SEO & GEO
How to Measure AI Search Visibility in 2026
How to measure AI search visibility in 2026 with Search Console's AI report, Bing AI Performance and GA4, plus a KPI table, regex, prompt set and monthly routine.

Key takeaways
In short:
- In 2026 there is no single metric for AI search visibility. There are four data sources: Google Search Console, Bing Webmaster Tools, GA4 and prompt tracking, plus a "How did you hear about us?" question.
- Since 31 August 2026, the Search Console generative AI report has shown every site its impressions in AI Overviews and AI Mode. It shows no clicks and no queries.
- Bing AI Performance counts citations in Copilot and Bing's AI answers, including grounding queries and, since June 2026, Citation Share.
- Since May 2026, GA4 has reported visits from ChatGPT, Gemini, Copilot and similar tools as an "AI Assistant" channel. Clicks from Google's own AI answers still count as Organic Search.
- In prompt tracking, more distinct prompts beat more repetitions: 20 different prompts measure roughly three times more accurately than one prompt asked on 20 days.
AI search visibility is how often, and how, your business appears in AI-generated answers. It can appear as a cited source with a link, as a brand named without a link, or not at all. Our guide to optimising for AI search covers how to prepare your content. This article covers what comes next: how to tell whether it is working, and what to report to leadership. The tools described here work the same way across Europe. Where a report is limited to certain markets, we say so.
What changed in 2026 for measuring AI visibility?
Until early 2026, businesses relied mostly on third-party tools and manual spot checks. Now Google, Microsoft and Google Analytics provide their own data, though with gaps. As of November 2026:
| Date | Change | Source |
|---|---|---|
| 10 Feb 2026 | Bing Webmaster Tools launches "AI Performance" in public preview: citations in Copilot and Bing AI answers | Bing Webmaster Blog |
| 11 Mar 2026 | Search Console branded queries filter available to all eligible sites | Google Search Central Blog |
| 13 May 2026 | GA4 introduces the default "AI Assistant" channel (medium ai-assistant) |
GA4 release notes |
| 3 Jun 2026 | Search Console launches generative AI performance reports (to a subset of sites at first) | Google Search Central Blog |
| 16 Jun 2026 | Bing adds Intents, Topics, Citation Share and Compare (preview) | Bing Search Blog |
| 31 Aug 2026 | Search Console AI report rolled out to all websites worldwide | Google Search Central Blog |
| 13 Oct 2026 | Bing "Topics Portfolio" (beta) with opportunity queries | Bing Webmaster Blog |
| 3 Nov 2026 | Google's Martin Splitt on missing click data in the AI report: high volatility as a possible reason, no commitment either way on clicks | SEO Südwest (German) |
The rule for the rest of this guide: every source shows a different slice. Only the combination gives a useful picture, and even that is an approximation.
What does the AI report in Google Search Console show?
The generative AI performance report is a separate report in Search Console. According to Google's help centre, it counts your site's impressions in AI Overviews and AI Mode on Google Search. Discover has a separate report. In which European countries AI Overviews and AI Mode appear depends on Google's rollout. The country dimension shows you where your impressions come from.
What you get
- Impressions: how often links to your site were shown in an AI feature.
- Pages: which URLs appeared, assigned to the canonical URL.
- Countries and devices: desktop, tablet and mobile.
- Dates: days, weeks or months. According to the announcement, hourly data is also available.
- Search type filter: "Web: text-based" and "Web: multimodal" (for example Google Lens and Circle to Search).
- Export of the chart and table data.
What you don't get
- No clicks, CTR or position. The report shows impressions only. Clicks from AI Overviews and AI Mode are included in the overall performance report (Google help on clicks and impressions), but you cannot separate them there from classic search traffic.
- No queries or prompts. There is no query dimension. You can see which page was visible, but not for what.
- No data for small sites. If the report doesn't appear, Google says your site usually has too few impressions in AI features.
Two details affect how you read the report. The chart aggregates by property, so two links from your site in one AI answer count as one impression. The page table aggregates by page, so the totals differ. The usual performance report limits also apply, such as the 1,000-row cap.
