Do B2B buyers actually use AI search to choose vendors?

B2B buyers do use AI search to choose vendors, and the share is now high enough that the interesting question is what they use it for rather than whether they use it. Forrester's 2026 analysis of its Buyers' Journey Survey reported that 94% of B2B buyers used generative AI somewhere in their most recent purchase process, up from 89% the previous year, in a survey of roughly 18,000 business buyers globally.

Forrester's Buyers' Journey Survey, 2025 found that twice as many buyers named generative AI or conversational search a more meaningful source of information than named any other single source, ahead of vendor websites, product experts and salespeople. That is a ranking of self-reported usefulness rather than a measurement of behaviour, and it is worth holding both facts in mind: buyers say the AI answer mattered more, and they are not always reliable narrators of their own research.

What makes the number consequential for marketing is where in the process it sits. The work buyers describe doing with AI, comparing options and assembling a case, happens before they identify themselves. A company that is absent from those answers is not losing a deal at the end of the funnel. It is not entering the consideration set that the rest of the funnel is built on.

None of this means classic search stopped mattering, and the honest framing is additive rather than apocalyptic. Buyers still read vendor sites, still ask peers, still shortlist from review platforms. The change is that an additional step now sits in front of all of those, and that step summarises you using material you may not have written.

What do B2B buyers use AI for at each stage of a purchase?

B2B buyers use AI for different jobs at different stages, and the reported breakdown matters because each job rewards a different kind of page. The figures below are from Forrester's 2026 analysis of its Buyers' Journey Survey, covering the 94% of surveyed buyers who used generative AI during their most recent purchase.

What buyers used AI for	Share of AI-using buyers	What that rewards on your site
Comparing vendors against each other	55%	Honest comparison pages that describe the alternatives accurately
Researching products and capabilities	54%	Specific, factual pages: documentation, integrations, limits
Building the internal business case	47%	Quotable figures with named sources, and clear scoping

Read the top row first, because it is the one most B2B sites are least equipped for. A buyer asking an engine to compare four tools is asking for a judgement, and the engine assembles that judgement from whatever it retrieved. Pages that only argue for one vendor contribute little to a comparison, which is why they tend not to be quoted inside one.

The second row is the easiest to win and the most often neglected. Questions about whether a product does a specific thing have factual answers that exist on the vendor's own site or nowhere, and Grow and Convert's 2026 citation study found 86% of the references the models used came from industry-specific domains, often the vendors' own blogs.

The third row changes what a marketing page is for. When a buyer uses an engine to draft the case they will take to a budget holder, the sentences that survive into that document are the ones carrying a number and a source. A page full of adjectives contributes nothing to it.

What if your buyers say they do not use ChatGPT?

Buyers saying they do not use ChatGPT is weak evidence either way, because the behaviour increasingly does not require a deliberate visit to an AI tool. Google's AI Overviews answer queries in the results page, Microsoft Copilot sits inside the Office applications your buyers already have open, and in-product assistants summarise vendor options without anyone describing it as using AI search.

Ask the question differently and the answers change. Instead of asking whether somebody used ChatGPT, ask how they built their shortlist, where the names on it came from, and whether anything summarised the options for them. Buyers who would deny using AI search frequently describe reading a generated summary in the same conversation.

There is also a sampling problem in asking at all. The people you can ask are the people who contacted you, which by definition excludes everyone who read an answer that did not name you. Research interviews with won and lost deals systematically undercount the buyers who never arrived, and that group is exactly the one an AI visibility programme is aimed at.

The useful version of this check is not a survey. Take the ten questions a buyer would type at the start of a purchase in your category, run them, and look at which companies get named and which pages the answer used. That tells you what your prospective buyers would see without asking anyone to introspect about their own habits.

How do you check whether your own buyers use AI search?

