What the research says about AI chat on a business website

A buyer shortlisting a firm like yours now often starts in an AI chatbot, not a search engine. Then they arrive on your website, where a chat window offers to book a meeting. The two experiences do not match, and the research on both points the same way.

This is what we found, where the evidence is thin, and what it suggests a website assistant should do.

Buyers are already asking AI about you

G2's April 2026 research on software buyers found that 51% now start their research in an AI chatbot, up from 29% a year earlier. 69% said chatbot guidance made them choose a different vendor than they had planned, and 33% bought from a vendor they had never heard of.

That study covers software buyers, and we found no equivalent for consulting or services. Treat it as a signal about direction, not a measurement of your market. It still changes the question. If the first answer about your firm comes from someone else's chatbot, the one on your own site has to be better informed than that.

What most website chat does

The commercial tools describe themselves plainly. A 2026 comparison of B2B chat platforms from Simple Machines describes the best ones as working "like a tireless SDR": they spot high-intent visitors, ask qualifying questions, capture details and route the lead to a person. For consulting firms, the pitch is scope intake that filters for fit.

That is a lead form with a friendlier face. It serves the firm first, and visitors can tell.

Why visitors dislike it

California Management Review summarized a Gartner survey of 5,728 customers in April 2026. 64% preferred that companies not use AI for customer service, 53% would consider switching to a competitor that did not, and only 14% of issues were fully resolved in self-service.

The article lists what drives the frustration: bots that misunderstand the request, fail at complex problems, hand off badly to people, pretend to be human, or ignore who is asking. The survey is about customer service, not sales conversations. The expectations carry over, though, because a visitor judging your firm meets the same bot with the same patience.

The free tools are generic

An executive who wants a view of their own problem before a first call has two options. One is a generic score. Microsoft's AI readiness assessment, for example, is a 45-minute multiple-choice survey. The other is a project: tailored assessments are paid, and Augment Code's 2026 roundup of 17 tools puts the paid range at $15,000 to $500,000.

In our search we found nothing in between. A short, tailored, cited piece of thinking about the visitor's own situation looks like a gap, and a good job for an assistant.

Five things the research points to

Ask before you answer. Misunderstanding comes first on the list of complaints. Two good questions beat a fluent guess.
Show your sources. In G2's research, citations from review sites were the single most confidence-inspiring signal in a chatbot's recommendation, named by 45% of buyers. The same instinct applies to your own answers: link the page each one comes from.
Say it is an AI, and keep the human close. Do not pretend, and make the route to a person short and visible. The handoff should carry the whole conversation.
Give something usable first. A worked exhibit, a short brief or a list of questions to send back all help the visitor even if they never contact you. Ask for an email last.
Be honest about limits. Mark which numbers are the visitor's, which are your assumptions, and what you do not know yet.

What this evidence cannot tell you

Some of the figures here come from vendor reports and one from a vendor blog. Our own reading is that the direction is consistent across all of them, but none of them measures a consulting firm's website. That is why the next step is to measure on your own visitors, starting from a baseline, instead of relying on anyone's benchmark. We set out how in how to build and measure an AI assistant for your website.

If you are weighing an assistant for your own site, here is how we approach AI development.