Practical tip: divide AI impressions by Web impressions from the main performance report for the same period. According to Google, the AI report's data comes from that search type. The result is your AI impression share. Because the two reports count differently, treat it as a trend indicator, not an exact figure.
For brand demand, use a second feature: the branded queries filter splits brand and non-brand searches in the performance report. It works only for top-level properties, not path or subdomain properties, and only with enough query volume. Rising brand searches after more AI mentions are a signal, not proof. For how AI Overviews and AI Mode affect clicks, see our article on Google AI Mode and AI Overviews.
What does Bing AI Performance measure, and why is it worth checking even without Bing traffic?
Microsoft currently provides the most detailed first-party data on AI citations. According to the Bing Webmaster Blog, the AI Performance report in Bing Webmaster Tools shows when your content is cited as a source in Microsoft Copilot, in AI-generated summaries on Bing and in selected partner integrations.
| Metric | Meaning according to Microsoft |
|---|---|
| Total Citations | Number of times your content was shown as a source in the selected period, with no indication of placement |
| Average Cited Pages | Average number of your pages cited per day |
| Grounding Queries | Phrases the AI used to retrieve your content (sampled and generalised) |
| Page-level citations | Citations per URL |
| Intents, Topics (preview, since June 2026) | Grounding queries classified by intent and grouped into topic clusters |
| Citation Share (preview) | Your share of all citations for a grounding query |
| Topics Portfolio (beta, since October 2026) | Topics with their size, the number of sites cited, and opportunity queries where others are cited and you aren't |
Bing's help page is clear: a citation is not a click and not traffic. The data is sampled, so values for the same page and query can differ depending on the filter. Microsoft also states that Citation Share is not a ranking and does not reveal competitor domains.
It is worth checking even if Bing barely shows up in your analytics. Grounding queries are the only official data showing which phrases an AI system actually uses to retrieve your pages. For content planning, that beats any estimate. Opportunity queries also show topics where competitors are cited and you aren't.
How do you measure AI traffic in GA4?
The default "AI Assistant" channel
Since 13 May 2026, GA4 has set the medium to ai-assistant and the campaign to (ai-assistant) whenever the referrer matches a Google-maintained list of AI assistants. These visits appear in the default channel group as "AI Assistant". According to Google's definition, the channel covers sources such as ChatGPT, Gemini, DeepSeek, Copilot and Grok. Google's AI Overviews and AI Mode are explicitly excluded. Those clicks count as Organic Search.
Why add a custom channel group as well?
- Historical data: according to Google's help centre, custom channel groups can be applied to your reports retroactively. That gives you one continuous time series from before and after May 2026.
- Full control: you decide which sources count, including assistants that aren't on Google's list.
- UTM capture: ChatGPT adds
utm_source=chatgpt.comto its links. A rule on the source dimension catches these visits even when no referrer is passed.
Step by step: create an "AI assistants" channel group
- In GA4, open Admin → Data display → Channel groups. You need the Editor role.
- Click Create new channel group. It starts as a copy of the default group.
- Click Add new channel and give it a name such as "AI assistants".
- Click + Add condition group, choose the dimension Source, select matches regex and paste:
^(.*\.)?(chatgpt\.com|openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|bard\.google\.com|copilot\.microsoft\.com|copilot\.com|chat\.mistral\.ai|chat\.deepseek\.com|grok\.com|meta\.ai|you\.com|poe\.com)$
- Click Save channel. Then use Reorder to move the channel to the top, or at least above "Referral" and "Organic Search", and click Apply. GA4 assigns each visit to the first channel that matches.
- Click Save group. Analyse the data under Reports → Acquisition → Traffic acquisition with the new group as the primary dimension, and connect it to your key events such as form submissions or call clicks.
Google's own regex example in the help centre is very broad (it includes .*gpt.* and .*copilot.*, among others) and can capture unrelated sources. A list of specific domains is more precise, but you have to maintain it. Standard properties allow two custom channel groups with up to 50 channels each. Assistants change domains from time to time. Check the Source dimension in GA4 every few months to see which values actually arrive, and update the list to match.
