Generative engine optimization measurement: how to track citations in AI answers

Generative engine optimization needs a measurement plan

A buyer asks ChatGPT, Google or Copilot for suppliers in your market. The answer names several businesses and cites a few pages. Generative engine optimization (GEO) measurement shows whether your company appears, which page earned the citation and whether the visibility leads to an enquiry.

Rank tracking alone cannot answer those questions. An AI answer can cite a page that ranks below the first few classic results, and the cited sources can change when the wording, location or language of the prompt changes. A useful GEO report therefore connects prompts, citations, pages, markets and commercial outcomes.

The process below gives you a repeatable way to establish that baseline and measure progress without turning every chatbot response into a vanity metric.

Start with buyer questions, not a list of keywords

Choose questions that reflect a real buying decision. A multilingual software company might test "localization agencies for a French product launch", "technical SEO support for a German website" and "how to structure an international site for Belgium". Each question carries a market, a need and an expected type of answer.

Build the test set from four sources:

• sales calls and enquiries that reached your team;
• Search Console queries with clear commercial intent;
• questions on service pages and support tickets;
• competitor comparisons that buyers already make.

Keep the first set manageable. Twenty to forty prompts per market are enough to expose patterns while still allowing a person to review the answers. Store the exact wording because a small change can produce a different set of sources.

Translate the intent for each market rather than translating the sentence word for word. A French buyer in Belgium may name a different service, location or legal requirement from a buyer in France. The prompt set should preserve those differences because they are part of what you are measuring.

Record the answer, citation and landing page separately

A brand mention and a citation are related but different. The answer may name your business while linking to another source, or cite your article without naming the brand in the prose. Record both.

For every prompt, capture:

ChatGPT search can include inline citations and a source panel. OpenAI also states that inclusion is not guaranteed and that OAI-Searchbot must be able to crawl a site for its pages to be eligible. A technical check belongs beside the prompt test whenever the domain never appears.

Screenshots help with spot checks, but a structured record is easier to compare. Save the answer URL where the platform provides one and keep a short note when the citation supports only part of the response.

Use the platform reports where they exist

Manual prompts show the answers you chose to test. Platform data provides a broader view of citations and visibility that occurred outside that sample.

Bing Webmaster Tools reports AI citations across Microsoft Copilot, Bing and selected partner experiences. Its AI Performance report includes total citations, cited pages and samples of the grounding queries used to retrieve them. Bing describes those figures as citation activity, not a ranking or a measure of authority, which is the right way to read them.

Google's generative AI performance reports in Search Console provide dedicated views of visibility in AI Overviews, AI Mode and generative features in Discover. Compare that view with the standard Web performance report and the enquiries attributed to organic search.

Search Console also has a "Search generative AI" control that lets a site include or exclude its links and content from AI Overviews, AI Mode and generative features in Discover. Include is the default, and staying included keeps your pages eligible to be cited, so read the generative AI performance report first and decide from your own numbers. Google says the control is not a ranking or inclusion signal for other parts of Search, and it does not affect AI training, which Google-Extended handles.

Google states: "As of August 31, 2026, we've rolled out this control to all websites worldwide." Source: Google Search Console Help, Search generative AI

Google also says that AI features use the same foundational SEO requirements as Search. A page must be indexed and eligible to show with a snippet. Google does not require a special AI file or a special schema type, and structured data must match the visible page.

Those reports cover Google and Microsoft surfaces. Keep the manual prompt set for other systems and for questions where you need to inspect the wording of the answer itself.

Google's AI answers are also giving links more room. A page that is clearly the official source on a subject has more chances to be cited and clicked, so check that your key pages state who you are and what you offer in plain terms.

Google reported "continued and substantial increases to the presence of inline links" in AI Mode and AI Overviews on 15 October 2026, and is testing a section at the top of responses that surfaces official pages. Source: Google, AI Mode and AI Overviews updates

Measure coverage and citation share by market

One total citation number hides the decisions that matter. Group the results by market, topic and stage of the buying journey.

Three measures make a useful monthly view:

1. Prompt coverage: the proportion of tracked questions where the brand appears.
2. Citation coverage: the proportion where at least one page from your domain is cited.
3. Citation share: your citations as a proportion of all citations recorded for the same prompt set.

Report the numerator and denominator alongside each percentage. "Cited for 9 of 30 French prompts" is easier to assess than a bare 30 per cent, especially when the prompt set changes.

Segment the cited URLs as well. A service page citation supports a different decision from a journal article or a third-party profile. The mix shows whether the system understands what the company offers or only recognises it as a source of general information.

Track competitors by domain rather than by how often their name appears in prose. A publication, directory or partner may be cited often without competing for the work. Separating businesses from information sources keeps the comparison honest.

Connect AI visibility to enquiries

Citation growth is useful when it improves discovery and creates qualified visits or enquiries. Add the commercial layer without claiming that every conversion came from one answer.

Ask new contacts how they found the company and include AI assistants as a selectable answer. Keep the original free-text response too, since buyers may say "ChatGPT", "Google's answer" or the name of a cited article. Review referral traffic from AI platforms in analytics and annotate the dates when a cited page changed.

Compare four signals together:

• citations and brand mentions for the tracked prompts;
• visits to the cited landing pages;
• assisted and direct enquiries mentioning an AI tool;
• conversion quality by market and service.

The pattern matters more than a single attribution claim. Repeated citations for German technical SEO prompts, followed by visits to the German service page and relevant enquiries, give the team a reason to expand that topic. Citations with no visits can still build recognition, but they call for a different decision from citations that regularly start a conversation.

Review the cited pages and the gaps around them

The report should lead to page-level work. For every question where a competitor is cited and you are absent, inspect the sources the system selected.

Look for the information those pages make easy to retrieve: a direct answer, a named author, a checked date, original evidence, a clear comparison or a specific market example. Then decide whether an existing page can answer the question better or whether the subject deserves its own page.

Technical access comes first. Confirm that the cited candidate is indexable, linked from the site and available as text. Check the relevant crawler rules and the CDN because an allowed robots.txt file can still sit behind infrastructure that blocks a search crawler. Our technical SEO audit checklist covers the wider checks that support both classic search and AI features.

Keep structured data accurate and visible in the page itself. Schema can clarify the type and entities on a page, but it does not replace a clear answer or create eligibility on its own.

Run the same test on a fixed schedule

AI answers vary, so one test is an observation rather than a trend. Run the same prompt set monthly, using the same language, location and account conditions where possible. Record any deliberate change to the prompts instead of silently replacing them.

Use a small control group of prompts that stays fixed for trend reporting and a second group that can change as new products, competitors and buyer questions appear. The fixed group protects comparability; the changing group keeps the programme commercially relevant.

Review the results in this order:

1. crawler and indexing problems;
2. changes in citation coverage by market;
3. pages gaining or losing citations;
4. competitor sources appearing repeatedly;
5. visits and enquiries connected to those pages.

Generative engine optimization becomes measurable once the same questions, platforms and markets are checked consistently. The goal is a clear chain from buyer question to cited source, from cited source to useful visit and from useful visit to an enquiry your team wants.