---
title: AI visibility tracking: what to measure and how, without guesswork
description: AI visibility tracking explained: the metrics that matter, five data sources and their blind spots, how to pick prompts, and when a paid tracker is worth it.
canonical: https://www.postbing.com/blog/ai-visibility-tracking-guide
published: 2026-11-16
updated: 2026-11-16
author: The Postbing Team
---

# AI visibility tracking: what to measure and how, without guesswork

AI visibility tracking measures how often AI engines name your brand, cite your pages and name competitors when asked your buyers' questions. No single tool sees all of it. Combine a fixed prompt sample run every week, AI referral sessions in GA4, Search Console's AI feature data, and a "how did you hear about us" question, then report trends by engine rather than a single score.

Last updated: November 16, 2026

## What AI visibility tracking can and cannot tell you

It tells you your share of a sample of answers, not your share of all answers. There is no public search volume for prompts, so every tracker, paid or manual, runs prompts that someone chose. If the prompts are wrong, the chart is precise and meaningless. Measurement is one part of the wider [GEO playbook](/blog/generative-engine-optimization-guide).

It also cannot see personalised answers. Logged-in users get answers shaped by their history and settings, and engines rotate sources between runs. Treat every number as an estimate with wide error bars, and look for direction over months.

## The metrics that matter

Track a handful of rates per engine, per prompt, per week. Each one answers a different question:

| Metric | Definition | Answers |
|---|---|---|
| Mention rate | Runs where your brand is named ÷ total runs | Are we in the conversation? |
| Recommendation rate | Runs where you are the pick for the stated need ÷ total runs | Do engines choose us? |
| Citation rate | Runs where one of your URLs is a source ÷ total runs | Do our pages get used? |
| Cited pages | Which of your URLs appear, and how often | Which pages work? |
| Competitor share | Runs where each competitor is named ÷ total runs | Who is winning the answer? |
| Third-party sources | Non-brand URLs cited in your prompts | Where should we get mentioned? |

Mention and citation rates often move separately. You can be named from a review site without being cited, or cited for a definition without being recommended. Report both.

## Five data sources and their blind spots

Each data source sees a different part of AI visibility. Use at least three.

| Source | What it measures | Blind spot | Cost |
|---|---|---|---|
| Manual prompt log (logged out, 3 runs per prompt) | What a real logged-out person sees, with sources | Small sample, takes about 20 minutes a week | Free |
| API-based tracker | Same prompts at scale, consistent schedule | API answers can cite different sources than the consumer app | Paid, or included in some tools |
| GA4 AI channel | Sessions referred by AI apps | Visits with no referrer land in Direct; no data on unclicked mentions | Free |
| Search Console | Google AI feature traffic within the Web search type; Google also names a Generative AI performance report | Google only; no ChatGPT or Perplexity data | Free |
| Signup question | Buyers who say AI sent them | Memory is fuzzy; only counts converters | Free |

The manual method is written up step by step in [track AI citations for free](/blog/track-ai-citations-for-free). The GA4 setup, including a regex that works with GA4's full match rule, is in [AI search traffic in GA4](/blog/ai-search-traffic-in-ga4).

## How to choose the prompts you track

Pick 10 to 30 prompts that a real buyer would type, weighted toward money questions. A good set mixes:

- **Category prompts:** "best accounting software for a small bakery"
- **Comparison prompts:** "is [competitor] or [competitor] better for a team of three"
- **Alternative prompts:** "cheaper alternative to [competitor]"
- **Constraint prompts:** "[category] with no monthly fee"
- **Brand prompts:** "is [your brand] any good", "[your brand] pricing"

Write them the way people talk, with context and a constraint, not as keywords. Then freeze the set for at least a quarter. Changing prompts every week makes trends impossible to read. Add new prompts as a separate group instead of replacing old ones.

Keep the same prompts in every engine you track, and record which engine and mode you used (for example ChatGPT logged out with search, or Google AI Mode).