Checking whether your own buyers use AI search takes three sources of evidence, and none of them requires a vendor. The first is your own analytics, where referrals from ChatGPT, Perplexity, Copilot and Google AI surfaces show up as a small and usually growing slice of traffic; our guide to tracking leads from ChatGPT covers the attribution gaps, of which there are several.

The second is your sales calls. Adding one question to discovery, about how the buyer came to shortlist you, surfaces AI-assisted research faster than any dashboard, and it does it with the only population whose opinion you can verify. Record the answer in the CRM as a field rather than in call notes, so it becomes countable by the time somebody asks for evidence.

The third is the direct test, and it is the only one that sees the buyers you never met. Run your category's buying questions across the engines and read the answers as a prospect would: who is named, in what order, and from which sources. A company that appears in none of them has a measurable absence rather than a suspicion.

Expect the three sources to disagree, and expect the referral numbers to be the least informative of them. Most AI-assisted research produces no referral at all, because the buyer reads the answer and then searches for the winner by name, which arrives in your analytics as branded search. That is one reason traffic is a poor proxy for AI visibility, as our guide to whether AI search sends traffic works through in detail.

Is AEO worth it if your category has low AI search volume?

AEO can be worth it in a low-volume category and the arithmetic is different from classic search, because a handful of answers can carry a disproportionate share of a pipeline. A category with twenty serious buyers a year does not need thousands of impressions. It needs to be named in the answers those twenty people read, and in a B2B deal with a six-figure contract value one of them covers a year of work.

The variables that decide it are deal size, the length of the buying committee's research phase, and who currently owns the answer. A long committee-based purchase involves more research steps and therefore more chances for an engine to shape the shortlist. A small, fast, self-serve purchase is less exposed, because the buyer is often acting on a recommendation they already had.

The strongest argument against starting is a category where the answers are genuinely unformed, meaning the engines return vague or hedged responses and name nobody consistently. That is a thin market, not an opportunity, and the work there is usually demand creation rather than retrieval. The strongest argument for starting is the opposite finding: the engines answer confidently, name three companies every time, and you are not one of them.

What tips the decision in practice is how long the work takes to show. Our guide to how long AEO takes to work sets out the timeline, and the relevant part here is that a category where competitors are already named takes longer to break into than one where nobody is, which argues for finding out early rather than waiting for the volume to be obvious.

If you want the answer for your own category rather than in general, a free visibility assessment runs your buyer questions across the four main engines and reports who gets named today, which is the input this decision actually needs.

What happens if you wait a year to start?

Waiting a year costs you the compounding half of the work rather than a position you could buy back later. The structural fixes, crawlability and rendering and vocabulary, can be done whenever you like and take effect quickly. The part that cannot be caught up is coverage: the stock of pages answering questions your category gets asked, which takes as long to build in 2027 as it would have taken in 2026.

The other cost is that citation is a flow rather than a stock, so the companies currently winning are not banking a permanent advantage either. Trakkr's "The Half-Life of AI Citations", a longitudinal study of 857,000 daily reports covering 10,991 brands across eight AI models over ten months, found that 73.5% of the 108,650 citation URLs it tracked appeared exactly once and did not return, and that the median brand fell to half its peak citation count in 31 days.

That finding cuts both ways and it is worth being precise about which way. It means a competitor's current lead is less durable than it looks, so starting later is not hopeless. It also means nobody holds a result by doing nothing, so the entry price is a publishing and measurement habit rather than a one-off project.

The clearest signal that waiting is expensive is a category where buyers ask shortlist questions. Shortlists are winner-take-most: a buyer reading three names does not usually go looking for a fourth, and displacing one of those three is harder than appearing in a question nobody has answered yet.

A reasonable middle path exists and it is cheaper than a programme. Measure now and fix nothing, so that you have a dated baseline and a list of the questions you lose, then decide with a year of context rather than a year of guessing. The measurement is the part that cannot be reconstructed retroactively.

What else do people ask about B2B buyers and AI search?