There is a stat making the rounds that says half of B2B buyers now start their buying journey inside an AI chatbot. It shows up in webinars, sales decks, and LinkedIn posts, usually right before somebody tries to sell you an AI visibility package. We almost quoted it ourselves.
Then we checked. In the studies that actually measured it, the share of buyer journeys that begin inside a chatbot is in the low single digits. Not half. Not a quarter. Low single digits.
We rebuilt our own website this summer, and before we made a single decision we ran the popular AI marketing stats through a fact-check: pull the original source, see what it really measured, and throw out anything that could not stand on its own. Here is what fell apart, what held, and what it means if you sell expertise for a living.
Which AI stats failed the check?
“Half of B2B buyers start in a chatbot.” The studies that measured actual starting points put chatbots at the front of the journey a few percent of the time. Buyers lean on AI heavily in the middle of their research, but the “half start there” version is a game of telephone played with a much smaller number.
“AI visitors are 4.4x more valuable.” This one traces back to a single vendor dataset and gets repeated as if it were a law of nature. The premium is real in some datasets and much smaller in others, and honest people say so. Anyone quoting one universal multiple is selling you something.
“AI chat is the number one place buyers build their shortlist.” Stated as market fact, it did not verify. It may get there. It is not there.
Which stats held up?
AI research is nearly universal. In G2’s buyer research, 87 percent of B2B buyers said AI chatbots have changed how they research. Forrester puts AI use somewhere in the buying process at 94 percent. Buyers are not starting in a chatbot, but almost all of them are running AI somewhere in the journey.
AI sends fewer clicks than search. A lot fewer. Ahrefs modeled roughly 75 percent fewer clicks from AI interfaces than from classic search. Cloudflare measured the crawl side: Anthropic crawled about 38,000 pages for every visitor it referred, OpenAI about 1,100, Perplexity about 195. Google, for comparison, sits around 5. The machines read everything and send back a trickle.
The trickle is serious. On Ahrefs’ own site, AI referrals were half a percent of traffic but 12 percent of signups. That is one company’s data and the multiple will be smaller for most businesses; ecommerce datasets land closer to 1.3 to 1.5x. But the direction held everywhere we looked: fewer visitors, higher intent.
Plain writing wins. Unbounce found pages written at a 5th to 7th grade reading level converted at 11.1 percent versus 5.3 percent for professional-register copy. Not an AI stat, but it survived the same check, and it says a lot about which content the engines end up quoting.
What do the real numbers add up to?
A picture that is more interesting than the hype. Your buyers research with AI at nearly every step. The AI engines read your entire public body of work and send you a fraction of the traffic search used to. And the people who do click through arrive already convinced, because a machine they trust handed them your name.
That changes the job. Getting cited by the engines is the new getting found, and it goes to whoever publishes clear, direct answers to the questions buyers actually ask. Your website stops being a brochure for cold traffic and starts being the closing room for warm traffic. Both of those are things you can build for on purpose.
Why are we telling you this?
Because we sell in this category, and the fastest way to lose your trust is to pitch you with numbers that fall apart the moment you tug on them. The shift to AI-led buying is real. It does not need inflated stats to matter, and we would rather build on numbers that hold.
If you want to know where your own brand stands while everyone else argues about percentages, start with a straight read. The 5-Minute Authority Audit scores your offer, message, content, funnel, and proof, and shows you the one fix to make first.
