How to measure AEO and GEO performance: AI search metrics that mean something

Why does one screenshot prove nothing, and what should you measure instead?

AI answers change from one run to the next. In a study published in January 2026, SparkToro and Gumshoe had hundreds of volunteers run 12 brand-recommendation prompts 60 to 100 times each on ChatGPT, Claude and Google's AI, 2,961 runs in all. ChatGPT and Google's AI returned the same list of brands less than 1% of the time, and the same list in the same order less than 0.1% of the time. A screenshot showing your company “first in ChatGPT” records one draw, not a position.

What held steadier was how often a brand appeared across many runs. That is the measure to build on: your share of answers, meaning the runs that name you divided by all runs, broken down by engine and by topic. There is no stable “ranking in AI” to report, so do not pay for one.

• ·Share of answers: the share of runs in which you are named.
• ·Citation rate: the share of runs that link to one of your pages as a source.
• ·Share of voice: your mentions as a share of all brand mentions for the same prompts, against competitors you choose.
• ·Accuracy: whether the engine describes what you do, where and for whom correctly.
• ·Cited sources: the third-party pages the engines quote for your questions, which becomes your list for reviews, press and outreach.

How many runs do you need?

Enough that the number means something. If you are named in 6 of 20 runs, your share is 30%, but a standard 95% confidence interval runs from roughly 15% to 52%. At 60 of 200 runs the share is still 30%, and the interval narrows to roughly 24% to 37%. Report the margin of error next to every share, and treat a move inside it as noise.

Record the conditions of every run: engine, mode (for example AI Overviews or AI Mode), date, country, and whether the account was logged in. Answers differ across all of these, and a change in conditions can look like a change in visibility.

How do you choose the prompt set?

The prompt set is the list of questions you track. Get it wrong and the numbers are precise but irrelevant. Build it from what buyers actually ask, not from your keyword list.

• ·Start with real questions: sales call notes, enquiry forms, Search Console queries, and the grounding queries in Bing's AI Performance report.
• ·Cover the stages of a decision: category (“best systems integration agency for manufacturers”), problem (“how to connect our ERP to the shop floor”), comparison (“X vs Y”) and brand (“is X any good”).
• ·Write several phrasings of each question. When SparkToro asked people to write their own prompts for the same need, the wording varied widely, yet the same brands kept appearing, so a spread of phrasings tells you more than one perfect prompt.
• ·Twenty to fifty questions is a workable first set for most B2B companies; group them into topics so you can see where you are strong and where you are absent.
• ·Track the engines your buyers use, typically ChatGPT, Google's AI Overviews and AI Mode, Gemini, Perplexity, Copilot and Claude.
• ·Freeze the set. Change it at most once a quarter and log what changed, or month-on-month comparisons stop meaning anything.

How do you track AI referral traffic in GA4?

Since May 2026, GA4 has an AI Assistant channel in its Default Channel Group. When a visit's referrer matches Google's list of AI assistants, GA4 sets the medium to “ai-assistant” and the campaign to “(ai-assistant)”. Google's definition names ChatGPT, Gemini, Deepseek, Copilot and Grok as examples, and its release note also names Claude. You will find it under Reports, Acquisition, Traffic acquisition.

Google AI Overviews and AI Mode are deliberately excluded from that channel. GA4 counts their clicks in Organic Search, so they cannot be separated there.

• ·Google has not published the full list of AI referrers, and the default channel group cannot be edited. Build a custom channel group as well, with an AI channel above Referral that matches sources such as chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and claude.ai.
• ·ChatGPT adds utm_source=chatgpt.com to the links it sends, according to OpenAI's publisher FAQ, so those visits carry their source even when the referrer is lost.
• ·Some AI visits arrive with no referrer at all, for example from apps or copied links, and land in Direct. Treat your AI numbers as a floor, not a total.
• ·Look at landing pages for the AI channel: they show which of your pages the engines actually send people to.

How do you tie AI visits to leads and revenue?

Visits are a proxy. What matters is whether AI-referred buyers become enquiries and then customers, and that lives in your CRM, not in GA4.

• ·Capture source, medium, campaign and landing page in hidden fields on every form, and pass them to the CRM with the lead.
• ·Store both the first touch and the last touch, because many buyers research in an assistant and come back later by searching your name.
• ·Add a free-text “How did you hear about us?” field. Answers like “ChatGPT suggested you” catch the visits analytics files as Direct or branded search.
• ·Report AI-sourced leads, qualified opportunities and won work as counts. The numbers are often small, and a percentage of a small number misleads.

What do Bing and Google report themselves?

Bing Webmaster Tools added an AI Performance report in public preview in February 2026. It shows how often your pages are cited as sources in Microsoft Copilot, AI-generated summaries in Bing and some partner integrations: total citations, average cited pages per day, page-level citation counts, trends over time and a sample of the grounding queries that led to them. In June 2026 Bing added previews of Intents, Topics, Citation Share (your percentage of all citations for a grounding query) and Compare. It reports citations, not clicks, and Bing says it respects robots.txt and other content-owner controls.

Google does not offer an equivalent report. Its documentation says sites appearing in AI Overviews and AI Mode are counted in the Search Console Performance report under the Web search type, together with ordinary results. For Google's AI features, your own repeated runs remain the only direct measure of being named.

How often should you report, and how does Seamwire do it?

We start with a free AI visibility check: ten real buyer questions put to four engines, showing who gets named today and whether AI can read your site. An audit extends that to fifty or more questions across five engines with repeated runs, and wires AI visits into your analytics and CRM. A programme then tracks the set weekly and reports monthly.

• ·Weekly: run the full prompt set across the chosen engines, enough times to keep the margin of error useful.
• ·Monthly: report share of answers by engine and topic with its margin, share of voice against named competitors, the sources engines cite, AI visits, AI-sourced leads, and what changed on the site that month.
• ·Quarterly: review the prompt set and the competitor list, and note any model or product changes that may explain a shift.
• ·Do not react to a single week. Answers vary, and engines change their models without notice.