// Answer
How do I track when AI engines cite my brand?
Not by Googling. AI answers are sessioned, personalized, and ephemeral — you need a clean-room pipeline that runs your prompt set against every engine on a fixed cadence and aggregates what it finds.
// Why it’s hard
Why you can’t just Google it.
AI engines don’t behave like a SERP. The answer you see when you ask is shaped by your session, your location, your prior conversation, your account’s personalization, and the random sampling that makes generation interesting. Two people can ask the same question and get different brands cited. The same person can ask twice in the same hour and get different answers. Search Console doesn’t see any of this. Ahrefs doesn’t see any of this. The surface is real, the impact on demand is real, and the only way to measure it is to build — or buy — a pipeline that re-runs your prompts under controlled conditions and aggregates over enough samples to be meaningful.
// Manual
Manual tracking and why it doesn’t scale.
The first move every team makes is logging into ChatGPT once a week and typing “best CRM for early-stage startups.” This will work for a month. It will not work at scale. You can’t cover six engines, hundreds of buyer-intent prompts, three languages, and weekly cadence by hand. You can’t reproduce results. You can’t aggregate. And you can’t catch a competitor pulling ahead in week three because you only checked twelve of your three hundred prompts. Manual is fine for spot-checks. It is not a measurement program.
// What to measure
What a real tracking pipeline measures.
- Share of voice. The percentage of tracked prompts in your category where an engine names or cites you. The single most useful headline number.
- Citation count and rank. When you’re cited, where in the answer? First citation versus seventh matters more than it should.
- Sentiment toward your brand. Being named neutrally is fine. Being named as the cautionary tale is not. A real tracker labels the polarity.
- Sources cited in your place. The most actionable number nobody else gives you: which competitors and which third-party sites win the citations you should be winning?
- Cross-engine deltas. Where Perplexity loves you but ChatGPT doesn’t, you have a known asymmetry to investigate.
- Time series. The point of measurement is movement. A snapshot is decoration; a trend line is operations.
// Buying checklist
What to look for in a tracking tool.
- Engine coverage. ChatGPT, Perplexity, Google AI Overviews, Bing Copilot, Claude, Gemini. If a tool covers three of those, you’ll learn a third of the answer.
- Prompt density. Tracking thirty prompts is theater. Tracking hundreds across the buyer journey is operating.
- Citation graph visibility. The tool should tell you not just whether you were cited but who was cited instead.
- Accuracy validation. Does the tool periodically verify its extractions against the actual engine output? AI engines change formats. Trackers that don’t validate drift.
- Cadence. Daily at minimum on high-stakes prompts. Weekly is the floor for the long tail.
// Vizelo
Vizelo’s approach.
Vizelo runs your category prompt set against every major answer engine on an hourly cadence, normalizes the citation data, and surfaces share of voice, the sources cited in your place, and the deltas that matter week over week. It’s GEO measurement built to operate from rather than report on. Start free to see your current footprint in under a minute.
FAQ
Common questions.
Can Google Search Console show me ChatGPT citations?
No. Search Console reports on Google Search and Google Discover. It has nothing to say about ChatGPT, Perplexity, Claude, or any non-Google engine. Some Search Console reporting on AI Overviews exists but it’s limited to the Google surface.
Why doesn’t manual checking work?
Answers are personalized to your session, location, and prior conversation. Two people asking the same question get different answers. You can’t generalize from a single laptop. You need a clean-room prompt run that’s reproducible, repeated, and aggregated across hundreds or thousands of prompts.
What’s the difference between a citation and a mention?
A citation is a linked source under the answer (most explicit on Perplexity and AI Overviews). A mention is your brand named inside the answer text, with or without a link. Both matter. A good tracker reports both and labels them clearly.
How many prompts should I be tracking?
Enough to be statistically meaningful and to cover the buyer journey: category-defining queries, comparison queries, jobs-to-be-done queries, and brand queries. For most B2B categories that lands between 100 and 1,000 prompts per language.
How often should the prompt set re-run?
Daily at minimum for high-stakes prompts, weekly for the long tail. AI engines change fast enough that monthly cadence is too slow to catch movement, and quarterly is decorative.