How to measure and increase brand visibility in ChatGPT, Perplexity, and AI Overviews

Measuring AI visibility means running a fixed set of questions in ChatGPT, Perplexity, and Gemini, recording whether your brand shows up, in what position in the answer, with or without a link, and repeating that consistently over time. This is quite different from looking at ranking position, and you can set up this routine in an afternoon, with no automation at all, just a spreadsheet.

This article shows the manual method first, because it is what any team can run without a tool budget, and then where automation actually pays off.

In a Google search, the same keyword tends to return the same top 10 for similar users. In an LLM, two people asking the same question at different times can get answers with different brands, because the model generates the answer on every call instead of querying a fixed index. That changes the measurement logic: instead of "I'm in position 3," the question becomes "in how many runs of the same question do I show up."

This instability also shows up as a recurring complaint among SEO professionals: the same brand can be well cited in ChatGPT and practically invisible in Gemini, with the difference explained by which sources each model prioritizes, not by site quality.

1. Mention rate. The percentage of prompt runs in which the brand appears, cited or not. Simple formula: answers that mention the brand divided by the total answers tested, times 100 (tropk.ai).
2. Citation rate with link. Of the mentions, how many come with an actual clickable link to your domain. Being mentioned by name with no link is common, and it is worth less than a citation with a source.
3. Competitive share of voice. In the same answers, which competitors show up and how often relative to you.
4. Position within the answer. Whether the brand appears in the first lines or only at the end of a long list.
5. Sentiment and framing. How the AI describes the brand when it mentions it (positive, neutral, as a "cheaper" alternative, and so on).
6. Drift. How much the answer changes between identical runs on the same day and across different days.

In practice: if you can only measure one thing, measure mention rate with a linked citation, broken down by platform. It is the metric closest to "generated a real opportunity for traffic or recognition."

1. Define 20 to 40 prompts that represent how a customer would search, split into categories: discovery ("best X tools"), comparison ("X or Y"), direct recommendation ("which X do you recommend for Z"), and niche-specific technical questions.
2. Run the same prompts in ChatGPT, Perplexity, and Gemini, always in a new session (no conversation history), so previous context does not bias the answer.
3. Log it in a spreadsheet: prompt, platform, whether the brand appeared, whether it came with a link, position in the answer, competitors cited, date.
4. Repeat weekly or biweekly, never daily: the natural variance between runs is too high for a daily cycle to produce a useful signal, and it just becomes noise.
5. After 4 to 6 cycles, compare mention rate over time and cross it against changes you made on the site.

Mistakes that invalidate this process: testing only prompts that already include your brand name (that measures recognition, not discovery), mixing language and region without reason, and swapping the prompt list every round, which prevents any historical comparison.

A few numbers help calibrate expectations before running your own test:

• Across 680 million citations analyzed, Wikipedia appears in 7.8% of ChatGPT citations (Profound.io, via tropk.ai), which shows that broad reference sources have a structural advantage over niche sites.
• An SE Ranking analysis of 63,987 sites points to ChatGPT as responsible for 78% of global referral traffic coming from AI (SE Ranking, via tropk.ai), meaning that if you can only monitor one platform, this is the priority.
• The correlation between web mentions (press, forums, other sites talking about the brand) and AI citation was measured at 0.664 by Ahrefs (via tropk.ai), reinforcing that PR and third-party mentions carry as much weight as the site's own content.

Correlation is not causation: these numbers show a market pattern, not a guarantee that replicating the tactic produces the same result in your niche.

Two real patterns worth checking before concluding "the AI doesn't know me":

• Citation without a brand name. Sometimes the model uses information from one of your pages to compose the answer, but names only the generic domain, or not even that, leaving the brand out of the visible text. Looking only at "was I mentioned by name" underestimates your real reach.
• Comparison and alternatives pages (/vs and /alternatives) tend to get cited more often than generic "best X" content, according to people who track this closely in SEO forums. It is worth having this type of page if your product gets compared often.

// real case: on the vast majority of sites I track, ChatGPT is the AI that brings the most traffic. Measuring the other AIs matters, but knowing how to split effort between them matters too.

The manual spreadsheet gets tiring after a few dozen prompts and three platforms. AI visibility tools solve scale, but two checks are worth doing before signing up: ask to see a raw answer (not just the aggregate score), and ask how the tool handles the natural variance between runs, because part of the market's justified skepticism comes from expensive tools that do not show a clear methodology behind the number they display.

Once you have the citation metric in place, it is also worth looking at the reflection in Search Console: check the article on AI Overviews and the CTR drop to separate AI visibility from traditional organic traffic.

A good start is a list of around 20 real prompts from your niche and a first round in ChatGPT and Perplexity, before paying for any tool. With that first snapshot, it gets much easier to decide what is worth automating. This article is part of the pillar GEO in practice: how to get crawled, understood, and cited by AI search engines.