How to Monitor Brand Mentions in AI Search (Without Fooling Yourself)
By Keith Schilling · August 7, 2026 · 6 min read
Monitoring brand mentions in AI search sounds like a solved problem — surely there's a Google Alerts for ChatGPT? There isn't, and the reason is structural: AI answers aren't published anywhere. Each one is generated on demand, shown to one person, and gone. You can't subscribe to them. The only way to know what engines say about you is to ask them yourself, systematically.
That said, there's a right way to do this at every budget, including free. Here are the three tiers, and the traps in each.
Tier 1: the manual spot check (free, twenty minutes)
Write down ten questions a real buyer would ask in your category — "best [category] for [segment]," "[you] vs [rival]," "[category] pricing." Ask them in ChatGPT, Perplexity, and Gemini. Tally three things per answer: were you mentioned, were you recommended, was your site cited as a source.
The trap: doing this once and treating it as truth. AI answers are stochastic — the same question re-asked produces a different brand list often enough that a single pass can miss you (or flatter you) by pure chance. If you're doing manual checks, ask each question at least twice on different days before you believe anything.
Tier 2: watch the exhaust in your analytics
AI engines leak evidence into tools you already have. Check referral traffic for chatgpt.com and perplexity.ai — those sessions mean an engine cited you and someone clicked through. Watch branded search volume for unexplained lifts, which often follow being recommended somewhere you can't see. And ask your sales team to log every "I found you through ChatGPT" — it's the cheapest attribution data you'll ever collect.
The trap here is the inverse of Tier 1: this data only shows you the wins. The deals where a competitor got recommended and you didn't produce no referral, no session, no signal. Analytics tell you AI search is working for you; they can never tell you how often it's working against you.
Tier 3: systematic tracking
At some point — usually the point where a number has to go in a board deck — you need the systematic version: a fixed prompt set covering your category's real buying questions, run across multiple engines on a schedule, with every answer stored and scored. Repeat runs per prompt, because of the randomness problem above. Competitor scoring, because "we're mentioned 40% of the time" means nothing until you know your rival is at 70%.
This is the method Treyci runs — every prompt, four engines, three runs each, monthly — but the design principles hold whoever builds it: same prompts every cycle (or trends are meaningless), multiple engines (they disagree wildly; in one category we measured, Grok mentioned brands in 81% of answers while Perplexity did in 43%), and stored verbatim answers so you can read *why* an engine said what it said.
What to do with what you find
Monitoring is only worth the effort if it feeds decisions. The three outputs that matter: the questions where you never appear (your content roadmap), the domains engines cite in your category (your PR target list), and your trend against named competitors (your proof the work is working). If your monitoring setup can't produce those three lists, upgrade the setup before you spend another hour reading answers.
Treyci tracks your brand across ChatGPT, Perplexity, Gemini, and Grok — every prompt run three times, every month, from $99.
See plansHow we measure →