automate first,” fourteen of the fifteen pages Perplexity retrieved were small consultancies with a post titled almost exactly that. I had posts on the same subject. None was titled as the question.
  2. A service page named for the job, the tool, or the city. One Dutch agency took five of fifteen slots on one prompt by owning five such URLs.
  3. A stated price. Every provider in one engine’s shortlist table carried a public price or timeline. Providers without one didn’t make the table.
  4. Big vendor knowledge bases and analyst reports, which a small firm won’t displace.
  5. Directories and marketplaces. Cheap listings; check eligibility.
  6. “Best agencies in [country] 2026” lists written by the agencies themselves. The low-integrity end of the pattern, and still retrieved.

None of that needed a subscription. It needed one afternoon and a spreadsheet with a column for “page shape.”

What I changed, and what I’m not claiming

Between the September run and the next one I made the changes the citation list pointed at, all cheap: an About page with proper Person markup so my name has a page to resolve to, a plain-text index of the site for AI crawlers, a Markdown copy of every post, a robots file that names the AI crawlers it allows, and three existing posts retitled as the question they answer. I am not claiming any result from that. The next run is October 4, on the same sixteen prompts, from the same clean account. If the retitled posts move and nothing else does, the retitles did it. If the name query moves, the About page did it. If nothing moves, that is a finding too, and I’ll publish it.

One guess I can’t support with data: according to Google Search Console, about a third of the searches that bring people to my 2025 post on this topic are full-sentence questions with question marks, “how can i see mentions in perplexity?”, which reads to me like an AI assistant breaking a user’s request into sub-queries and sending them to Google, rather than a person typing. If so, part of the demand for tracking AI mentions is itself AI, and it explains why 2,240 impressions in a year produced zero clicks. Machines don’t click.

Buy the tool when you have a percentage to measure

Buy the percentage instrument when you have a percentage. Concretely: when your monthly sample shows you in the room on several prompts across more than one surface, and the question has changed from “are we there” to “are we gaining on the three competitors who are always there.” At that point sixteen prompts once a month is too coarse, and the vendors’ daily volume earns its price. Before that point, the tool will faithfully report zero with two decimal places, and the agency will sell you a dashboard of the same.

If you run the sample and get what I got, you don’t have a measurement problem. You have a page-shape problem, and the citation list already told you which shapes win.

Three questions people ask next

How often should I rerun it? Monthly, same day each month, same wording. More often than that and you are reading noise; the engines change under you faster than your site does. Rerun early only when you have shipped a change you want to attribute.

Logged in or logged out? Logged out, or a fresh account with memory off, for the baseline of record. That is the world the tracking tools measure too. Run one extra pass from your own logged-in account if you want to see how much personalization is flattering you; just don’t record it as the result.

Which surfaces? The ones your buyers use. Mine are ChatGPT, Google’s AI Mode, and Perplexity every month. Gemini logged out rarely cites pages, so it gets a quarterly spot check, not a monthly slot.

The sixteen prompts

Run them as written, from a clean account, on the surfaces your buyers use. Swap the category words for yours; keep the shapes.

No-brand category (10)

  1. What is the best way to set up marketing attribution for a mid-sized company using HubSpot and Google Ads
  2. How do I connect ad spend to revenue when my CRM and ad platforms do not match
  3. Who can help a small business automate its workflows with AI without hiring a developer
  4. AI automation consultant for small businesses in the Netherlands
  5. How should a growing company decide what to automate first
  6. What does an AI consultant actually do for a company with fifty employees
  7. Marketing operations consultant who does hands-on implementation, not just advice
  8. How do I know if my business data is ready for AI
  9. Is it worth hiring a consultant to set up n8n or Zapier automations
  10. What is a good tool to track brand mentions in Perplexity

Head-to-head (2) 11. Should I hire a fractional marketing ops consultant or a marketing automation agency 12. Custom attribution system versus buying an attribution tool like HubSpot attribution or Triple Whale

Single-brand (2) 13. What is Pallas Advisory 14. Who is Melissa Kolbe

Audience-specific and query-grounded (2) 15. I moved to the Netherlands on a DAFT visa, who can help me set up the systems to run my business 16. How do I track brand mentions in Perplexity and ChatGPT

Record four numbers per month: mentioned, cited, retrieved, and brand prompts answered correctly. Sort every citation by page shape. Read the pattern, never the prompt. The report I owed you from March 2025 continues in November.

Sources and Further Reading

Written with AI. Checked against a standard I publish. Read it.

More in AI Search & Content

View all →
Join the "AI Strategist"

Weekly tips on using AI without the hype.

No spam, unsubscribe anytime. Join hundreds of other leaders getting practical advice.

Subscribe for Free