How to Measure AI Search Visibility (GEO) in Google & Bing

What is AI search visibility?

AI search visibility is how often AI-generated answers show, cite or link to your pages: Google’s AI Overviews and AI Mode, Microsoft Copilot and Bing’s AI summaries, and assistants such as ChatGPT, Perplexity and Gemini. You can measure a useful part of it for free with first-party data: Google’s generative AI performance report, Bing Webmaster Tools’ AI Performance report, the query data in Search Console and Bing, and AI referral traffic in your analytics. None of them shows everything, so the work is in combining them.

GEO, AEO and SEO: what Google and Bing say

GEO (generative engine optimization) and AEO (answer engine optimization) are the labels for this work. The two search engines frame them differently:

• Google says optimising for generative AI search is still SEO. Its guide to optimising for generative AI features explains that AI Overviews and AI Mode are rooted in its core ranking and quality systems, and retrieve pages from the Search index.
• Bing calls its AI Performance report “an early step toward Generative Engine Optimization (GEO) tooling”, and reports citations as a metric in their own right.

Both views lead to the same practical conclusion. The fundamentals haven’t changed, but the measurement has. Clicks and rankings no longer capture all of your visibility, so you need a few extra numbers next to them.

The four first-party signals you can get for free

Only the third row is available through the Search Console and Bing APIs, which is the row GSC MCP Server reads. The first two are dashboards you export from, and the fourth lives in your analytics tool.

Google’s generative AI report: impressions, not clicks

Google’s generative AI performance report went live for all sites on 31 August 2026. It counts impressions in AI Overviews and AI Mode, combined, with no clicks, CTR, position or queries, and it isn’t in the Search Console API. The number to build from it is AI share: a page’s AI impressions divided by its total Web impressions for the same dates. Our full guide to the Search Console generative AI report covers the rest.

Bing’s AI Performance report: citations and grounding queries

Bing Webmaster Tools opened AI Performance as a public preview on 10 February 2026. It covers Microsoft Copilot, AI-generated summaries in Bing and select partner integrations, which Microsoft doesn’t name. Unlike Google’s report, it counts citations and shows the queries behind them.

The report has grown quickly since launch:

• March 2026: grounding queries were mapped to cited pages, so you can click a query to see which of your pages it cites, or a page to see which queries cite it.
• 16 June 2026: Microsoft added Intents, Topics, Citation Share and Compare in preview. Intents label grounding queries as informational, commercial, navigational and so on; Compare overlays a previous period.
• 13 October 2026: a Topics Portfolio tab (beta) groups grounding queries into topics, with approximate topic size, the number of sites cited, and opportunity queries where other sites are cited and yours isn’t. Microsoft says the numbers are directional and smaller sites may not see it.

What Bing’s report doesn’t give you

• No clicks. Every metric is citation-based. A citation is not a visit.
• No user prompts. Grounding queries are the phrases the AI generated to retrieve content. They are often more specific than anything a person typed, because a single question can fan out into several searches.
• No documented API. Bing Webmaster Tools’ API documents traffic, crawl and submission methods, but nothing for AI Performance. Take a monthly export or a copy of the tables you track.
• No named ChatGPT coverage. ChatGPT search runs its own crawler, OAI-SearchBot. Don’t read Bing citations as ChatGPT citations.

For the setup side of Bing, from verification to IndexNow, see Bing Webmaster Tools for ChatGPT and Copilot visibility.

Proxy signals in Search Console and Bing query data

The AI reports tell you where you appear. Your query data tells you what that is doing to clicks. Three proxies are worth tracking every month; if a metric looks odd, check how Search Console calculates CTR and position.

1. CTR compression at a steady position

When an AI answer satisfies the searcher, your position can hold or even improve while CTR falls. In Google, every link in an AI Overview shares the Overview’s position, so a cited page can look higher than ever and still lose clicks. Look for pages and queries where position is flat or better and CTR is down by a fifth or more.

Google impressions before 27 April 2026 are unreliable, so judge year-on-year on clicks only.

2. Question and long, conversational queries

Questions and long queries are where AI answers appear most. Track their share of your impressions and their CTR over time with a regex filter in Search Console, such as ^(how|what|why|when|where|which|who|can|does|is|should)\b for questions or ^(\S+\s+){7,}\S+$ for eight words or more. The Search Console regex cheat sheet has tested variants. Rare queries are anonymised and left out of query tables, so this segment always undercounts; follow the trend, not the total.

3. Search appearance and other SERP changes

Group Search Console data by searchAppearance to see which result features your impressions come from. Google doesn’t document an AI Overview or AI Mode search appearance, so don’t expect one. The point is to rule out other explanations, such as the end of FAQ rich results on 7 May 2026.

Joining Bing citations to clicks

The most revealing check combines engines. Export your most-cited pages from Bing’s AI Performance report and ask how those pages perform in classic search on both Google and Bing:

A page cited often but rarely clicked is doing brand work, not traffic work. A page that ranks well on Google but is never cited by Bing is worth checking for Bing indexing problems.

