Check this page with an assistantOpens a chat asking it to summarise this article and name the evidence behind each claim.
Claude opens with the prompt on your clipboard: Anthropic does not support prefilled prompts on the web, and we would rather copy it than ship a button that drops it.
The Numbers, and the One That Matters Most
G2 surveyed 1,076 B2B software buyers in March 2026. Fifty-one percent said they start research with an AI chatbot, up from 29% in April 2025. Seventy-one percent use one somewhere in the process, and 53% now find that research more productive than traditional search, up from 36%.
Adoption is the least interesting part. The number that should change how you think is that 33% purchased from a vendor they were not previously familiar with, and 69% chose a different vendor than they had originally planned, on the strength of what the assistant said.
Those two together describe something search never did at this scale: a shortlist assembled by a third party, before the buyer has visited anyone's website, with enough authority to overturn a pre-existing preference. A buyer who arrives at your site from that process arrives having already been sorted.
The Unknown Vendor Advantage
A third of buyers purchasing from a vendor they had never heard of is the most commercially interesting finding in the study, and it cuts both ways. It means brand recognition is a weaker moat than it was, and it means an unknown company can be placed in front of a buyer by something they trust.
Historically the small competitor's problem was that nobody searching the category knew to look for them. A results page rewards accumulated standing, so the incumbent won the discovery step and the challenger had to buy attention. An assistant assembling a shortlist from what its sources say is a different selection mechanism, and it is one where being well-described matters more than being well-known.
The corollary is uncomfortable for incumbents. If a third of buyers will purchase from an unfamiliar vendor on a recommendation, then being the established name in a category is worth less than being the name that the sources assistants read describe most accurately and most favourably.
Why the Mention Is Worth More Than the Click
Eighty-five percent said they thought more highly of a vendor when a chatbot mentioned it in a recommendation. That reframes what appearing in an answer is worth, because it produces an effect even when nobody clicks anything.
This is the honest counterweight to the zero-click complaint. A mention that produces no visit still moves the buyer, which means measuring the value of AI visibility purely in referral sessions understates it. It also means the absence of traffic from assistants is not evidence that being cited is worthless.
It cuts the other way too. If being named improves perception, being named unfavourably or as an also-ran damages it in the same channel, with the same borrowed authority. That is the shadow metric almost nobody instruments, covered in being mentioned is not being recommended.
What Actually Determines Whether You Are Named
Not your website, mostly. Citation research consistently finds assistants leaning on third-party sources, which means the shortlist is assembled from roundups, review platforms, and community discussion rather than from vendor pages.
| Buyer stage | What the assistant draws on | Where your work goes |
|---|---|---|
| Framing the problem | General explanations of the category | Low leverage; this is the commoditised part |
| Assembling a shortlist | Roundups, review platforms, community threads | Earning third-party citations |
| Comparing named options | Comparison content and reviews | Comparison and alternatives pages |
| Verifying a claim | Your own site, finally | Converting AI-referred visitors |
| Deciding | Pricing, fit, and whether you answered the constraint | Service pages and visible pricing |
Your site enters at stage four, to confirm a claim the assistant already made. That is a narrower job than most sites are built for, and it is why a page that opens by explaining the category loses a visitor who was sent there to check a specific assertion.
What This Does Not Mean
It does not mean search is over, that you should reallocate a budget on the strength of one survey, or that these percentages apply to your market. The study covers B2B software buyers, and transferring its numbers to a plumbing business would be exactly the overreach this blog keeps arguing against.
It is also a survey, which measures what people say they do. Stated behaviour and actual behaviour diverge, particularly on questions about new technology where there is mild social pressure to report adoption. The direction is corroborated by other evidence; the precise figures deserve the usual scepticism.
The proportionate response is to find out whether it is true for your buyers rather than assuming either way. Run the questions your customers actually ask through a few assistants, repeatedly, and see whether you are named. That is a measurement you can take this afternoon, and the method is in share of model.
Questions About the AI Buying Journey
- How many buyers actually start with an AI chatbot?
G2 surveyed 1,076 B2B software buyers in March 2026 and found 51% starting their research with an AI chatbot, up from 29% in April 2025. A larger share, 71%, use chatbots somewhere in the research process. The starting point is the number that matters, because it determines who gets into the consideration set at all.
- Does the AI recommendation actually change what people buy?
According to the same research, substantially. Sixty-nine percent said they chose a different vendor than they had originally planned based on chatbot guidance, and a third bought from a vendor they were not previously familiar with. That is a demand channel reshaping outcomes rather than merely informing them.
- Why does being mentioned matter beyond the click?
Because the mention itself changes perception. In that survey, 85% said they thought more highly of a vendor when a chatbot mentioned it in a recommendation. Being named functions as an endorsement from a source buyers treat as neutral, which is a different thing from appearing in a list of search results.
- Is this only a B2B software phenomenon?
The study covers B2B software buyers specifically, so that is the population the numbers describe. Similar behaviour is widely reported in other considered purchases, but you should not transfer a percentage from this survey to a different market. Use it as evidence about software buying and as a directional signal elsewhere.
- What should a small business actually do about this?
Find out whether assistants name you on the questions your buyers ask, and fix the reason if they do not. That is usually presence on the third-party sources assistants retrieve rather than anything on your own site, which makes it an outreach and reputation problem rather than a content production one.