## How often to track, and how many runs

Weekly is enough for most sites, with three runs per prompt per engine. One run is noise, because the same question returns different sources. Ten prompts, three runs and three engines is 90 answers a week, which is about the limit of what one person will keep logging by hand week after week.

If you use a tool, more runs reduce noise but do not fix a bad prompt set. Look at monthly averages: a move from 3 named runs in 30 to 9 in 30, held for two months, is a signal. A jump in one week is not.

## A monthly AI visibility report on one page

Keep it short enough to read in two minutes:

1. **Mention and citation rate by engine,** this month vs last month vs three months ago.
2. **Top three competitors' mention rates** on the same prompts.
3. **Top cited third-party URLs** in your prompts, with whether you appear on each.
4. **AI referral sessions** from GA4 and any signup answers naming an AI engine.
5. **One action** for next month, tied to a specific gap (for example "get reviewed on the two list pages cited in 40% of runs").

Do not report a combined "AI visibility score" without showing its parts. Blended scores hide which engine moved and why.

## When a paid tracker is worth it

A paid or bundled tracker is worth it once you track more than about 30 prompts, several sites or several markets, because manual logging stops scaling. It is not worth it before you know which prompts matter. Start manual for a month, then automate the prompt set you trust. Our roundup of [AI SEO tools](/blog/best-ai-seo-tools) covers options.

When you evaluate one, ask four questions: does it use the API or the consumer app, how many runs per prompt, does it store the cited URLs, and can you export raw answers. Google's AI optimization guide also warns to "be wary of third-party tools that promise ranking success or claim to use 'internal' Google metrics".

## Where Postbing's tracker fits

Postbing includes AI visibility tracking in every plan: each week it asks your site's buyer questions through the AI providers' APIs (ChatGPT, Claude, Gemini and Perplexity are supported, depending on configuration) and records brand mentions, domain citations, positions, which of your articles were cited, and competitor mentions. Question limits run from 5 per site per week on the free trial and BYOK Lite to 40 on Agency Pro, as of November 2026 ([pricing](/pricing)).

Its limits: it is API-based, so pair it with a monthly logged-out manual check, and it does not track Google AI Overviews as a separate surface. It does not guarantee any change in visibility.

## FAQ

### What is a good AI mention rate?
There is no published benchmark we could verify, and rates depend on your prompts and category. Compare yourself with your top competitors on the same prompts and track your own trend over months.

### Can Google Analytics track AI visibility?
Only the clicks. GA4 can group sessions from AI apps into a custom channel, but it cannot see answers that mention you without a click, and visits without a referrer land in Direct.

### Why do paid trackers and my manual checks disagree?
Most trackers query APIs, and API answers can cite different sources than the consumer app a logged-out person uses. Sampling also differs: number of runs, time of day and prompt wording all change results.

### How many prompts should I track?
Start with 10 that a real buyer would type, run three times each per engine weekly. Add more only once you have a month of stable data.

## Sources

- [Google: Optimizing for generative AI search](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide): Generative AI performance report, warning about third-party tools, accessed 16 November 2026
- [Google: AI features and your website](https://developers.google.com/search/docs/appearance/ai-features): AI feature traffic in the Web search type, accessed 16 November 2026
- [Google Analytics Help: Custom channel groups](https://support.google.com/analytics/answer/13051316): AI assistants channel example, accessed 16 November 2026
- Nicholas Dulait, "The KING of GEO" post on X (13 November 2026): logged-out runs, three runs per prompt, API vs app caveat, signup question, accessed 16 November 2026

## Related reading

- [Generative engine optimization: the practical guide](/blog/generative-engine-optimization-guide)
- [Track AI citations for free with a weekly log](/blog/track-ai-citations-for-free)
- [AI search traffic in GA4](/blog/ai-search-traffic-in-ga4)
- [Postbing for agencies](/agencies)
