The short answer.
An AI chatbot can help a B2B website, but only in a narrow role: answering clear questions from your own approved content, collecting context, and routing a buyer to the right person fast. Outside that role it is more likely to cost you enquiries than win them.
That conclusion comes from three bodies of evidence. Surveys show that many people do not yet trust AI or the companies using it. Independent tests show that even the best language models still add unsupported claims when summarising documents. And courts and regulators now treat what a chatbot says as what the company says. None of this means “never use one”. It means scope it tightly and design the exit to a human first.
| Situation | Likely to help | Likely to hurt |
|---|---|---|
| Question type | Factual and repeatable: “Do you integrate with Xero?”, “Which regions do you cover?” | Judgement calls: pricing for a specific deal, legal or contract terms, fit for an unusual case |
| Source of answers | Your own approved pages and documents only | The open web or a general model’s memory |
| Timing | Outside office hours, or when the team cannot reply within the hour | When a person could have replied just as fast |
| Buyer stage | Early research and routing to the right specialist | Late stage, when the buyer wants a named person |
| Data collected | Just enough to route: company, need, timescale | Personal or sensitive details before any value is given |
| Exit | A visible “talk to a person” at every step | Loops, dead ends, or a form hidden behind the bot |
If you are weighing a chatbot because enquiries are low, check the basics first. Our conversion rate optimisation work usually starts with the page and the form, because a chatbot cannot fix an unclear offer. It also helps to know whether traffic or conversion is the constraint before adding a tool.
What a B2B buyer wants from your site.
A B2B buyer on your website usually wants three things: to confirm you do what they need, to see evidence you have done it before, and to reach a competent person without friction. A chatbot is only useful if it makes one of those faster.
Consider who is visiting. A finance director checking whether you support their accounting system wants a straight yes or no with a link. An operations lead with an unusual requirement wants to describe it once and know a specialist will read it. A procurement manager wants your security documents. Each needs something different from a chat window, and two of the three are better served by a clear page and a short form.
Usability research has said this for years. The Nielsen Norman Group’s study of chatbots found users saw them as less helpful than humans but faster, and that bots which duplicate what a website already does rarely justify themselves. Its advice holds up: “Be upfront about using a bot and not a human,” “Offer an escape hatch in the form of a real human, a phone number, or a link to a different interaction channel,” and “Don’t be overly ambitious: create bots for simple tasks.” That study is from 2018 and small (8 participants), so treat it as design guidance rather than proof. But every newer source in this guide points the same way.
Before adding anything new, our guide to auditing a B2B landing page for trust covers the evidence and objections a buyer looks for. If those are missing, a chatbot will spend its time answering questions the page should have answered.
The trust gap, in numbers.
Large surveys in 2024 and 2025 show that trust in AI, and in companies using it, is low or falling. A chatbot starts every conversation with that deficit.
The detail behind each bar is worth knowing:
- KPMG and the University of Melbourne surveyed more than 48,000 people across 47 countries (fieldwork November 2024 to January 2025) and found only 46% willing to trust AI systems.
- Pew Research Center surveyed 5,410 US adults in August 2024: 59% had not too much or no confidence in US companies to develop and use AI responsibly, and 55% wanted more control over how AI is used in their lives.
- The Stanford AI Index 2025, using Ipsos data, reported that confidence that AI companies protect personal data fell from 50% in 2023 to 47% in 2024.
- Salesforce, which sells AI customer-service software, reports that 72% of customers say it is important to know if they are communicating with an AI agent, and 64% believe companies are reckless with customer data.
For a B2B site, the practical meaning is simple. Say clearly that it is an AI assistant, say what it can and cannot do, and make the human route obvious. Pretending the bot is a person, or making people work to escape it, spends trust you have not yet earned.
When a chatbot helps.
A chatbot helps when it removes waiting or searching for a buyer who already knows what they want to ask. The strongest cases are out-of-hours questions, routing, and helping your own team answer faster.
Out-of-hours answers to factual questions
If buyers in other time zones arrive when your team is offline, an assistant that answers “Do you support SAP Business One?” from your approved integration page, links to the page, and offers to pass on a message is useful. It does not need to sell; it needs to be right and to say when it does not know.
Routing to the right person
Many B2B firms have several specialisms. A short guided exchange (“Which of these best describes your need?”) that sends the buyer’s context to the right specialist, with an honest reply time, can beat a generic form. Speed of the human follow-up still decides the outcome; our guide on the first reply after a form covers what the response-time research says.
