For marketing leads: measuring GEO and reporting it upward

This guide is for the marketing lead who has to answer “is GEO working?” to a CEO or budget owner. It lists the data that actually exists, what each source can and can’t show, how to build a monthly report, and which promises to refuse. Setting it up takes a day or two. Running it takes a few hours a month, plus the time to sample prompts.

The short version

1. Use the platforms’ own reports first: Search Console for Google, Webmaster Tools for Bing.
2. Track AI referral traffic in analytics, knowing much of it arrives untagged.
3. Run a fixed prompt panel in the assistants your customers use, with enough repeats.
4. Connect GEO to leads by asking customers and tagging CRM records.
5. Report one headline trend with its range, plus a few concrete before-and-after answers.
6. Promise effort and measured movement, never rankings or a fixed position.

1. Start with the platforms’ own reports

Google and Microsoft both report AI visibility directly. These are the only numbers that come from inside the systems.

Search Console’s Generative AI performance report shows impressions: how often links to your site were shown in AI Overviews and AI Mode. Data can be grouped by page, country, date or device.

Officially documented · Google Search Console Help

Sites appearing in Google’s AI features “are included in the overall search traffic in Search Console”, within the Performance report’s Web search type.

Officially documented · Google Search Central

Bing’s AI Performance report counts citations in Copilot and Bing AI answers, with the grounding queries behind them. Microsoft: “AI Performance does not measure rankings, authority, performance, or importance. It simply shows which content was cited.”

Officially documented · Microsoft Bing

What to pull each month:

Source	Metric	Use it for
Search Console, Generative AI report	AI impressions by page	Which pages show in AI Overviews and AI Mode
Search Console, Performance	Clicks and impressions (Web)	Overall Google search trend, AI included
Bing Webmaster Tools, AI Performance	Citations, cited pages, grounding queries	Which pages Copilot cites, for what

Details: Generative AI performance report: AI Overviews and AI Mode in Search Console, Bing AI Performance report. Neither report covers ChatGPT, Perplexity or Chinese assistants.

2. Track AI referral traffic, and its gaps

Visits from AI assistants can be measured when the browser passes a referrer.

Google Analytics 4 has an AI Assistant channel in its default channel group. It covers visits referred by assistants such as ChatGPT, Gemini, DeepSeek, Copilot or Grok, and sets the medium to ai-assistant. Google’s own AI Overviews and AI Mode are excluded from it and count under Organic Search.

Officially documented · Google Analytics Help

Google Analytics announced the AI Assistant channel on May 13, 2026, naming ChatGPT, Gemini and Claude as examples.

Officially documented · Google Analytics Help

Much AI-influenced traffic still won’t be labelled. In-app browsers often drop the referrer. Many people read an answer, then type your name into a search box or your URL into the browser.

DATA17’s 2026 guide argues that when a user learns a brand name from an AI answer and then types the URL, the visit is recorded as direct traffic. It concludes that GEO needs its own visibility metrics rather than relying on traffic data. This is reasoning, not a measured share.

Hypothesis · DATA17 AI Research Lab · 解数智能 (unpublished report, not publicly available; readers can't check it)

So report AI referral sessions as a floor, not the total. Watch branded search and direct traffic alongside it.

Google says it has seen that clicks from results pages with AI Overviews “are higher quality (meaning, users are more likely to spend more time on the site)”. It suggests tracking conversions and time on site in tools such as Google Analytics.

Officially documented · Google Search Central

That is Google’s statement about its own product. Check it against your conversion data rather than repeating it to leadership as fact.

AI referral traffic is the part you can see. Report it as a floor, never as the whole effect.

3. Sample prompts properly

For assistants with no report, you have to ask them and count. One run measures nothing, because answers change from run to run.

Observed in 2025-09, more than 9 months ago. Answer engines change often; re-test before relying on this.

Miaozhen’s September 2025 study (2,025 runs of five typical questions, 135 per model, three Chinese models) found results reach basic stability only at 30 or more runs per prompt, and 60 or more for low-visibility brands.

Observed in practice · observed 2025-09 · 中国商务广告协会AI营销应用工作委员会 · 秒针营销科学院

A 2026 study of Pilates studios in Hefei found API and consumer-app results can disagree sharply. DeepSeek recommended the brand in 69% of API runs but 1% of app runs.

Observed in practice · observed 2026-08 · 腾讯新闻（报道合肥图灵象普拉提行业GEO白皮书）

Rules for the panel:

• Build it from real customer questions, grouped by intent. Report category questions apart from brand-name questions.
• Freeze it. Version it if you must change it.
• Record product, mode, account state, location and date for every run.
• Report each platform separately, with the share of runs and its range, for example “41% ± 9 pts, n = 30”.