The limits of GA4
- Missing referrers hide visits: links opened from apps or copied and pasted often arrive without a referrer and land in "Direct". October 2026 showed how fragile this is. According to Search Engine Roundtable, ChatGPT links carried neither UTM parameters nor a referrer for about a week after the GPT-6 launch. ChatGPT traffic dropped in reports while Direct rose. The report attributes this to a bug that was fixed on 15 October.
- Consent: in the EU and UK, GA4 records only partial data without consent, depending on your setup. This applies to AI traffic as much as to any other channel.
- Google stays hidden: in GA4, clicks from AI Overviews and AI Mode cannot be told apart from ordinary Google clicks.
Any AI share you see in GA4 is therefore a floor, not a total. Be wary of anyone who promises an exact total: nobody can fully measure it today.
How does prompt tracking work, and how reliable is it?
Prompt-tracking tools ask ChatGPT, Gemini, Perplexity or Google's AI Mode the same questions at regular intervals. They then check whether your brand is named or your domain is cited. It is the only method that also captures mentions without a link and competitors. It has two structural weaknesses.
1. Answers fluctuate heavily. SISTRIX analysed 82,619 prompts over 17 weeks. Each week, Google's AI Mode replaced 56% of its cited sources and ChatGPT 74%. AI Overviews were completely stable for just over half of the queries. A single check is a snapshot.
2. Breadth beats repetition. In an analysis of 10,826 measurement series published on 27 October 2026 and covering four AI systems and eight countries, the typical error for a budget of 20 queries was:
| How 20 queries are spent | Typical deviation from the true value |
|---|---|
| 1 prompt on 20 days | ±34 percentage points |
| 20 different prompts on one day | ±9.5 percentage points |
| 100 different prompts on one day | ±4.2 percentage points |
According to SISTRIX, you need about 70 prompts for ±5 points and about 200 for ±3. There is also a selection effect: in self-chosen prompt sets, the brand's own visibility looked better than in a neutral market sample in 63–80% of comparisons. Your own set therefore shows the trend reliably, but usually flatters the level. Judge changes by weekly averages, because daily values jump around even when nothing has changed.
What this means in practice: a fixed set is ideal for tracking the effect of your own work. It cannot back up claims like "AI recommends us in 40% of answers".
Starter set: 20 prompts by intent
Build your set from real customer wording: email enquiries, questions your sales team hears, and queries from Search Console. The grid below uses a fictional accounting firm in Amsterdam as an example. Write the prompts in the languages your customers actually use. If you sell in several European countries, track each language as its own set.
| Intent | Count | Example prompt (fictional) |
|---|---|---|
| Understand the problem | 4 | "What do I need to know about tax in my first year as a BV founder?" |
| Solution/category | 4 | "Is an accountant worth it for a sole trader with €80,000 revenue?" |
| Vendor shortlist | 4 | "Which accounting firms in Amsterdam specialise in e-commerce?" |
| Comparison/criteria | 3 | "How do I recognise a good accountant for start-ups?" |
| Price/effort | 2 | "How much does monthly bookkeeping by an accountant cost?" |
| Local/constraints | 2 | "Accountant in Amsterdam with digital receipts and English-language advice" |
| Brand (control) | 1 | "What does [firm name] offer and who is it for?" |
Rules for clean measurement:
- Freeze the set. Don't reword existing prompts. Add new ones and analyse them separately.
- Query neutrally: sign out or use a private window, and start a new chat for each prompt. Record the date, system and location.
- Record for each answer: brand named (yes/no), domain linked (yes/no), competitors named, and whether the facts are correct (yes/no).
- Analyse brand prompts separately. They test whether what the AI says about you is accurate, not how visible you are.
Doing this by hand takes about half a working day for 20 prompts across three systems. A tool pays off once you need more prompts, more systems or weekly runs.
Why does every form need "How did you hear about us?"
AI assistants often influence buying decisions without generating a click. Someone asks ChatGPT for three providers and then searches for your name on Google. GA4 records that as a brand search or a direct visit. Only the person can tell you where they first heard of you.
