What AI visibility is, and how to measure it without fake precision

AI visibility is how often AI assistants such as ChatGPT, Gemini and Perplexity mention or cite a brand when people ask them questions the brand should be relevant to. It's measured by asking a fixed set of realistic buyer questions repeatedly and counting, across all the answers, how often the brand appears — alongside its competitors.

It's the measurement side of GEO — see GEO vs SEO vs AEO for how the disciplines relate. The important word is often. AI answers aren't a ranked list that stays put. They're generated fresh each time, and they vary. So AI visibility is a rate, observed over many answers, not a position.

Why "rank in ChatGPT" is the wrong unit

In SEO, a page ranks third for a keyword and, give or take personalisation, stays third until something changes. AI answers don't work that way. In SparkToro and Gumshoe's study of nearly 3,000 responses, ChatGPT and Google's AI gave the same list of brands in less than one run in a hundred, and the same order roughly once in a thousand. The authors' conclusion: visibility percentage across many prompts and runs is a reasonable metric; ranking position is not.

Any tool or report that says "you're #3 in ChatGPT" is describing one answer. The next answer might say something else. That's why the metrics below are all rates over a sample. Why ChatGPT gives different answers every time explains the mechanics.

The six metrics worth reporting

Mention rate

The headline number. Always show it with its counts: "mentioned in 71 of 240 answers (30%)". The counts tell the reader how much evidence sits behind the percentage, and stop small changes being over-read.

Citation rate

Engines that search the web cite sources. If the client's own website is among them, the engine is using the client's content to form its answer. Mentions and citations move independently — a brand can be recommended on the strength of review sites without its own site ever being cited, and vice versa — so track both.

Share of voice

The client's share of all mentions among the brands you track. It turns visibility into competitive position. The formula, and the counting rules that make it trustworthy, are in how to calculate AI share of voice.

Prompt coverage

Group prompts by how consistently the brand appears in their answers. One simple scheme:

• Visible: mentioned in at least half the answers to that prompt.
• Weak: mentioned sometimes, but less than half.
• Absent: never mentioned in the period.

"Absent while competitors appear" is the most actionable list in any AI visibility report.

Engine split

ChatGPT, Gemini and Perplexity search and cite differently. Always report each engine as well as the total; see ChatGPT vs Gemini vs Perplexity.

Cited sources

The domains engines cite when answering the client's questions — review sites, directories, media, competitors' pages. This is where the work is: you can't edit ChatGPT, but you can influence the sources it reads.

A worked example

Take a tiny log: three prompts, each asked once a week for four weeks on one engine — 12 answers. The client is Northwind; the tracked competitors are Veloz and Cyclo Co. (fictional).

From those 12 answers:

• Mention rate: 7 of 12 answers mention Northwind — 58%.
• Citation rate: 5 of 12 answers cite Northwind's website — 42%.
• Share of voice: Northwind 7, Veloz 6, Cyclo Co. 5 tracked mentions — Northwind 7 ÷ 18 = 39%.
• Prompt coverage: visible on two prompts (3 of 4 and 4 of 4), absent on one — while both competitors appear there.

The averages look healthy; the coverage line shows the problem. That's why the six metrics are reported together, and why a real prompt set has about 20 questions on several engines rather than three on one.

What to be careful with

• Single "visibility scores". A score from 0 to 100 that blends mentions, position and sentiment is easy to chart and hard to explain. If you use one, make sure you can show the counts underneath.
• Position. Being named first is nice; it's also the least stable property of an AI answer.
• Sentiment. Useful when a brand is described negatively, but automated sentiment on short mentions is noisy. Read the answers before you report it.
• Small samples. Ten answers is an anecdote. See how often to track AI visibility for how many answers you need before calling a change real.
• API vs app. Tools either call the AI providers' APIs or collect answers from the consumer apps. API answers can differ from what one person sees in the app. Neither reproduces one person's screen; both can measure patterns.

What a measurement set-up looks like

1. A fixed prompt set per client — about 20 realistic buyer questions across intents. How to choose them.
2. 3–5 tracked competitors, with aliases.
3. A fixed schedule — every prompt on every engine, weekly or daily.
4. Every answer stored, with mentions and citations detected the same way each time.
5. Monthly rates with counts, per engine, compared with the previous month.
6. Evidence on demand — any number can be opened to the answers that produced it.

That's the method Ranxii automates: about 20 prompts and up to 5 competitors per client, asked on ChatGPT, Gemini and Perplexity every week — roughly 240 answers per client per month — with every metric shown as "X of Y answers" and linked to the stored text.

Connecting visibility to traffic

Visibility is upstream; clicks are downstream. OpenAI says ChatGPT adds utm_source=chatgpt.com to links in its answers, so its referrals can be identified in analytics. Reporting both — how often the client is recommended, and how many visits those recommendations send — answers the inevitable "so what?" Our AI visibility report template shows how to put it together.

Sources

• SparkToro: AIs are highly inconsistent when recommending brands or products (January 2026)
• OpenAI Help Center: Publishers and developers FAQ