Would a chatbot help your customers get answers?

01

What is an AI chatbot, and how is it different from a rule-based one?

A chatbot is software that converses with website or messaging visitors. A rule-based chatbot follows fixed scripts and menus. An AI chatbot uses a language model to write replies. Many products combine both, using rules for sensitive steps and AI for open questions.

The difference changes what can go wrong. A scripted bot can only say what you wrote, so it is predictable but fails on unexpected questions. An AI bot copes with varied wording but can say things nobody wrote, which is useful and risky. AI chatbots are often set to answer from your own approved content, a method called retrieval augmented generation (RAG). The UK Government AI Playbook warns that RAG tools are susceptible to indirect prompt injection.

A hybrid design uses rules for bookings, payments and handover, and AI only for open questions. It gives you control where it matters most.

02

What can a chatbot genuinely do well for a small business?

Chatbots do five jobs reasonably well: answering repeated factual questions, collecting enquiry details, offering booking slots, covering out-of-hours messages and passing complex cases to a person with the conversation attached.

• Frequently asked questions, where the answer is already on your site.
• Lead qualification, asking the same few questions every time.
• Booking, showing real availability and confirming slots.
• Out-of-hours cover, saying when a person will reply.
• Escalation, passing the chat and its history to a person.

Each works best when the answers come from content you wrote and keep current. Where the bot has no content, it should say it does not know. Left alone, an AI bot may guess.

03

What can a chatbot not do, or not do reliably?

A chatbot cannot take responsibility, use real judgement or be relied on to be accurate every time. It may invent facts, misread an upset customer or reveal information if manipulated. It should not give legal, medical or financial advice.

The UK Government AI Playbook says language models can create content that appears plausible but may be factually incorrect, and that it is difficult to produce an AI system giving 100% accurate outputs under all conditions. It adds that they are not domain experts and not a substitute for professional advice.

Typical weak points are prices, stock and policies it was never given or that have changed, complaints and distressing situations, ambiguous questions where it should ask rather than guess, and attempts to override its instructions.

04

Who is responsible if my chatbot gets something wrong?

You are. A chatbot speaks for your business, and UK consumer protection law bars traders, and people acting for them, from giving consumers false or misleading information. Check your supplier contract, but do not assume it removes your duties to customers.

The CMA's guidance on the unfair trading provisions of the Digital Markets, Competition and Consumers Act 2024, which apply from 6 April 2025, defines a trader as including a person acting on the trader's behalf. It says traders should not give consumers objectively false information and that factual claims should be supported by evidence. The guidance does not mention chatbots, so applying it to chatbot answers is our reading of the principle, not a ruling.

A notice saying "this chatbot may make mistakes" helps set expectations but does not replace getting the answers right.

05

How much does a chatbot cost?

Cost depends on the type of chatbot and who builds it, and prices change often, so this guide gives no figures. Expect some mix of subscription, usage charges, set-up, content preparation, testing, upkeep and staff time for handover.

The costs people overlook are usually ongoing: keeping content current, reading transcripts, fixing wrong answers and staffing the handover. Use a small, time-limited pilot to learn them before committing to a long contract.

06

Does my business actually need a chatbot?

Probably only if you get many repeated questions, enquiries outside staffed hours or slow first responses, and have accurate written content. If enquiries are few, complex or sensitive, a clearer website, FAQ page or contact form may do more.

Official figures show how common AI use has become, not whether it works. The ONS reported that in June 2026 about 28% of businesses with 0 to 9 employees and about 35% of those with 10 or more used at least one AI technology. Among adopters with 10 or more employees, only 10% described their use as extensive. Earlier ONS analysis found that difficulty identifying business use cases, cost and lack of expertise were the most commonly reported barriers. The ONS does not report chatbots separately.

07

What do UK data protection rules require when I use a chatbot?

If the chatbot handles personal data, UK GDPR applies. Tell people what you do with their data when you collect it, have a written contract with the supplier, collect only what you need, and consider a data protection impact assessment, because AI can count as innovative technology.

ICO guidance is under review after the Data (Use and Access) Act 2025, and the automated decision guidance is still in draft, so check the latest versions. Collect only what you need, and ask visitors not to type sensitive details.

08

How should a chatbot hand over to a person?

A good handover is easy to trigger, fast and keeps context. Offer a visible "talk to a person" option, send the transcript to whoever picks up, set honest expectations on waiting time, and provide a callback or email option when nobody is available.

Hand over automatically when the conversation involves a complaint, a payment problem, health, legal or safety concerns, signs of distress or repeated misunderstanding.

The ICO's draft automated decision guidance says human intervention cannot be tokenistic: reviewers should be able to influence the outcome and have authority to change the decision. The AI Playbook likewise calls for meaningful human control.

09

How do I keep a chatbot accurate and safe after launch?

Limit it to approved content, test it with real and awkward questions before launch, review conversations regularly and fix wrong answers quickly. Give it the least access it needs, because people may try to manipulate it, and keep logs so problems can be traced.

The AI Playbook also calls for testing before deployment and monitoring afterwards for drift, bias and hallucinations. Review a sample of transcripts every week and update content whenever prices or policies change.

10

How should I trial a chatbot without much risk?

Run a narrow pilot. Choose one job, such as FAQs or booking, write the approved answers, test before launch, label the bot clearly with an easy route to a person, then review transcripts weekly and judge it on wrong answers, handovers and customer feedback.

1. 01
   Pick one job

   Choose a task with plenty of repeat questions and low risk if an answer is slightly off.

2. 02
   Write the approved answers

   Use your own policies and website content and name an owner for keeping it current.

3. 03
   Label it and add handoverFull step in the download ↓
4. 04
   Test before launchFull step in the download ↓
5. 05
   Launch small and review weeklyFull step in the download ↓

Reeve Consult offers AI consultancy and web development. If you want a second opinion on whether a chatbot suits a workflow, you can contact us. No specific saving or result is promised.