How to set up self-reported attribution:
- Add an optional field to your contact form and ask in the first call: "How did you hear about us?"
- Offer answer options that name AI explicitly: Google search · AI assistant (ChatGPT, Gemini, Perplexity, Copilot …) · referral · social media · event · other. Add a free-text field.
- Store the answer as its own CRM field and review it monthly, separately for leads and closed deals.
- Collect only what you need: keep the field voluntary, in line with the GDPR's data-minimisation principle.
Self-reported answers are imprecise because people remember selectively. But they capture exactly what the tools cannot see: influence without a click.
Which KPIs actually reflect AI visibility?
| Metric | Source | What it tells you | What it doesn't tell you |
|---|---|---|---|
| AI impressions (AIO + AI Mode) | Search Console AI report | How often Google linked your pages in AI answers, and which pages | Clicks, queries, position, prompt wording |
| AI impression share | Search Console (AI report ÷ total Web) | How much AI surfaces weigh in your Google visibility | An exact share, because the counting methods differ |
| Citations and cited pages | Bing Webmaster Tools | How Copilot and Bing AI use your content | Clicks, traffic, visibility in ChatGPT or Gemini |
| Citation Share, grounding queries | Bing Webmaster Tools | Your share per query, and the phrases the AI uses to retrieve content | Rankings, competitor domains, complete data (it is sampled) |
| "AI Assistant" sessions | GA4 | Measurable visits from ChatGPT, Perplexity, Gemini and others | Visits without a referrer or UTM, clicks from Google's AI answers |
| Key events from AI traffic | GA4 | Enquiries and purchases after an AI click | AI influence without a click |
| Mention rate in your prompt set | Prompt tracking (manual or tool) | Trend in brand mentions and citations, including competitors | Your true market level (selection effect), usage volume |
| Branded queries | Search Console branded filter | How demand for your name develops | What caused the increase |
| Leads citing an "AI assistant" | Form/CRM | Business impact, even without a click | Exact attribution to a platform or answer |
Monthly routine: check AI visibility in 90 minutes
Run the checks on the same day each month and compare with the previous month and, once you have the data, the same month last year.
- Search Console (20 min): export the AI report for the last full month. Note the top 10 pages by AI impressions, mark winners and losers, and calculate your AI impression share. Use the branded filter to record clicks from brand searches.
- Bing (15 min): export total citations, cited pages and the top 10 grounding queries. Use Compare against the previous month and save the opportunity queries for content planning.
- GA4 (20 min): review sessions, engagement rate and key events for the "AI assistants" channel by source and landing page. Also check how Direct is trending: a sudden jump can signal a tracking problem like the one in October 2026.
- Prompt set (20 min with a tool, longer by hand): calculate the monthly average mention and citation rate per system. Note new competitors and any wrong statements about your business.
- CRM (10 min): count leads and deals that named an AI assistant.
- Decide (5 min): pick no more than three actions. Examples: expand a page using the opportunity queries, correct a wrong AI statement with clear facts on your site, or refresh a page that is being cited.
Monthly tracking sheet (one row per month):
| Month | AI impr. (GSC) | AI share | Bing citations | AI sessions (GA4) | AI enquiries (GA4) | Set mention rate | "AI" leads (CRM) | Actions |
|---|---|---|---|---|---|---|---|---|
| Oct 2026 | … | …% | … | … | … | …% | … | … |
| Nov 2026 | … | …% | … | … | … | …% | … | … |
How do you report AI visibility to management?
Leadership wants to know whether AI search brings or costs enquiries, and whether the budget is working. A one-page AI report is enough:
- Three numbers with trends: AI impressions (Google), AI sessions with enquiries (GA4) and leads that named an "AI assistant" (CRM).
- One sentence of context, for example: "AI answers are showing our service pages more often. Measurable enquiries are still small, but more leads name AI as their first touchpoint."
- State the uncertainty openly: Google gives no AI click data, Bing shows only citations, GA4 shows a floor, and the prompt set shows trends, not market level.
