Abstract illustration of search results being summarized into a single AI-generated answer

AI Search for B2B Marketing: What to Publish and Measure When Clicks Fall

AI Overviews and AI Mode answer buyer questions before a click happens. Here is what the evidence says, what Google and OpenAI document, and how to publish and measure for it.

By Shayan Salehi 8 min read

Image: ColdAI

AI search changes B2B discovery by answering the buyer's question on the results page, so fewer people click and more of them meet your brand inside a summary. Pew Research Center found that users clicked a result link in 8% of visits with an AI summary, against 15% without one. The response is to publish material worth citing, keep it crawlable, and measure influence beyond clicks.

What actually changes for B2B discovery

For a decade, B2B marketing treated search as a click funnel. A buyer typed a query, scanned ten links, opened three, and landed on your site where your forms and nurture sequences took over. AI Overviews and AI Mode compress the first two steps into one generated answer with a handful of links attached.

The change matters most for the long, specific questions that B2B buyers ask. Pew's analysis found that 8% of one- or two-word Google searches produced an AI summary, while 53% of searches with ten or more words did. Pew studied U.S. adults rather than business buyers, so treat that as a directional signal. Still, a query such as "how do we compare vendors for supplier risk monitoring when procurement and security both sign off" is exactly the kind of long search that now gets a synthesized answer.

Two consequences follow. First, your page can influence a purchase decision without receiving a visit. Second, the pages that get cited are chosen for how well they answer a specific sub-question, not for how well they rank for a head term.

What the click data says

Pew Research Center's analysis covered the browsing activity of 900 U.S. adults in March 2025 and 68,879 unique Google searches, of which 12,593 produced an AI summary. Searches with a summary led to a click on a traditional result link in 8% of visits. Searches without one led to a click in 15% of visits.

The summary itself did not make up the difference. Pew reported that users clicked a link inside the AI summary in just 1% of visits to pages that showed one. Users also ended their browsing session more often after seeing a summary: 26% of pages with a summary, compared with 16% of pages without.

Pew also looked at what gets cited. Wikipedia, YouTube and Reddit together made up 15% of the sources listed in AI summaries, and government sites accounted for 6% of linked sources in summaries against 2% in standard results. Most summaries (88%) cited three or more sources, so a summary is a shortlist of references, not a single winner.

Gartner's February 2024 press release predicted that traditional search engine volume would fall 25% by 2026 as AI chatbots and other virtual agents take over some queries. That is a forecast, not a measurement, and your own Search Console data is the better test of whether it applies to your audience. Plan for a lower click yield per impression either way.

How Google chooses what to cite

Google's Search Central documentation on AI features and your website is unusually direct. It states that the best practices for SEO remain relevant for AI Overviews and AI Mode, and that there are no additional requirements to appear in them. To be eligible as a supporting link, a page must be indexed and eligible to be shown in Google Search with a snippet.

The documentation also explains that both features may use a technique it calls query fan-out, which issues multiple related searches across subtopics and data sources to build a response. In practice, one buyer question can pull in pages that would never rank for the original wording. A page that answers one narrow sub-question well, such as data residency options or implementation timelines, can be cited even when it has no chance on the broad head term.

Google adds that AI Overviews are shown only when its systems judge them additive to classic Search, and that they often do not trigger. It also says you do not need new machine-readable files, AI text files, or special schema.org markup. Structured data should match the visible text on the page, which is a standing rule for any markup.

Be skeptical of any vendor that sells a special format for AI visibility. The documentation says no such format is required for Google.

Answer engines beyond Google

Buyers also use assistants that run their own retrieval. OpenAI's crawler documentation describes separate user agents with separate jobs. OAI-SearchBot surfaces sites in ChatGPT's search features, GPTBot crawls content that may be used for training, and ChatGPT-User fetches pages when a user asks. OpenAI states that each robots.txt setting is independent, and that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers, although they can still appear as navigational links.

That gives you a concrete decision to make deliberately. A single rule that blocks every AI crawler also blocks the search agent. If you want to be cited, allow the search agent. If you object to training use, address the training agent separately.