AI referral traffic in your analytics

Citations only matter commercially if some of them send people to you. Google Analytics 4’s default channel group now includes an AI Assistant channel for visits from sources such as ChatGPT, Gemini, DeepSeek, Copilot or Grok. GA4 sets the medium to ai-assistant when the referrer matches its list.

• Google’s own AI features aren’t in it. The channel explicitly excludes AI Overviews and AI Mode. Those clicks land in Organic Search, mixed in with all other Google organic visits.
• Not every assistant passes a referrer. Some app traffic arrives as Direct, so treat the channel as a floor.
• Check the Referral channel too. If you see assistants such as Perplexity or Claude there, build a custom channel group that matches the session source with a pattern like .*(chatgpt|perplexity|claude|copilot|gemini|deepseek).*.

We explain how GA4 counts AI Assistant and Organic Search traffic in our Search Console vs GA4 guide. The useful report is landing pages by AI Assistant sessions, with conversions. Set it beside Bing’s cited pages and Google’s top AI pages. Pages that appear on all three lists are your proven AI assets; protect and update them first.

Check the basics: can AI systems reach your pages?

Measurement is pointless if the pages can’t be retrieved. Each system has its own gate:

• Google: the page must be indexed and eligible for a snippet, and the property must be included under Settings → Search generative AI. Blocking Google-Extended doesn’t remove you from Search.
• Bing and Copilot: the page must be in Bing’s index, with bingbot allowed. Bing’s own advice for AI answers stresses freshness and recommends IndexNow.
• ChatGPT search: OpenAI says sites that block OAI-SearchBot won’t be shown in ChatGPT search answers. Blocking GPTBot only affects training.

A hosted Search Console and Bing MCP server such as ours can check the Google and Bing side in one request with multicall: inspect_url for index status, canonical and robots.txt state, list_sitemaps for Google sitemaps and Bing feeds, and get_bing_crawl_stats for crawl errors, robots.txt blocks and pages in Bing’s index. The technical SEO workflow for developers goes deeper.

What you can’t measure (yet)

Be explicit about the gaps in any report you send. As of October 2026:

• ChatGPT, Perplexity, Claude and Gemini app citations. As far as we know, none of them offers site owners a citation report like Bing’s. You see their clicks in analytics, not their citations.
• Clicks from Google’s AI features. Counted, but mixed into Web clicks in Search Console and Organic Search in GA4.
• Prompt volume. No engine publishes how often people ask a given question in an AI assistant. Grounding queries and fan-out queries are machine-generated, not demand data.
• What the answer said about you. Accuracy, sentiment and whether a competitor was recommended need manual checks or a paid tracker.

Paid AI visibility trackers fill some of these gaps by running a fixed set of prompts through several assistants and recording who gets cited. That is a sample, and answers vary by run, location and account, so use them for direction. Use the first-party numbers above for anything you report as fact.

llms.txt, chunking and “AI markup”: what Google and Bing say

A lot of GEO advice is about files and formatting. Here is what the engines themselves say:

• llms.txt. The llms.txt proposal suggests a Markdown file at your site root that gives AI agents a concise guide to your content. Google’s AI guide says Google Search doesn’t use llms.txt and that creating one neither helps nor harms your visibility there. It is fine to publish one for other tools that read it. Bing’s AI Performance announcement doesn’t mention it.
• Chunking. Google says there is no need to break content into tiny pieces for AI, and no ideal page length.
• Structured data. Not required for Google’s AI features, and there is no special schema. It is still worth using for rich results.
• Structure and evidence. Bing recommends clear headings, tables and FAQ sections, supporting evidence and fresh content. On-page FAQs are still fine as content, even though Google stopped showing FAQ rich results in May 2026.

For measurement, the lesson is simple: don’t put “has an llms.txt” or “has AI schema” on your scorecard. Track outcomes (impressions, citations, clicks and conversions), not inputs the engines say they ignore.

A monthly AI visibility scorecard

Run this in the first week of each month, once Search Console’s 2–3 day reporting delay has passed. The template below uses illustrative numbers for a mid-sized B2B site.

Read the example as a story: AI visibility rose on both engines, clicks dipped slightly, CTR on questions slipped, and AI Assistant visits and conversions grew. That is a site gaining presence in AI answers while classic clicks compress, which is a very different message from “traffic is down 2%”.

The monthly routine

1. Export the Google generative AI report (chart and Pages) for last month.
2. Export Bing AI Performance: citations, cited pages, grounding queries, and Citation Share for your ten priority queries.
3. Pull the search data with one prompt (below).
4. Note AI Assistant sessions and conversions from GA4.
5. Inspect the top cited pages for indexing problems.
6. Write three sentences: what changed, why you think it changed, and what you’ll do next.

Agencies can drop this table straight into the monthly client reporting workflow. If you haven’t connected Search Console and Bing to your assistant yet, the setup guide for Claude, ChatGPT and Cursor takes a few minutes, and our explainer covers what a Search Console MCP server is.

Frequently asked questions