- Actions and the next review date, with no raw-data appendices.
Avoid reporting single tool scores such as an "AI ranking" or "AI score" without explaining the method. You will have to explain every swing. For a sense of what ongoing support costs, see our article on GEO pricing.
Conclusion
AI search visibility can be measured in 2026, just not with a single number. Search Console shows where Google links you in AI answers. Bing shows how often Copilot cites you. GA4 shows the visits that actually arrive. Prompt tracking shows the trend, including competitors, and "How did you hear about us?" captures influence without a click. Review these five sources monthly using the same routine and you will see real progress and be able to report it honestly.
If you want tracking, channel groups and reporting set up properly, our web analytics service can help. For content and technical work that AI systems can cite, see Contentoren's generative engine optimisation (GEO) service, from €4,500 setup (incl. VAT). We can review your starting point in a no-obligation call: get in touch.
Sources
Accessed 6 November 2026
- Google Search Central Blog: "Search Generative AI performance reports in Search Console" (3 Jun 2026, updated 31 Aug 2026)
- Search Console Help: "Generative AI performance report (Search)"
- Search Console Help: "What are impressions, position, and clicks?"
- Google Search Central Blog: "Introducing the branded queries filter in Search Console" (20 Nov 2025, updated 11 Mar 2026)
- Google Search Central: "AI features and your website"
- SEO Südwest: "Google: keine Klickdaten für KI-Features in Search Console …" (3 Nov 2026, German)
- Bing Webmaster Blog: "Introducing AI Performance in Bing Webmaster Tools Public Preview" (10 Feb 2026)
- Bing Search Blog: "New AI Visibility Insights in Bing Webmaster Tools: Intents, Topics, Citation Share, Compare" (16 Jun 2026)
- Bing Webmaster Blog: "Introducing Topics Portfolio (Beta) in AI Performance" (13 Oct 2026)
- Bing Webmaster Tools Help: "AI Performance"
- Google Analytics Help: release notes (13 May 2026: AI Assistant traffic)
- Google Analytics Help: "Default channel group"
- Google Analytics Help: "Custom channel groups"
- Search Engine Roundtable: "OpenAI Drops UTM Parameters From ChatGPT Results (GPT-6)" (14 Oct 2026, updated 15 Oct 2026)
- SISTRIX: "AI Citation Drift: Wie stabil sind Quellen in AI-Suchergebnissen?" (28 Apr 2026, German)
- SISTRIX: "Breite schlägt Wiederholung: Wie man AI-Sichtbarkeit verlässlich misst" (27 Oct 2026, German)
Frequently asked questions
Can Google Search Console show how often my site appears in AI Overviews?
Yes. Since 31 August 2026 the generative AI performance report has been available to all websites. It shows impressions in AI Overviews and AI Mode by page, country, device and date. It does not include clicks, queries or positions.
How do I identify visitors from ChatGPT, Perplexity and Gemini in GA4?
Since May 2026 GA4 has assigned recognised AI assistants to the default channel 'AI Assistant' (medium ai-assistant). For your own analysis that also applies to past data, add a custom channel group with a source regex. Clicks from Google's AI Overviews and AI Mode are still counted as Organic Search.
Does Bing show whether Copilot cites my content?
Yes. The AI Performance report in Bing Webmaster Tools counts citations in Microsoft Copilot, AI-generated summaries in Bing and selected partner integrations. It shows cited pages and grounding queries, and since June 2026 also Citation Share, but no clicks.
How many prompts do I need for reliable prompt tracking?
According to a SISTRIX analysis of 10,826 measurement series, 20 prompts give only a rough estimate with about ±10 percentage points of error; around 70 prompts are needed for ±5 points. More distinct prompts help more than repeating the same prompt. A fixed set of 20 prompts still works well for tracking trends through weekly averages.
Which AI visibility KPIs should I report to management?
Report a few robust figures: AI impressions from Search Console, citations from Bing, sessions and enquiries from AI assistants in GA4, and the number of leads who answer 'How did you hear about us?' with an AI assistant. Always add a note on what these numbers cannot measure.


