All posts
AI search

Google AI Search Links: What SEO Teams Should Measure

Google is making links in AI answers easier to find. Here is what changed and how to measure whether your pages are seen, visited and useful.

STSearchFIT team14 minutes read

Google AI search links are becoming more prominent, but prominence is not the same as traffic or business impact. On 15 October 2026, Google said it was showing inline links higher in AI Mode and AI Overviews more often, while testing a prominent section for official websites. SEO teams should measure source visibility, resulting visits and business outcomes separately, because no single current report connects all three.

The update changes how people can discover sources in Google’s AI answers. It does not establish that every site will receive more clicks, or that a page’s position in an answer guarantees a visit. This article explains what happened, who is affected and what teams can measure now. For a wider view of SearchFIT’s AI search features, see how visibility monitoring can sit alongside the reporting you already use and SearchFIT’s AI search resources.

What Google changed in AI Mode and AI Overviews

In its 15 October announcement, Google said it had increased the number of inline links in its AI experiences and was showing those links higher in responses more often. An inline link is a clickable word or phrase inside an AI-generated answer that takes a person to a source page. The change is about how people encounter and access sources in the answer, not a promise that any particular page will be selected.

Google also said it was testing a prominent section near the top of some responses for one or two official websites. The company describes this as a test, so teams should not assume that the section appears for every query, market, account or device. It is intended to help people find an official source when they need to complete a task or get exact information. Whether it appears for your brand is something to check, not a new universal placement to count on.

The announcement covered more than inline links. Google said it was expanding prominent article carousels to more locations and languages, adding thumbnails to some related-article links at the end of AI responses, and testing clickable creator or publisher names on video results in AI Mode. Each format creates a different way for a source to be discovered. A page linked within an answer, an article surfaced in a carousel and a publisher name attached to a video are not interchangeable impressions of the same thing.

Google also reported that use of its Preferred Sources feature had grown more than sevenfold since the start of 2026. That is a Google-reported adoption figure for the feature. It does not tell marketers how often their particular site appears, how many people click its links or how much traffic it receives. Keep that distinction clear when you report the announcement internally.

This update follows Google’s 6 May 2026 announcement about article suggestions, subscription labels, previews of public discussions and firsthand perspectives, and website previews on desktop. Google’s examples included links to reporting about Seoul’s stream restoration and New York’s High Line in an answer about urban green spaces, and subscription links in an answer about kid-friendly Nashville events. Those examples show that source discovery can happen through several types of links and previews, not only a conventional citation in the body of an answer. Google’s May announcement provides the background to the October changes.

For SEO teams, visibility in Google’s AI experiences is no longer well described by a ranking alone. A page might be selected as a source and linked in an answer. The brand might be named without a link. A page could appear in a carousel or a related-article section, or it could be absent. Those outcomes have different implications for discovery, referral traffic and brand understanding.

A more noticeable link may make a source easier to spot. But its location does not tell you how many people saw it, whether they clicked it or what they did after arriving. The link may answer a question without requiring a visit. Or a person may click through because the answer provides a useful starting point but not enough depth. Measure the observable steps separately instead of treating the presence of a link as proof of a traffic gain.

This distinction matters to different organisations in different ways:

  • Agencies need to explain changes to clients without presenting a handful of screenshots as a complete visibility estimate. A consistent sample and a clear separation between source appearances, visits and outcomes make reporting more useful.
  • B2B SaaS teams need to know whether AI answers surface the right pages for use cases, integrations, comparisons, alternatives and buying questions. A product mention without a link is different from a visit to a relevant product or proof page.
  • E-commerce teams may care whether product information is represented accurately, whether official product pages are surfaced and whether any resulting visits lead to add-to-carts or purchases. An appearance alone does not establish a sale.

The practical question is not simply, ‘Did we get cited?’ It is: which page appeared, in what format, for which kind of question, in which Google surface, and did the visibility connect to a meaningful next step? A useful AEO measurement and ROI guide for B2B SaaS teams can help frame that final question around business outcomes rather than mention counts alone.

AI Overviews and AI Mode are separate measurement surfaces

Google says that AI Overviews and AI Mode can use query fan-out: the system may run multiple related searches across subtopics and data sources to develop a response. Google also says the experiences can use different models and techniques. As a result, the pages linked in one experience may differ from the pages linked in the other. Google’s Search Central guidance on AI features explains these features and the way Google describes their operation.

That is why a spot-check in an AI Overview should not stand in for visibility across all of Google AI Search. The same applies to one AI Mode response. And neither should be treated as a proxy for what people see in ChatGPT, Perplexity, Claude or Gemini. These are separate services with different interfaces and systems. If you report across them, keep their measurements distinct.

Question wording also matters. A person may ask a broad, nuanced question, while the system explores related subtopics behind the scenes. A team’s test prompt is therefore a useful monitoring sample, not a complete record of every search involved in an answer. It can show whether a defined set of important questions is returning a source, but it cannot establish the full universe of queries that might lead to an appearance.