Full protocol: Measuring AI visibility by sampling prompts. Some vendor guides suggest asking each question only two or three times. That is cheaper, but the error at that sample size is wider than most month-to-month changes.

Google warns site owners to be wary of tools claiming internal Google metrics: “No third-party tool has access to our internal ranking or AI systems.”

Officially documented · Google Search Central

Third-party GEO tools sample answers from outside, as you would. Ask any vendor how many runs, which products and which locations sit behind a number (Claim: GEO tools can see your rankings inside Google's AI).

4. Connect GEO to business results

Leadership cares about leads and revenue. The link from an AI answer to a sale is indirect, so build the evidence deliberately.

Yisou Baiying’s 2026 white paper proposes four ways to attribute leads: direct (the customer says they came from AI), assisted (customers mention using AI during the sale), trend (brand search, site visits and enquiries improve after GEO work) and comparison (optimised questions against unoptimised ones and competitors). It recommends adding an AI-source field to forms and the CRM. These are proposals; the paper reports no validation.

Hypothesis · 一搜百应 GEO 研究院 (unpublished report, not publicly available; readers can't check it)

Practical steps:

• Add “an AI assistant (ChatGPT, Doubao, etc.)” to “How did you hear about us?” on forms.
• Ask sales to log it in the CRM when a prospect mentions AI.
• Keep a control group of questions you haven’t worked on. Change on worked questions only means something next to them.

5. What to put in the report

R3’s 2026 advertiser white paper observes that GEO vendors have no shared KPI definitions, sampling or observation windows, so their results are hard to compare. It advises that early projects use visibility, recommendation rate, share of voice and accuracy of core messages as KPIs, with a reasonable observation period, rather than judging GEO like performance advertising.

Hypothesis · R3（胜三） (unpublished report, not publicly available; readers can't check it)

DATA17’s guide suggests reporting a single headline number, the mention rate, with one sentence on the trend. It suggests placing an AI answer from three months earlier beside today’s answer to the same question, and keeping the other metrics in an appendix.

Hypothesis · DATA17 AI Research Lab · 解数智能 (unpublished report, not publicly available; readers can't check it)

A one-page monthly report:

1. Headline: mention rate on the category panel, per platform, with range and run count, against the baseline.
2. Platform data: Google AI impressions and Bing AI citations for priority pages.
3. Traffic: AI Assistant channel sessions and conversions, marked as a minimum.
4. Accuracy: answers that describe you wrongly or out of date, and what you did about them.
5. Evidence: two or three before-and-after answer pairs.
6. Next month: the work planned, not a forecast of results.

Show the trend, the range and the sample. A number without its sample size is a guess with a decimal point.

What not to promise

Even the vendor-written ROI white paper separates influenceable metrics from results that cannot be promised. It says GEO can raise the probability of being understood, cited and recommended, but should not promise fixed rankings, a fixed recommendation order or fixed sales.

Hypothesis · 一搜百应 GEO 研究院 (unpublished report, not publicly available; readers can't check it)

• No “first place in ChatGPT.” No platform documents a stable ranking position, and answers vary run to run.
• No causal claims from one change. Microsoft says trend changes in its report can’t be tied to a specific edit (Bing AI Performance report).
• No fixed timelines for category questions. Broad “best X” questions usually move last (A 90-day GEO plan for a company brand, starting from zero).
• No numbers bought from manipulation. Planted mentions risk penalties and, in China, regulatory action (Claim: seeding brand mentions across the web boosts AI visibility, GEO regulation in China).

If an agency promises any of these, see Before you hire a GEO agency.

How you’ll know it’s working

• The platform reports show more AI impressions or citations on priority pages, over several months.
• The category panel’s mention rate rises by more than its sampling error.
• Wrong descriptions fall, and stay fixed.
• More leads cite AI as a source, and the trend holds when you compare worked and unworked questions.

Evidence at a glance

Recommendation	Evidence
Use Search Console and Bing AI reports	Officially documented
Track GA4 AI Assistant channel, as a floor	Officially documented (channel); hypothesis (the floor)
Run 30+ repeats per prompt; measure in the real app	Observed in practice
Distrust tools claiming internal Google data	Officially documented
Attribute leads through forms, CRM and controls	Hypothesis
Report visibility, accuracy and trend, not rankings	Hypothesis (industry advice)
Never promise fixed rankings	Hypothesis, consistent with platform docs