AI chatbots for marketing: what UK brands should know

What to take away

AI chatbots earn their place when they capture buying intent, not when they answer every question.
UK GDPR applies to chat conversations, so you need a lawful basis, a clear notice and a route to a person.
Trust falls when a bot hides its nature or blocks escalation to a human.
Measure qualified conversations and assisted revenue rather than message volume.
Start with one job, such as booking demos or handling delivery questions, then widen it.

Where a marketing chatbot fits

Most UK marketing teams meet chatbots through customer service first.

The marketing value sits upstream of that: qualifying a visitor, capturing a need and routing them to a human or a booking page while interest is still warm.

On a pricing or product page, a tightly scoped bot can handle two or three recurring objections and offer a demo slot.

That is a marketing job, not a support job. Support metrics such as deflection rate can look strong while lead quality stays flat.

Examples to assess include Intercom, HubSpot, Tidio and Zendesk. Check whether the relevant product meets your needs for handover to a person, integrations, transcript controls and approval of answers before choosing.

Captured details then need a follow-up sequence, and our comparison of AI email marketing tools for UK businesses covers what is worth automating.

Chatbot types and what each one is good for

Platforms differ less in language quality than in what they can do with a conversation.

Chatbot types compared

Rules-based

Typical use: Answering shipping and returns
Cost: Cheap
Predictability: Predictable
Governance need: Low
Can act: No

Retrieval bot

Typical use: Explaining a service
Cost: Moderate
Predictability: Content-bound
Governance need: Approve answers
Can act: No

Generative assistant

Typical use: Booking calls
Cost: Higher
Predictability: Messy phrasing
Governance need: Logging and retention
Can act: Yes

Typical marketing use

Rules-based decision tree: Answering shipping, returns or pricing questions on a product page
Retrieval bot over your own content: Explaining a service or matching a visitor to the right package
Generative assistant with tool access: Booking calls, checking order status, drafting a quote
Voice or messaging bot on WhatsApp, SMS or phone: Following up on abandoned enquiries

What to check before buying

Rules-based decision tree: Whether it can hand over to live chat without losing the transcript
Retrieval bot over your own content: How source content is refreshed and who approves answers
Generative assistant with tool access: Logging, retention and whether it can act as well as talk
Voice or messaging bot on WhatsApp, SMS or phone: Consent for the channel and the staffing cost of handover

Rule-based bots are cheap and predictable. Generative assistants cope with messy phrasing but need governance, because they can produce an answer nobody approved.

What UK data protection law expects

When a chatbot conversation contains personal data, collecting or using it requires a lawful basis. The ICO's Guidance on AI and data protection explains how transparency, data minimisation and human review apply to AI systems.

The UK GDPR has six lawful bases: consent, contract, legal obligation, vital interests, public task and legitimate interests. Choose a basis for each purpose based on the circumstances and what is necessary: legitimate interests may suit lead qualification after a documented purpose, necessity and balancing assessment, while consent is appropriate when processing genuinely depends on a freely given choice. Do not treat consent as a universal default. Showing a different offer based on what a visitor types can amount to profiling, which the UK GDPR defines. A decision based solely on automated processing that has legal or similarly significant effects engages Article 22 restrictions and safeguards; separately, people have a right under Article 21 to object to processing for direct marketing.

Keep transcripts for a set period and say what that period is. A short notice in the chat window, linked to your privacy page, does more for trust than a policy nobody opens.

Getting the basics right: a short sequence

Getting the basics right

Write down the single job the bot will do and the data it needs for that job.
Name a lawful basis for each field and record it in your privacy notice.
Tell users they are talking to a bot in the first message, before they type.
Give a one-click route to a human and publish the hours when that is staffed.
Set a retention period for transcripts and review a sample every month.

Consumer trust decides the result

The CAP Code requires marketing communications to be obviously identifiable as such; rule 2.1 covers this. The Advertising Standards Authority (ASA) administers and enforces the CAP Code. These rules apply to automated replies and chatbot prompts as much as to paid ads, so approve the wording.

The handover matters just as much. A bot that loops twice before offering a person does more damage than having no bot at all.

Measuring whether it earns its keep

Message volume tells you very little. Track qualified conversations, booked meetings and revenue influenced by a chat session.

The fair comparison is against your existing forms and live chat, over the same period and the same traffic sources.

Agree a written definition of a qualified conversation. Then track it in your own analytics rather than in the vendor dashboard. Our marketing analytics buyer's checklist sets out the fields worth capturing.

Common questions

Does a chat widget need consent like other marketing tags?

Under regulation 6 of the Privacy and Electronic Communications (EC Directive) Regulations 2003 (PECR), storing information on or accessing information from a user's device, including through cookies or similar technologies, generally requires consent unless an exception applies, such as when it is strictly necessary to provide a service the user requested. Check what the vendor loads and prevent non-essential storage or access until consent. If you use chat details for email or text marketing, PECR generally requires consent for individual subscribers unless the soft opt-in applies; different rules apply to corporate subscribers.

Can a chatbot make product claims?

Anything the bot says is your marketing communication. Claims must be accurate and substantiated, so approve a fixed list of answers rather than letting it improvise.

Should the bot admit it is not human?

Yes, and early. Deceiving customers about the nature of the interaction damages trust and creates avoidable regulatory risk. A simple label plus a route to a person is enough.

What about WhatsApp and other messaging channels?

Messaging channels bring their own consent and cost questions, and voice bots add call recording duties. Pilot on your website first, where you control the notice, before adding a channel.