AI visibility tools track what AI assistants - ChatGPT, Perplexity, Gemini, Copilot, and AI Overviews in Google Search - say about your brand: whether you are mentioned, cited or recommended when buyers ask questions in your category, and how that compares to competitors over time. The need exists because assistant answers are dark by default: millions of buying conversations happen inside engines that send you no referrer, no impression data and no alert when a rival becomes the default recommendation. AI visibility tracking turns that darkness into a measurable surface. This page covers the methodology that makes the numbers trustworthy, the capabilities to demand from any tool, and the mistakes that make LLM visibility data worse than no data.
What AI visibility tracking is
Mechanically, every serious tool does the same thing: it maintains a set of questions your buyers plausibly ask, puts them to the live engines on a schedule, records the full answers, and extracts the signals - brand mentions, page citations, recommendation position, sentiment, and the competitors appearing alongside you. Run on a cadence, that produces the two numbers a program is managed by: your citation share (of the answers composed for your question set, what fraction cite or recommend you) and its trend against each named competitor. Everything else a dashboard shows - sentiment deltas, ChatGPT brand mentions by week, engine-by-engine breakdowns - is decoration on that core. The practice this measurement serves is answer engine optimization; the tracking is its instrument panel.
The methodology that matters
Four disciplines separate real measurement from screenshots. A fixed question set: the same questions, phrased the same way, re-asked every cycle - change the questions and you have a new experiment rather than a trend. Per-engine tracking: ChatGPT, Perplexity, Gemini and AI Overviews retrieve differently and cite differently, so a blended "AI score" hides exactly the variation you need to act on. Archived answers: the raw text of every response, stored - because assistants answer probabilistically, because stakeholders will ask "says who?", and because the archive is what turns "we think we improved" into evidence. Citation share as the headline: a percentage against named competitors, benchmarked from a recorded baseline. The full metric definition and its edge cases live in our citation share reference, and the end-to-end process in the AI visibility audit playbook.
The capability checklist
| Capability | Why it matters |
|---|---|
| Fixed, editable question sets | Trends require constant inputs |
| Multi-engine coverage, reported separately | Engines disagree; the disagreement is the insight |
| Raw answer archive | Every score auditable back to text |
| Mentions and citations tracked as distinct signals | Brand recommendation and page citation are different assets |
| Named-competitor benchmarking | Share is only meaningful against someone |
| Scheduled runs with repeat sampling | Probabilistic answers need repeated draws |
| A route from finding to action | Export, brief or executed fix - the loop must close |
The mention-versus-citation row deserves emphasis because it is the most commonly collapsed distinction in LLM visibility reporting. An assistant can recommend your product by name while citing a third-party review, or cite your docs inside an answer that recommends a competitor. The two signals have different causes - how assistants choose sources unpacks the mechanics - and different fixes, so a tool that merges them into one number has thrown away the actionable half of the data.
Common measurement mistakes
Two subtler failures round out the list. Treating one answer as truth: the same engine asked the same question twice can cite differently, so anything worth reporting needs repeated sampling before it is a finding. And measuring only yourself: visibility is a share, and a rising mention count during a quarter when your competitor's rose faster is a loss dressed as a win. The corrective for all five mistakes is the same boring discipline - fixed sets, all engines, archives, baselines, competitors - which is why the methodology section above is the real buying guide.
Where AstroFabric fits
AstroFabric's AI Visibility agent runs exactly the loop this page describes - fixed question sets per engine, archived answers, citation share against a baseline - as one of eight agents on a platform for growth, revenue and digital operations. The difference in shape: findings become missions for sibling agents, so the citation gap turns into drafted content and the retrievability failure turns into a technical fix on the same system that measured it. The fair comparison: dedicated analytics products - the specialists cataloged in our GEO tools landscape - go deeper on pure measurement, with prompt-volume research and enterprise reporting a focused product can afford to build. Teams with a staffed visibility program often want that depth; teams where the people reading the dashboard are also the people doing the work usually want the loop closed by the platform itself.
Frequently asked questions
What are AI visibility tools?
Software that tracks what AI assistants - ChatGPT, Perplexity, Gemini, Copilot and Google’s AI Overviews - say about your brand: putting a fixed set of buyer questions to the engines on a schedule, archiving the answers, and measuring mentions, citations and recommendations against competitors over time.
How do I track ChatGPT brand mentions?
Define the questions your buyers actually ask, put them to ChatGPT on a repeating schedule, archive the full answers, and count mentions and citations against a recorded baseline - with repeat sampling, because answers vary between runs. Tools automate the cadence; the methodology is what makes the number real.
What is the difference between a brand mention and a citation?
A mention is the assistant naming your brand in an answer; a citation is the assistant sourcing your pages. They move independently - you can be recommended via third-party coverage without a single citation, or cited in answers that recommend rivals - and they have different fixes, so track both separately.
What is LLM visibility?
The umbrella term for how visible a brand is in large language model outputs - the same territory as AI visibility tracking: share of answers that mention, cite or recommend you across engines, measured on a fixed question set with an archived trail.
What are the most common AI visibility measurement mistakes?
Cherry-picked prompts that flatter, no recorded baseline, single-engine data generalized to all of AI, single answers treated as truth despite probabilistic variance, and tracking your own numbers without competitor share. Fixed question sets, multi-engine coverage, archives and baselines prevent all five.
Sources
- Google Search Central - AI features and your website
- GEO: Generative Engine Optimization (the original research paper)
- llms.txt - the proposed standard for LLM crawler guidance
Every playbook on this blog ships as a runnable mission.
Open a workspace and the playbook library is waiting - describe the outcome and the agents carry it end to end, on your plan's monthly credits.