Short answer
Measure AI search visibility on a fixed set of prompts your buyers ask, run several times a month in each AI engine. Record whether your brand is named, whether your pages are cited and which competitors appear instead, then report share of voice and citations by engine. Add the visits and leads the assistants send, from Google Analytics, Search Console and Bing’s AI Performance report. Single answers vary too much to read on their own.
Key takeaways
- AI answers change between runs: in SparkToro’s 2026 test, ChatGPT and Google’s AI repeated the same list of brands less than once in 100 runs, so visibility is a percentage across many runs.
- Measure inside the answers (mentions, citations and share of voice by engine) and after them (AI referral visits, leads and branded searches).
- Google counts AI Overviews and AI Mode in Search Console’s Web results and in Google Analytics’ Organic Search, and Google Analytics’ AI Assistant channel covers ChatGPT, Gemini, Copilot and others.
- Bing Webmaster Tools reports the citations your pages earn in Copilot and Bing’s AI summaries, with the grounding queries behind them.
- A fixed prompt set, run monthly against named competitors, turns noisy answers into a trend you can act on.
What AI search visibility is
AI search visibility is how often AI assistants name and cite your brand when people ask them the questions your buyers ask, measured against your competitors. It is also called AI visibility or LLM visibility, and it has two layers. Inside the answers: whether the brand is mentioned, whether its pages are cited as sources, and how its share compares with the other brands named. After the answers: the visits, leads and branded searches those answers produce.
Both layers count. A mention shapes how a buyer sees a category even when nobody clicks, and a click is the part analytics can follow to a sale. Visibility is the scoreboard of answer engine optimization (AEO): the work I describe in what is answer engine optimization (AEO) is judged on these numbers.
Why AI answers need their own measurement
Three things make AI answers harder to measure than search rankings.
They change from run to run. In SparkToro’s January 2026 study, 600 volunteers ran 12 prompts through ChatGPT, Claude and Google’s AI results 2,961 times. ChatGPT and Google’s AI returned the same list of brands less than once in 100 runs, and the same list in the same order about once in 1,000 [1]. Rand Fishkin’s conclusion is that visibility as a percentage, across dozens to hundreds of prompts each run several times, is a reasonable metric, and a “ranking position” inside an AI answer is meaningless [1].
Mentions and citations tell different stories. Answers built from what a model already knows name brands without linking to them, and in Ahrefs’ study of 66.6 million ChatGPT answers from August and September 2026, only 14.4% contained fan-out queries, the searches that bring back pages to cite [7].
Analytics sees only the clicks, and it files many of them with search. Google counts visits from AI Overviews and AI Mode in Search Console’s Web search results [2] and in Google Analytics’ Organic Search channel [3], next to the classic results.
The metrics that matter
I report seven numbers, the first four from the answers and the last three from analytics:
| Metric | What it measures | Where it comes from |
|---|---|---|
| Mention rate | The share of tracked answers that name your brand | Prompt tracking, by engine |
| Citation rate | The share of tracked answers that link to one of your pages | Prompt tracking, and Bing’s AI Performance report for Copilot [4] |
| Share of voice | Your mentions as a share of all brand mentions in the tracked answers, yours and your competitors’ together | Prompt tracking against a fixed competitor set |
| Fan-out coverage | The share of the searches behind each priority prompt that a page on your site answers | Fan-out queries from Gemini’s API or a tracking tool |
| AI referral sessions | The visits the assistants send | Google Analytics’ AI Assistant channel [3] |
| AI-assisted conversions | The leads and sales those visits produce | Conversion tracking that keeps AI sources apart |
| Branded demand | Branded searches and direct visits that follow AI mentions | Search Console and analytics [6] |
Share of voice is the headline number, because it accounts for the competition: a 30% mention rate means one thing when the leader sits at 25% and another when the leader sits at 70%. On one program for a leading news portal, share of voice in AI answers rose from 9% to 41% against its top five competitors.
How to build the prompt set
The prompt set decides what every number means, so I build it before anything is measured:
- Start from the questions buyers ask before they buy, in their own words: category questions, comparisons, the problems the product solves and questions about the brand itself.
- Add the fan-out searches behind the priority prompts, the narrower searches an assistant runs to answer them, as I explain in how query fan-out works.
- Group the prompts by intent and buying stage, so a fall on comparison prompts reads differently from a fall on informational ones.
- Fix the competitor set: the brands that appear most often in the answers, plus the ones your sales team meets.
- Run every prompt several times in each engine every month (Google AI Overviews and AI Mode, ChatGPT, Gemini, Perplexity and Copilot), and record each answer.
- Keep the set fixed for the quarter, so monthly changes reflect your work, then refresh it as the category moves.
The set needs dozens of prompts at least, and a broad category a few hundred, to give percentages that hold from one month to the next [1]. I build and run the sets with tools I build, in production since 2022.
Where the data comes from
Five sources cover the two layers:
- Prompt tracking: mention rate, citations and share of voice per engine, from running the prompt set, with your own tooling or a tracking tool.