Helping the people who answer
The strongest evidence for AI in customer conversations is about helping staff, not replacing them. A study by Erik Brynjolfsson, Danielle Li and Lindsey Raymond of 5,179 customer-support agents found that an AI assistant raised issues resolved per hour by 14% on average, and by 34% for novice and lower-skilled agents, with “minimal impact” on experienced ones. It also improved customer sentiment.
Note what was tested: AI assisting a human agent, who stayed in the conversation. For many B2B companies that is the better first project. Our framework on what to automate first with AI explains why internal, reviewable workflows usually come before customer-facing ones. If you would bring in outside help to build one, our guide to what an AI automation agency does covers how to tell a builder from a reseller.
When it hurts: accuracy.
Language models still make things up, even when given the right document. If your chatbot invents a feature, a price or a policy, your company is responsible for it.
Vectara, an AI company, maintains a public leaderboard that measures how often models add unsupported content when asked to summarise a short document they have been given, which is close to what a website chatbot does with your pages. In its October 2026 update the best model scored a 1.8% hallucination rate, and the widely used models shown below ranged from 7.0% to 15.1%:
Even a few per cent matters when the error is about your offer. The legal position is now clear in at least one common-law jurisdiction. In Moffatt v. Air Canada (2024 BCCRT 149, decided 14 February 2024), the airline’s website chatbot told a customer he could claim a bereavement fare refund within 90 days of buying a ticket. The airline’s actual policy said otherwise. Air Canada argued the chatbot was “a separate legal entity that is responsible for its own actions”. The British Columbia Civil Resolution Tribunal called that argument “remarkable”, held the airline responsible for the information on its own website, and ordered it to pay the difference: 812 Canadian dollars plus interest and costs.
The amount was small; the principle is not. If your chatbot tells a prospect your service includes something it does not, or quotes terms you do not offer, expect to be held to it.
How to cut the risk
- ✓Answer only from approved content: your pages and documents, not the open web.
- ✓Link every answer to its source page so the buyer can check it.
- ✓Ban topics: specific prices, contract terms, legal or compliance assurances, and anything about named clients go to a person.
- ✓Prefer “I don’t know, here is who does” over a confident guess.
- ✓Read transcripts weekly for the first three months and fix the source pages, not just the bot.
When it hurts: privacy and the law.
A chatbot collects personal data, often more than a form does, and in Europe it now carries a disclosure duty. Plan for both before launch.
Under the EU AI Act, Article 50 requires providers to ensure that AI systems “intended to interact directly with natural persons” are designed so that “the natural persons concerned are informed that they are interacting with an AI system,” unless that is obvious to a reasonably well-informed person. Article 50 applies from 2 August 2026. If you sell into the EU, label the assistant clearly as AI at the start of every conversation. It is good practice everywhere else, and it is what the trust data in chapter 3 suggests buyers want anyway.
Privacy is the other half. Chat transcripts contain names, emails and often details about a buyer’s business problems. Treat them like any other personal data: say in your privacy notice what the chatbot collects and why, keep only what you need for the enquiry, set a retention period, and check where your chatbot vendor stores and processes transcripts and whether it uses them to train its models. If your buyers work in regulated sectors, ask the vendor for its data-processing terms before you sign. Our guide on explaining your use of AI on your website covers what to tell visitors.
When it hurts: friction.
A chatbot hurts conversion when it gets in the way of a buyer who was ready to act. The common forms of friction are easy to spot and easy to avoid. If you are planning a new site, design them out as part of the website redesign checklist rather than after launch.
| Pattern | What the buyer experiences | Fix |
|---|---|---|
| Auto-open pop-up | A chat window covers the page they came to read, often on mobile | Open only on click; use a small, labelled launcher |
| Form hidden behind the bot | No way to send a normal enquiry without chatting | Keep the contact form and email visible on the page |
| Data before value | Asked for email and phone before any answer | Answer first; ask for contact details only to hand off |
| Fake human | A name and photo that turn out to be a bot | Say “AI assistant” plainly, from the first message |
| Loops | The bot repeats options or misunderstands, with no way out | After two failed turns, offer a person automatically |
| Promised callback that never comes | “Someone will be in touch” and nobody is | Give a real reply time and route to a named team inbox |
If several of these describe your current site, a chatbot is not the problem to solve. Start with the enquiry path itself; our walk-through of why websites fail to generate leads lists the usual leaks.
Designing the handoff to a human.
