Answers
How do you track AI visibility for your brand?
Tracking AI visibility means running a fixed set of buyer prompts across ChatGPT, Perplexity, Claude, Gemini, and Google AI Overviews on a repeating schedule, then recording whether your brand is mentioned, how prominently, and in what light. The output is a share-of-voice trend, not a single score.
Updated 2026-07-30
TL;DR
Key takeaways
- Fix the prompt battery before you measure anything; changing prompts destroys the trend.
- Run weekly at minimum, per engine, and store the full answer text.
- Score three things: mention, prominence, sentiment.
- Compare against a named competitor set, not against an absolute target.
- Join mention windows to CRM first-touch to report influenced pipeline.
Step 1: build a prompt battery
A prompt battery is a fixed list of 50 to 300 buyer prompts, spread across problem, category, comparison, and brand intent, that you never edit mid-quarter.
Source prompts from real buyer language: sales call transcripts, support tickets, the questions your SDRs get on cold calls, and the long-tail question keywords your category already generates. Avoid prompts that name only your brand; those flatter the number and teach you nothing.
Balance matters. A battery that is 80 percent brand-intent prompts will show 90 percent share of voice and predict nothing about pipeline. Weight toward the unbranded category and comparison prompts where the shortlist is actually formed.
- Problem intent: "how do I prove AI search drives pipeline".
- Category intent: "best AI visibility tracking platform for B2B".
- Comparison intent: "Attribufi vs Profound".
- Brand intent: "is Attribufi worth it".
Step 2: run on a schedule, across every engine
Run the full battery weekly against each engine separately, because coverage differs sharply between ChatGPT, Perplexity, Claude, Gemini, and AI Overviews.
A brand can hold 40 percent share of voice in Perplexity and near zero in ChatGPT for the same prompts, because retrieval sources and recency windows differ. Averaging engines together hides the gap that would tell you where to work next.
Store the full answer text, not just a boolean mention flag. The raw text is what lets you audit sentiment later, spot which competitor is being recommended over you, and see exactly which of your pages was cited.
Step 3: score mention, prominence, and sentiment
Score each run on three axes so a first-sentence recommendation is never counted the same as a passing footnote.
Share of voice alone is a blunt instrument. A mention in the opening recommendation drives materially different buyer behavior than a citation buried in a list of eight alternatives. Prominence weighting turns a flat count into something that correlates with pipeline.
Sentiment catches the failure mode that raw counts miss entirely: being consistently mentioned as the expensive or enterprise-only option. That reads as visibility and behaves as disqualification.
Step 4: join mentions to pipeline
Join AI visibility to revenue by matching mention windows against CRM first-touch dates and reporting sourced versus influenced pipeline in dollars.
This is the step most teams skip, and it is the one that decides whether the program keeps its budget. The defensible model is a window join: when share of voice rises for a prompt cluster, look at accounts entering the CRM in the following window whose first touch was direct or AI-referred, and report those as influenced rather than sourced.
Influenced is an honest word. It does not claim causation, it claims a documented correlation with a stated window, and that is enough for a CFO conversation when it is presented consistently quarter over quarter.
Common tracking mistakes
The most common mistakes are editing the prompt battery mid-quarter, checking manually and irregularly, and averaging engines into one number.
Manual spot checks feel productive and produce noise. Answers are regenerated per request and vary with phrasing, personalization, and recency, so a single check tells you almost nothing. Only repeated runs of a stable battery produce a signal you can act on.
- Do not change prompt wording mid-quarter; version the battery instead.
- Do not measure only branded prompts.
- Do not average engines; report them separately.
- Do not report mentions without prominence and sentiment.
FAQ
Frequently asked questions
How often should I check AI visibility?
Weekly is the practical minimum for a stable trend, and daily is useful only for high-volatility categories. Anything less frequent than weekly makes it impossible to attribute a share-of-voice change to a specific content release.
Can I track AI visibility manually?
You can for a handful of prompts, but it does not scale or stay consistent. A 100-prompt battery across five engines run weekly is 2,000 answers a month, which is why teams automate the runs and the scoring.
What is a good AI share of voice?
There is no universal target. Judge it relative to a named competitor set in the same prompt battery. Moving from third to first mention position on your comparison prompts matters more than any absolute percentage.
Does AI visibility show up in Google Analytics?
Only partially. Some assistants pass a referrer and many do not, so analytics undercounts AI-driven discovery. That gap is why mention-window joins to CRM data are used alongside analytics rather than instead of it.
Next step
See where your brand actually stands.
The free grader runs a sample prompt battery against your domain across ChatGPT, Perplexity, Claude, Gemini, and Google AI Overviews, then returns a scored report you can download as a PDF. No credit card.