For Google, the controls work differently. Google's documentation points to nosnippet, data-nosnippet, max-snippet and noindex as the way to limit what Search shows, and Google-Extended for limiting training and grounding in some other Google systems. Its May 2025 guidance adds that more restrictive preview permissions will limit how your content is featured in its AI experiences. Review your snippet settings alongside your crawler rules, since a restrictive max-snippet value can quietly reduce your eligibility.

What to publish

Google's guidance for AI search points back to its people-first content principles. In its May 2025 post on succeeding in AI search, Google tells site owners to focus on unique, non-commodity content, and notes that users in AI experiences ask longer and more specific questions along with follow-ups. Its self-assessment questions ask whether content provides original information, reporting, research or analysis, goes beyond the obvious, and demonstrates first-hand expertise.

Applied to B2B, that suggests a content plan built around evidence a summarizer cannot invent:

  • Original data and methods. Publish benchmarks, survey results, or measurement approaches with the method stated, since a cited number needs a source behind it.
  • Plain definitions and decision criteria. A page that defines a term, states when each option fits, and lists the trade-offs is easy to cite for comparison queries.
  • Constraints and edge cases. Buyers ask about integration limits, regulatory scope, and failure modes. Pages that state them honestly answer the questions summaries struggle with.
  • Named authors with first-hand experience, whose pages show how the conclusion was reached.
  • Answer-first structure. Open each section with the direct answer, then give the evidence. This also helps human readers who skim.

Avoid producing dozens of thin pages that restate what already exists. Google's questions explicitly ask whether content copies or rewrites other sources without adding substantial value, and AI summaries make commodity pages even less necessary.

Keep the technical basics in order as well. Google lists allowing crawling in robots.txt and by any CDN or hosting infrastructure, making content findable through internal links, keeping important content in text form, and supporting text with quality images and video.

What to measure when clicks fall

Start with what Google tells you it reports. Its documentation says sites appearing in AI Overviews and AI Mode are included in the overall search traffic in Search Console, in the Performance report under the Web search type. It does not describe a separate AI report, so you cannot isolate AI-driven impressions there. Google also says clicks from pages with AI Overviews tend to be higher quality, meaning users are more likely to spend more time on the site. That is Google's observation, so test it against your own conversion data.

Because the click is no longer the unit of value, build a measurement set that does not depend on it:

  1. Impressions and click-through rate by query length. Split queries into short and long. If impressions hold while click-through rate falls on long queries, you are probably being shown but not visited.
  2. Branded search volume and direct visits. A buyer who sees your name in a summary may search for you next, so track these as lagging indicators of influence.
  3. Conversion rate and engagement per landing page. Fewer clicks that convert at a higher rate can leave pipeline unchanged.
  4. A fixed prompt set. Choose 30 to 50 buyer questions, run them monthly across the assistants your buyers use, and record whether you are cited, how you are described, and which competitors appear. The method is manual and noisy, so treat movements as signals, not as precise rankings.
  5. Self-reported attribution. Add an open-text "how did you first hear about us" field to forms and ask in discovery calls.

Do not report a single blended score. Keep each measure separate, with its method written down, so changes in the assistants' behavior do not get mistaken for changes in your performance.

Where to start

Start with an audit rather than a rebuild. Most of the work is about knowing where you stand and removing blockers.

  1. Check indexing and snippet eligibility for your 20 most important pages, using Search Console and the URL Inspection tool Google points to in its documentation.
  2. Review robots.txt and your CDN rules for OAI-SearchBot and other search agents, and decide separately about training agents.
  3. Pull 12 months of Search Console data and segment by query length, so you have a baseline before the next change.
  4. Build the fixed prompt set from sales calls and support tickets, and run the first round of checks.
  5. Pick five pages where you hold real first-hand evidence, and rewrite them answer-first, with the method and the limits stated.

Treat this as a measurement problem before a content problem. A baseline you trust lets you judge each later change on its merits, and it keeps the team from chasing forecasts, including the Gartner one, in place of its own data.

Tags:ai-searchai-overviewsai-modeb2b-marketinganswer-enginessearch-consolecontent-strategymeasurement

Your starting point

What would you like to move forward?

Start with the outcome that feels closest. We'll help give the next step a useful shape.

Let's find your next move

One conversation. A clearer direction.

Tell us what you want to improve. A few sentences are enough to start.

shayan@coldai.org