For comparisons to be meaningful, record the surface and conditions each time. At minimum, include the date, exact prompt, market or country, device and whether you checked AI Mode or an AI Overview. If you use a prompt-monitoring platform, make sure those details are visible in exports and reports. A combined score that hides which surface or prompt produced an appearance can obscure the change you are trying to understand.

What Search Console can and cannot tell you

As of 31 August 2026, Google said its Generative AI performance report had rolled out to websites worldwide. The report covers impressions in AI Overviews and AI Mode. Google documents views by page, country, date and device, and notes that the report may not appear for a site that has not received enough impressions in its generative AI features. See Google’s Search Console report documentation for the current details.

This makes the report useful for answering a specific question: are pages on my property receiving impressions in Google’s AI features, and how is that visibility changing across the available dimensions? It is a valuable page-level signal. It is not a complete AI search analytics suite, and its limits should shape the conclusions you draw.

The documented report does not show the query that triggered an AI impression. It also does not provide a dedicated AI-feature click or conversion total. Its chart counts impressions at the property level, so multiple links from the same site in one AI result count as one chart impression. The page table groups results by page. Google also warns that chart and table totals can differ because they use different aggregation. Treat those views as useful but not identical counts.

Clicks from AI answers can still be part of Google’s reporting. Google’s Search Console help says that a click on an external link in an AI Overview or AI Mode counts as a click. Google’s Search Central guidance says traffic from AI features is included in the overall Web search type. That is not the same as a dedicated AI click total, because the current Generative AI report is an impressions report. For AI Overviews, Google says all links in the overview are assigned the same search-result position, so Search Console cannot tell you whether a link appeared higher or lower in the answer. The Search Console explanation of impressions, position and clicks sets out how those metrics are counted.

Measurement questionUseful signalWhat it does not establish
Are pages appearing in Google AI features?Generative AI report impressions by page, country, date and deviceWhich exact query caused the impression or how prominent the link was
Are appearances producing Google clicks?Standard Search Console Web clicks and Google organic sessions in analyticsA separate total of AI Overview or AI Mode clicks
Is the right page being linked and described correctly?Repeated manual checks or a prompt-monitoring workflowA complete measure of all AI visibility or a stable estimate from one check
Is visibility commercially useful?Leads, sign-ups, add-to-carts, purchases and revenue in analytics or a CRMAutomatic attribution from a particular AI appearance to a conversion

Use the report according to what it can show. Do not turn its impression count into a click estimate, or claim that it identifies the exact question behind an appearance. Those would be conclusions the documented data does not support.

1. Define a useful set of questions

Start with questions your customers or prospects actually care about. Use sales calls, support tickets, site search, keyword research and relevant Search Console queries to build the list. Include questions at different stages, rather than testing only broad informational prompts or only product names.

For example, a team might include:

  • How does [product category] work for [specific use case]?
  • What is the difference between [option A] and [option B]?
  • What is the best [category] for [company type or need]?
  • What are alternatives to [brand]?
  • How much does [solution] cost?

Group the questions by audience, intent and funnel stage. For an agency, that might mean questions about service fit and evidence of results. For B2B SaaS, include use cases, integrations and alternatives. For e-commerce, include product comparisons, sizing, materials, price and availability. The result should be a manageable monitoring sample that reflects business priorities.

Be clear that the sample is not a complete list of what people ask Google. Query fan-out means a test prompt does not necessarily match every related search involved in producing an answer. Keep the wording stable for comparisons, but add new questions when customer needs or product lines change. Record why each prompt matters so the team can revisit the sample rather than letting it grow without purpose.

2. Check each surface separately and repeat the checks

For every test, record the date, exact query, market, device and surface. Keep AI Overviews and AI Mode in separate rows or views. If you check other assistants, record each one separately as well. This makes it possible to see whether a change applies to one experience or appears across several.

Capture what happened in a consistent way. Did the answer mention the brand? Did it link to a specific page on your site? Was the link inside the answer, in a carousel or in a related-article section? Which competitors appeared? Did the response describe the product or organisation accurately? A simple yes-or-no citation field loses much of the context that makes an observation actionable.

Repeat checks on a regular schedule and compare trends, not isolated screenshots. Search experiences can change, and a single response is only one observation under one set of conditions. Keep screenshots or result notes if they help your team review what changed, but do not present them as a representative sample of all searches. One SEO practitioner discussion about measuring AI visibility raises questions about citations versus mentions, competitor visibility and how to avoid combining different AI surfaces. Treat it as an example of questions practitioners ask, not as representative measurement data.

3. Use the Search Console report as a page-level review queue

Review the Generative AI performance report for pages gaining or losing impressions, then segment by the dimensions Google provides, such as country, device and date. Use the page list to decide what to inspect next. For example, ask whether the pages gaining impressions are relevant to your target market, whether informational articles appear while product or service pages do not, and whether a change followed a meaningful content or technical update.

Check that the pages you care about can appear in Google Search. Google’s guidance says pages need to be indexed and eligible to appear with a snippet in Search. Review crawl access, internal links and whether important information is available as text. If a key page is missing from your review, first establish whether Google can find and use it before assuming that a content edit will change AI visibility.