The handoff is the most important part of a B2B chatbot. Decide when the bot stops, what it passes on, and how fast a person picks it up, before you write a single prompt.
- 01 · ASKIs the question answered on an approved page?
- 02 · ANSWERYes: answer briefly and link the page
- 03 · TRIGGERNo, or price, terms, or a buying signal: offer a person
- 04 · CAPTURECollect name, company, need and best contact
- 05 · ROUTESend the full transcript to the right person, with a reply time
Two failed answers in a row should also trigger step 03.
Seven handoff rules
- Offer a person from the first message, not only when the bot fails.
- Hand off on buying signals: price, timeline, procurement, “can we talk”, or a named competitor.
- Pass the whole transcript so the buyer never repeats themselves.
- Tell the buyer exactly what happens next: who will reply, through which channel, and by when.
- Route to a named team, never a shared inbox no one owns.
- Outside hours, say so, and give the real time a person will reply.
- Record the handoff as an enquiry in your analytics and CRM, so you can measure it.
Assistant: “Hi, I’m an AI assistant. I can answer questions about our services from our website, or put you in touch with a specialist at any point.”
Buyer: “Do you handle ISO 27001 readiness for a 120-person software company, and roughly what would it cost?”
Assistant: “Yes, ISO 27001 readiness is one of our services; our page explains the steps and typical timeline [link]. Cost depends on your current controls, so a specialist would need to ask a few questions. Shall I pass this to our compliance lead? They reply within one working day.”
Why it works: it discloses that it is AI, answers the factual part with a source, refuses to guess a price, names who picks it up and when.
An invented exchange for a fictional firm, written to show the pattern.
The decision scorer.
Tick each statement that is true for your business. The scorer tells you whether an AI chatbot is likely to help now, and what to fix first if not. Nothing you tick is sent anywhere.
Readiness: 0 / 100
Tick what is true today. If you are unsure, leave it unticked.
The weights favour content and ownership over technology. A good tool with nobody reading transcripts will drift; a simple tool with a clear owner and a fast human handoff usually does fine.
Measuring whether it works.
Judge a chatbot by qualified enquiries and buyer experience, not by the number of conversations. Conversations are easy to inflate; pipeline is not.
What to measure
- Enquiry rate on pages with the chatbot, before and after launch, counting both forms and chat handoffs.
- Share of handoffs that become qualified opportunities in your CRM.
- Time from handoff to first human reply, against the time you promised.
- Answer accuracy from a weekly sample of transcripts: right, wrong or should have handed off.
- Form submissions, to check the chatbot is not simply replacing enquiries you would have got anyway.
Run it as a test where you can. Launch the assistant on a set of pages and leave comparable pages without it for a period, then compare qualified enquiries, not chats. If the numbers do not move, remove it; a widget that adds no enquiries still adds page weight and a privacy obligation. Our guides on website lead tracking and judging by qualified opportunities show how to set this up.
Questions owners ask.
Should a B2B website have an AI chatbot?
Only if buyers ask repeatable factual questions, often arrive when no one can reply quickly, and you have someone to own the content and a fast human handoff. Otherwise improve the pages, form and follow-up first.
Do chatbots increase conversion rates?
There is no reliable vendor-neutral figure for B2B websites. The best evidence shows AI helping human support agents resolve more issues, not bots replacing them. Test it on some pages against others and judge by qualified enquiries.
Are we responsible for what our chatbot says?
Expect to be. In Moffatt v. Air Canada (2024), a Canadian tribunal held the airline responsible for wrong information its website chatbot gave a customer and rejected the argument that the chatbot was a separate legal entity.
Do we have to tell visitors it is AI?
In the EU, Article 50 of the AI Act requires people to be informed they are interacting with an AI system, from 2 August 2026, unless it is obvious. Elsewhere it is still good practice, and surveys show most people want to know.
When should the chatbot hand over to a person?
Whenever the answer is not on an approved page, on any question about price, terms or legal matters, on buying signals such as timelines or procurement, and after two unsuccessful answers. Always offer a person from the first message.
Sources: KPMG and University of Melbourne, Trust in AI 2025; Pew Research Center, 3 Apr 2025; Stanford AI Index 2025, public opinion; Salesforce, State of the AI Connected Customer; Nielsen Norman Group, The User Experience of Chatbots, 2018; NBER Working Paper 31161; Vectara hallucination leaderboard; Moffatt v. Air Canada, summary and references; EU AI Act, Article 50. All accessed 10 November 2026.