How to Track Your Brand Mentions in ChatGPT, Gemini and Perplexity

A one-afternoon manual method, an API-based upgrade, and the three numbers worth trending: mention rate, citation rate and AI share of voice.

You cannot improve what you do not measure, and AI visibility is easy to leave unmeasured because it never shows up in your analytics — the visit that never happened leaves no trace. The good news: a usable tracking practice takes one afternoon to start and about an hour a week to keep running. If you are brand new to the topic, What Is Answer Engine Optimization (AEO)? sets up the vocabulary; this article assumes it.

Start with a frozen prompt set

Before any method, write down the questions. Pick 10–20 prompts shaped the way buyers actually ask: category ("best CRM for solo founders"), comparison ("X or Y for a two-person team"), problem ("how do I keep projects on track with no manager"), and brand ("what do you know about [your product]"). Freeze the set — same wording, same order — and record it in a spreadsheet with columns for engine, prompt, date, mentioned (yes/no), cited (yes/no), sentiment, position in the answer, and notes. The discipline matters more than the wording: a frozen set is what turns scattered spot-checks into a trend. Decide what you will change on your site after you have a baseline, not before.

The manual method: ask, record, repeat

The zero-tool version is exactly what it sounds like: run every prompt in every engine, by hand, and fill in the row. Start each prompt in a fresh session — a conversation that already contains context about your brand contaminates the result. For Perplexity, read the citation list and note whether your domain appears; for Gemini, check the grounded sources where shown; for ChatGPT, note whether it browsed or answered from memory. Paste the raw answer into the sheet or screenshot it, because answers drift. Run the set weekly or biweekly: caching, model updates and your own content changes all move the numbers. The manual method's limits are a small sample and judgment calls in the sentiment column — acceptable at the baseline stage, and exactly what pushes you to automation later. Engine-specific reading: The Perplexity Citation Guide and Gemini Brand Visibility.

The automated method: run the set through APIs

The upgrade reuses the identical frozen prompt set, executed by a script on a schedule — a cron job or a scheduled GitHub Action is enough. Call each provider's API, store the raw JSON response, and parse a few fields: does the brand name appear in the answer text, does one of your domains appear in the cited URLs, which competitor names appear. Run each prompt three to five times to smooth sampling variance, and pin the model version in your log — answers change when providers ship, and you want to distinguish "the model changed" from "our content worked". One caveat: even at temperature zero, browsing-grounded answers vary run to run, because the retrieval step returns different pages. You will be calling several providers with the same prompt set; this is where a multi-protocol LLM gateway earns its keep — one config routes the identical prompts to every model and upstream you test.

The metrics that matter

Three numbers cover most decisions. Mention rate is the share of prompts where your brand appears in the answer — your headline number, per engine. Citation rate is the share of prompts where a page of yours is cited as a source — it measures the retrieval half of the problem. AI share of voice is your mentions divided by all brand mentions in the set, competitors included — the number to trend monthly once the first two are stable. Formulas, sampling pitfalls and benchmarking are in Measuring Share of Voice in AI Answers.

The tool landscape, in one sentence

Dedicated AEO trackers — prompt-set monitoring, citation capture, share-of-voice dashboards — are shipping fast, DIY scripts remain fine at small scale, and we reviewed the whole field in The AEO Tool Landscape: A Roundup; our own tracker, AEOscope, is coming — join the waitlist to shape it.

Keep reading

Part of PZTools·AEOscope·PolicyForge·LaunchKit