C
Citations.io
Start free
Playbook · Learn

How to track AI visibility.

A repeatable, five-step playbook for measuring how AI engines mention your brand — and what to do with the data.

TL;DR
To track AI visibility: build a 100–500 prompt set your customers actually ask, run it daily across ChatGPT, Perplexity, Gemini and Claude, tag every brand mention and citation, roll up to a per-engine score, and act on the share-of-voice gaps versus competitors.

Step 1 — Build your prompt set

Don't guess. Your prompt set is the foundation; if it doesn't reflect what real buyers ask, your tracking is theatre. Build it from four sources:

  • Search queries — top informational and commercial queries from GSC and Semrush in your category.
  • Sales conversations — questions prospects ask reps in discovery calls.
  • Support tickets — what existing users ask in chat and email.
  • Competitor prompts — "best X", "X alternatives", "X vs Y" formulations.

Aim for 100–500 prompts across category, comparison, problem and product intent.

Step 2 — Choose your engines

Track the four majors at a minimum:

Step 3 — Run prompts daily and capture verbatim answers

Two non-negotiables: daily cadence and verbatim capture. Daily, because answers shift constantly. Verbatim, because aggregate counts mean nothing without the underlying text — you need to inspect mentions in context for sentiment, accuracy and competitor framing.

Step 4 — Tag mentions, citations, rank, sentiment

For each captured answer, record:

  • Whether your brand is mentioned (yes/no).
  • Position in any listed group of brands.
  • Which URLs were cited as sources.
  • Sentiment of the mention (positive / neutral / negative).
  • Which competitors were mentioned alongside you.

This produces your raw AI visibility score inputs.

Step 5 — Act on the gaps

The metric isn't the goal — moving it is. Each week, review:

  • Prompts where competitors win and you don't. Identify which sources the AI cited; pursue mentions on those sources.
  • Negative-sentiment mentions and hallucinations. Correct the record by publishing canonical content on your own domain.
  • Citation drops. Pages you used to be cited from that you no longer are — usually means a competitor displaced you on the source.

Tooling vs. doing it manually

For a 10-prompt audit on one engine, a spreadsheet works. For ongoing monitoring across hundreds of prompts and four engines daily, you need a platform. Compare the major AI visibility tools or run a free check below.

Frequently asked questions

Can I track AI visibility manually?+
For 5–10 prompts on one engine, yes — open ChatGPT, run each, log the answer. Beyond that you need tooling. AI answers are stochastic (different each run) and you need daily multi-engine coverage to spot trends, which manual tracking can't deliver at scale.
How many prompts should I track?+
Most teams start with 50–200 high-intent prompts covering category, comparison, problem and product queries. Enterprise programs track 1,000–5,000 across markets and personas. Quality of the prompt set matters more than raw count.
How often should AI visibility be tracked?+
Daily for active monitoring; weekly is the minimum useful cadence. AI engines update their retrieval and source pools frequently, and a single news cycle can shift answers within hours.
Do I need to track every AI engine?+
Track the four majors at minimum: ChatGPT, Perplexity, Gemini / Google AI Overviews, and Claude. They each behave differently and your visibility on one is not a proxy for the others.
Run a free AI visibility check →