- Bing Webmaster Tools: its AI Performance report shows the citations your pages earn in Microsoft Copilot, Bing’s AI summaries and selected partner integrations, with the grounding queries the AI used to retrieve them [4]. Since June 2026 a preview adds intents, topics and citation share, your share of all citations shown for a grounding query [5].
- Google Search Console: the Generative AI performance report shows the impressions your pages earn in AI Overviews and AI Mode, by page, country and device [8], and the Performance report’s Web search type counts those impressions and their clicks in the same totals as the classic results [2].
- Google Analytics: the AI Assistant channel groups visits from assistants such as ChatGPT, Gemini, Deepseek, Copilot and Grok, assigned when the referrer matches Google’s list of assistants, while visits from AI Overviews and AI Mode count as Organic Search [3].
- Server and CDN logs: requests from OAI-SearchBot, PerplexityBot and Googlebot show whether the engines can fetch the pages you want cited.
Google Analytics labels a visit as AI Assistant only when its referrer matches the list of assistants [3]. A visit that arrives with no referrer, such as a link opened from some apps, counts as Direct, so the AI Assistant channel shows the minimum the assistants send.
From visibility to leads and revenue
Visibility is the leading number, and revenue is the one a business runs on. I connect them with conversion tracking that keeps every source apart: source-segmented, cookieless and server-side, across organic, paid, referral and AI search traffic, with the data in Google Analytics and Looker Studio. That shows which assistant sent a visit, which page it landed on and whether it became a lead or a sale.
Count the effect that comes without a click too. Similarweb’s panel data, reported by SparkToro in June 2026, found that brands recommended in AI answers gained more direct visits and branded searches than brands the answers left out [6]. Branded impressions in Search Console and direct visits in analytics belong in the same report.
On one program for an e-commerce brand, citations in AI answers for its buying guides drove a 38% rise in AI-referred sessions and a 19% rise in assisted conversions in 60 days.
A worked example: one month of AI visibility data
The company and its numbers here are illustrative. A project management software company tracks 60 prompts across five engines, three runs each, for 900 answers a month, against four competitors.
This month the brand is named in 162 answers, a mention rate of 18%, and the leading competitor in 41%. The brand’s pages are cited in 54 answers, most of them on Perplexity and Copilot. The tracked answers contain 1,080 brand mentions across all five brands, so the brand’s share of voice is 15%. Bing’s AI Performance report lists 230 citations across 14 of its pages, and Google Analytics shows 410 sessions in the AI Assistant channel and 9 demo requests from them.
The gaps become the content strategy for next month. The brand is missing from every answer to its 12 comparison prompts, where the leader’s comparison pages are cited. It has no page for three of the fan-out searches behind its best prompt, a hole in its topical authority. And two engines still describe it by a product name it retired, an entity gap fixed on its own pages and profiles. The plan is two comparison pages, three pages for the missing searches, the corrected name everywhere, and outreach to the two review sites the engines cite most. The same 60 prompts run again next month, and the report shows the change engine by engine.
I run this as a monthly service, AI visibility tracking, on its own or inside a program.
Frequently asked questions
What is a good AI share of voice?
It depends on the category and on the competitors in your prompt set. Read it against the leader in your own set and against your own trend: a rising share on a fixed set of prompts is the result that counts.
Does Google Search Console show AI Overviews traffic?
Yes. Its Generative AI performance report shows your impressions in AI Overviews and AI Mode by page, country and device, and the Performance report’s Web search type counts them, with their clicks, together with the classic results.
How many prompts should I track?
Dozens at minimum, and a few hundred for a broad category, each run several times per engine. A small set makes month-to-month swings look bigger than they are, because single answers vary so much.
How often should AI visibility be measured?
Monthly, on the same prompt set, so the changes reflect your work. Run an extra check after a launch or a model update, and refresh the prompt set each quarter.
Can I track my ranking position in ChatGPT?
Position inside an AI answer changes from run to run, so it tells you little. Track the share of answers that name you and the share that cite your pages, across many runs.
Sources
- SparkToro, Rand Fishkin, “NEW Research: AIs are highly inconsistent when recommending brands or products; marketers should take care when tracking AI visibility”, 28 January 2026. sparktoro.com
- Google Search Central, “AI features and your website”, last updated 10 December 2025. developers.google.com
- Google Analytics Help, “[GA4] Default channel group”, read 28 October 2026. support.google.com
- Microsoft Bing Webmaster Blog, “Introducing AI Performance in Bing Webmaster Tools Public Preview”, 10 February 2026. blogs.bing.com
- Microsoft Bing Blog, “New AI Visibility Insights in Bing Webmaster Tools: Intents, Topics, Citation Share, Compare”, 16 June 2026. blogs.bing.com
- SparkToro, Rand Fishkin, “New Research from Similarweb: How AI Brand Mentions Influence Direct Visits & Traditional Search Queries”, 28 June 2026. sparktoro.com
- Ahrefs, Louise Linehan, “How ChatGPT Does Query Fan-Out (15 Million Queries Studied)”, 12 October 2026, corrected October 2026. ahrefs.com
- Search Console Help, “Generative AI performance report (Search)”, read 28 October 2026. support.google.com