As of 13 August 2026, AI can only partly add an AI chatbot to your business website.
This still needs a person who signs their name to it.
Can you do it?
15 minutesto a draft.
2 hoursto something you’d act on.
Cost, all in£0/month
Skill neededchat-fluent
Who has to check ityou
What the alternative costsSiteGPT is an alternative purpose-built tool that creates a chatbot trained on your website and answers visitor questions.
If this goes wrong: the bot gives a confident but incorrect answer, mishandles a complaint or fails to pass a customer to a person, and your business has to repair the customer relationship.
What to actually do
Hand it to a person
The route this page recommends
A person who owns the outcome does this end to end, worth it when the failure is dear.
Use a tool built for this
Second choiceDo it yourself
The distant thirdA chat interface, chat-fluent skill, and roughly 2 hours until you can act on the result.
How to actually do it
- Open your website, help centre, FAQs, terms, returns policy, delivery information and current contact page, then remove superseded or contradictory content.
- Write down the human support email or phone route, opening hours, expected response route, complaint process and the topics that must always go to a person.
- Choose a purpose-built chatbot platform such as SiteGPT, upload or connect only the approved pages and documents, and configure the bot to answer from those sources.
- Paste the prompt and your business information into the chatbot configuration, replacing every bracketed slot with current facts and leaving unknown items marked as needing business input.
- Run the ten generated test conversations and add your own questions about refunds, cancellations, accessibility, complaints, personal data and requests outside the chatbot's scope.
- Compare every answer against the current website and policies, test the human hand-off on desktop and mobile, correct any unsupported answer, then embed the widget only after a named person agrees how live conversations will be monitored.
Prompt
Help me plan and configure an AI chatbot for this UK business website: [WEBSITE URL]. Its job is to handle straightforward tier-1 customer-support questions and send anything uncertain, sensitive, urgent or complaint-related to a human. Business context: - Business name: [BUSINESS NAME] - What we sell or provide: [SHORT DESCRIPTION] - Human support contact and opening hours: [CONTACT DETAILS AND HOURS] - Topics the bot must never answer without a human: [TOPICS] - Customer data the bot may and may not request: [RULES] - Approved support sources: [PASTE HELP CONTENT, FAQs, TERMS, RETURNS POLICY, DELIVERY INFORMATION AND CURRENT CONTACT DETAILS] Create: 1. A concise chatbot welcome message. 2. A list of supported tier-1 question types. 3. A list of prohibited topics and clear hand-off triggers. 4. A response policy that says the bot must use only the approved sources, must say when it does not know, must not invent policies or promises, and must offer the human contact route when needed. 5. Ten realistic UK customer test conversations, including an uncertain question, an angry complaint, a request for personal data, an urgent matter and a question outside the supplied sources. 6. A pre-launch checklist covering accuracy, escalation, privacy, accessibility, mobile display and human follow-up. Do not invent business facts, prices, delivery times, legal rights or contact details. Mark missing information as [NEEDS BUSINESS INPUT]. Keep the tone plain and helpful. Do not present the plan as legal advice.
Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.
What it gets wrong
- AI cannot decide your acceptable balance between automation, customer experience and commercial risk.
- AI cannot know which internal policy is current when your website and staff guidance disagree.
- AI cannot reliably recognise every unusual complaint, vulnerable customer or situation that needs a human.
- AI cannot take responsibility for inaccurate answers, privacy failures or customers who are not handed to a person.
- AI does not replace ongoing testing when products, prices, policies or opening hours change.
What caps this at PARTLY: verification cost, judgement under ambiguity and legal accountability.
How we scored this
Five axes, each scored nought to two by hand: ten means AI carries the task cleanly, and the thresholds that turn a total into YES, PARTLY or NO are published in the methodology. Each axis name links to its definition.
| Axis | Score (0–2) |
|---|---|
| Output | 2 |
| Inputs | 1 |
| Verification | 1 |
| Liability | 1 |
| Effort delta | 2 |
| Total | 7 / 10 |
The methodology and its thresholds are published in full.
FAQ
- Can AI build a chatbot for my website?
- Yes, current purpose-built tools can create a chatbot from your website and documents and provide an embeddable web widget. They do not remove the work of preparing accurate content, setting escalation rules and testing the live experience.
- What do I need to give an AI chatbot?
- Give it current FAQs, help pages, terms, delivery and returns information, contact details and clear human hand-off rules. Also specify what customer data it may request and which subjects it must not handle.
- Can a website chatbot handle customer complaints?
- It can collect basic information and route a complaint, but it should not be the final owner of a serious or emotional complaint. Configure it to offer a human route early and test that route with realistic complaints before publishing it.
- Is an AI chatbot responsible if it gives the wrong answer?
- Your business remains responsible for how the chatbot is used and for the customer impact of its answers. Treat its replies as customer-service output that needs monitoring, correction and a clear route to a person.
Nearby answers
Assessed by gpt-5.6-luna (gpt-5.6-luna) on 2026-08-13, second-checked by an independent model. Wrong somewhere? Email [email protected] and it gets re-checked.
The newsletter
AI news, new answers and product picks, straight to your inbox.