Because the documented report does not expose the triggering query, conventional Search Console queries can be clues, not proof. A practical way to investigate is to inspect the conventional queries associated with a page, select relevant question-shaped queries and test those in AI Mode. This narrows the investigation and can help you find gaps. It does not establish that one of those queries caused a particular AI impression.

4. Connect visibility to visits and outcomes carefully

Use your analytics and CRM to follow Google organic sessions, engagement and conversions for pages that appear in the AI report. Compare those trends with Search Console impressions, clicks and the conventional queries associated with the same pages. For an online shop, relevant outcomes may include add-to-carts and purchases. For a SaaS business, they may include sign-ups, qualified leads or demo requests. Choose outcomes that match the way your organisation measures value.

Do not label a rise in organic sessions as AI-generated traffic unless the data identifies it that way. Google’s current documentation places AI-feature activity in the overall Web search reporting, while the dedicated Generative AI report covers impressions. This gives teams signals about visibility and broader traffic, but not a clean, feature-specific path from an AI appearance to a click and then a conversion.

Where possible, compare similar pages or query groups and annotate important changes. Note when a page was updated, when a technical issue was resolved and when Google announced a relevant change. Separate branded from non-branded demand if that distinction matters to your business. These steps help explain what may have changed, but they do not remove all attribution limits or prove that one change caused a movement in traffic.

5. Review whether the source page is useful and clear

Google’s guidance does not introduce a special technical requirement for AI Overviews or AI Mode. It says standard Search fundamentals still apply: pages should be crawlable and discoverable, important information should be available in text, and pages should offer a good experience. Google also says structured data should match visible page content. These are practical checks for an SEO team, not a special recipe for earning an AI link.

At an editorial level, make the page’s purpose clear. Put a direct answer near the top, then provide the detail and evidence a reader needs to assess it. Keep prices, product information, examples and other time-sensitive facts current. Explain who a recommendation suits, define technical terms and make the next step easy to understand. Link related pages together so both readers and search systems can discover useful context.

Do not make changes solely to force a citation. Google says there is no additional AI-specific optimisation or special markup required. Improve accuracy, usefulness, accessibility and clarity because those are sound practices for people and conventional search as well. A clearer page can support users who arrive from any source, even when you cannot attribute a visit to a particular AI answer.

A prominent source link may be easier to notice, but the appearance alone does not show whether the person clicked. Even a click does not tell you whether the visit became a lead, a sale or a useful brand interaction. Keep three reporting layers distinct: visibility in an AI answer, traffic to the site and business outcomes after arrival.

Google has said that overall organic click volume from Search to websites has been relatively stable year over year, and that users who click from pages with AI Overviews are more likely to spend more time on the destination site. Those are Google’s own observations, as described in its 2026 SEO in AI Search guide for marketers. They are not a guarantee for an individual site, market or query group.

Other research can reach different conclusions because studies ask different questions and use different methods. An August 2026 preregistered field experiment with 1,100 participants reported that removing AI Overviews and AI Mode increased clicks to publishers, while an AI Mode-only experience reduced clicks. The study on referral effects from AI in Search describes results under its experimental conditions. It does not predict the impact for every website, query set or market.

The sensible response is to assess your own pages and defined query groups. Where possible, keep a comparison group of similar queries or pages, separate branded from non-branded demand and annotate significant content and Search changes. Report AI impressions as visibility, clicks and sessions as traffic signals, and conversions or revenue as business outcomes. A single blended ‘AI SEO’ number can hide the difference between being seen, being visited and producing value.

What SEO teams should do now

Google’s October update makes links in AI responses more prominent in some ways, and adds or expands other discovery formats. The measurement implications are immediate even though the results will differ by site. Teams should start with a small, documented process rather than wait for a perfect attribution report.

A practical action list:

  1. Check the Search Console report. Confirm whether the Generative AI performance report is available for your property. Record the pages receiving impressions and the dimensions Google provides.
  2. Create a fixed question sample. Choose real questions tied to audience needs, products and buying decisions. Record the reason for each prompt and the surface where you test it.
  3. Track formats separately. Note whether a page is linked inline, shown in a carousel, included in related articles or surfaced in another format. Record brand mentions without links separately from linked sources.
  4. Repeat checks consistently. Keep the market, device, wording and schedule consistent enough to make comparisons. Treat individual results as observations, not a full market estimate.
  5. Connect the page to outcomes. Review relevant organic sessions and conversions in analytics or a CRM, while being explicit that these tools do not identify every AI-driven visit.
  6. Improve the source page for people. Make it accurate, discoverable and clear. Do not add unsupported claims or special markup in the hope of triggering an AI citation.
  7. Explain the limits in reports. State what the data measures and what it cannot establish. This protects client and executive decisions from false certainty.

Teams that need to formalise this process can review SearchFIT’s pricing and decide whether a dedicated monitoring and reporting workflow fits their needs. The most useful next step is to choose the questions that matter to your organisation, establish a baseline and review visibility, traffic and outcomes as three separate signals.

Get SearchFIT Weekly

What we are learning about getting brands cited in AI search, in your inbox every week. Unsubscribe any time.