Why customers hate support chatbots: what 173 one-star reviews and Gartner's data say
The complaints are mostly about unreachable humans and surprise bills, not wrong answers. A review-by-review look, with a six-point checklist for any vendor.
We make an AI support product, and before building it we read what angry people write about the existing ones. We expected "the bot gave a wrong answer" to be the main complaint. It was not.
What we read
We hand-coded 173 one-star Trustpilot reviews of 13 chat and support vendors, written between January 2025 and November 2026 (Intercom, Tidio, Zendesk, tawk.to, Crisp, LiveChat, Help Scout, Chatbase and others). Each review could carry more than one complaint.
Limits first. Trustpilot shows about 20 recent one-star reviews per vendor, and one-star reviews over-represent billing disputes. So the counts below rank the complaints. They are not rates across all customers.
| Complaint | Share of reviews | Vendors affected (of 13) |
|---|---|---|
| Vendor support slow, unreachable or scripted | about 40% | 11 |
| Billing and pricing | about 38% | 13 |
| Reliability and bugs | about 20% | not recorded |
| Setup and UX | about 18% | not recorded |
| Cancellation and auto-renewal | about 17% | not recorded |
| AI answer quality | about 13% | not recorded |
Sources: Trustpilot one-star pages for Intercom, Tidio, Zendesk, tawk.to and others, read 20 Nov 2026.
The surprising part: a good share of the 13% "AI quality" reviews are about the vendor's own support bot blocking a paying customer from reaching a person.
Complaint 1: "I cannot reach anyone"
The largest group is not about chatbots on websites. It is about the company that sold the chatbot. Reviewers describe paying for a support tool and then being unable to get support from it.
This matters for anyone shopping for a tool, because the buyer is the one who ends up stuck. It is also the same failure the shopper's own customers meet when your bot offers no way out.
Gartner's 2026 survey of 3,566 customers puts a number on that second group:
- 87% say companies using generative AI for service must provide access to a human (Gartner, Aug 2026).
- Only 27% would try a chatbot again after a bad experience (Gartner, Sep 2026).
- They were three times more likely to use a third-party tool such as ChatGPT than the company's own bot in their latest service interaction (Gartner, Jul 2026).
We saw these figures through Gartner's press releases. We did not buy the full reports.
Complaint 2: "The handoff made me start over"
Even when a bot does pass you to a person, the person often starts cold. Five9 reports that 83% of consumers re-explain their issue after an AI-to-human transfer, while 96% of business decision-makers believe their handoffs keep the context (Five9). Five9 sells contact-centre software, so read it as a vendor survey. The gap is still the point: the business thinks it is fine and the customer disagrees.
Complaint 3: "My bill doubled"
Billing shows up for all 13 vendors. The reviews are specific:
- Tidio: "They doubled my rate without notifying me", and a reviewer whose $150 became $500 a year with "0 notifications" (Trustpilot).
- JivoChat raised prices 250 to 300% for existing users (Trustpilot).
- A Help Scout reviewer went "from about $1,500 to $7,500+" a year (Trustpilot).
- A LiveChat reviewer reports an $805 charge "without my consent" (Trustpilot).
These are individual reviews, and we have not verified any of them with the vendors. They are useful as a map of what customers feel they were not told.
Per-resolution pricing adds a new kind of argument: who decides that a chat was "resolved". In an Intercom community thread a customer reported a real resolve rate of about 12% while being billed for "assumed resolutions", even when a human stepped in after a wrong answer. Intercom staff later wrote "we didn't communicate the billing logic accurately" (Intercom Community). An audit by a competitor found 9 of 16 conversations marked "resolved by Fin" had not been solved (Beri); that is a small sample from a source with its own product, so we treat it as a warning, not a measurement.
Complaint 4: "The bot made something up"
This is the smallest group in our sample, but the costliest when it happens. A tribunal held Air Canada liable for its chatbot's wrong bereavement-fare answer (Moffatt v. Air Canada, 2024 BCCRT 149). Cursor's support agent blamed a policy that did not exist for what was really a bug, and users cancelled (Help Scout).
The law is catching up
- EU AI Act, Article 50: since 2 August 2026, chatbots must disclose they are AI at the first interaction (Goodwin).
- California AB 1609: signed 28 September 2026, in force 1 January 2027. It bans presenting a bot as human, and for businesses over $500 million in revenue it requires a live agent within 15 minutes in business hours or an appointment within one business day (Quarles).
This is not legal advice. Neither law requires that the human sees what the customer already said, which is the handoff complaint above.
What to demand from any vendor
- A human route on every turn, not hidden behind three failed attempts.
- Context carried over. Test it: start a chat, hand off, and see whether the human has to ask again.
- A billing definition you can read, and evidence for each charge you can inspect.
- A spend cap that is on by default, so a launch day does not become a bill.
- A way to cancel in the product.
- A reply from the vendor's own support. Write to them before you buy.
What we are doing about it
We are building ChatSupportHQ around these six points, and we are not live yet, so we cannot show you results. What we have committed to publicly: a visible human route on every turn, an escalation offer after two failed turns, a charge only with evidence (visitor confirmation, the merchant's own system, or 72 hours without contact on any channel), and a hard monthly cap on by default. When we have real data from our own four sites we will publish it here, including the bad parts.
FAQ
Why do customers dislike chatbots? In our review sample the main drivers were unreachable humans and billing, not wrong answers. In Gartner's survey, 87% want a human option and only 27% would retry a chatbot after a bad experience.
Do customers prefer talking to a human? When AI is in the way, most say they want the option of one. They do not say they never want AI.
How should a chatbot hand off to a person? Visibly, early, and with the conversation attached so the customer does not repeat it.
Method and sources
173 one-star Trustpilot reviews, Jan 2025 to Nov 2026, 13 vendors, hand-coded by our research team, with multiple labels per review. Counts rank complaints; they do not measure prevalence. Gartner and Five9 figures are as published by those firms. Written 21 Nov 2026; we will correct anything we got wrong and mark the change.
ChatSupportHQ is a product of DishIs Technologies LLC.