The report The findings Ask a B2B chatbot for a person and it asks for your details first We asked 88 of them for a human and 4 produced one. The filed transcripts show bots naming the number of questions they would ask, then exceeding it Drift's customers did not switch, they quit 164 of the 319 study sites running Drift in 2021 had no chat at all by 2025. They were not abandoning a channel, they were closing an account on something nobody had answered Five years of AI hype, and B2B websites went backwards The share of B2B sites offering a buyer no way to ask anything went up across the window that produced modern language models Speed to Lead is timing a robot A metric that times every response is set by the machine, because the machine sends almost all of them The chat vendors caused their own demise An install was sold, the staffing decision was left for later, and five years on it has not been made The identification tax Buyers do not refuse people. They refuse being handled Speed and personalization did not separate the humans from the robots One signal told the two classes apart, and it was not tone, speed or a named sender. It was leaving the form They changed the chat vendor. They did not change who answers. 257 of the 585 sites identified in both waves changed supplier, 43.9% and a ceiling. Almost none of them changed who answers Method I run GTM Clarity, which sells chat software with a human handoff into this market, priced per conversion. So I have a commercial interest in what this research says. GTM Clarity has been running this research since 2021. Charts and figures may be reproduced with attribution and a link.