Skip to content

GEO courseModule 9 of 10: After the MentionLesson 9.3

Website Chat, AI Receptionists and Calls

A chat box on your site, or an AI that answers or screens your phone calls. Run well, they catch enquiries out of hours. Run badly, they state a wrong fee or rent on your behalf, and that statement is yours.

0.5% of your GEO effort4 min read~15 min/week

Why it matters

A chatbot trained on your whole site, old PDFs included, will quote fees you no longer charge, and you answer for it.

The evidence

Evidence: Official guidance

The ICO's own chatbot record as a model, and consumer law on statements made on your behalf. Applying them to agency chatbots is our reasoning: take advice.

Do this first

Ask your chat tool five questions you know the answers to, including "how much is my house worth?", and record every wrong or invented answer.

01What it is

A robot at the shop door that answers “what time do you open?” Great if it knows the answers. A problem if it makes things up, so there must always be a way to fetch a real person.

Think of it like

A new receptionist on their first day. You'd give them a script, tell them what they must never promise, and sit nearby for the first week.

02Why it matters

To the people searching

Buyers and tenants want quick, correct answers about availability, viewings and fees; a wrong one wastes a trip. Sellers and landlords may type in personal details, and need to know who sees them and that a person will follow up.

buyers

Want quick, correct answers about viewings and availability.

tenants

Ask about fees and pets, and act on the answer.

sellers

May type in details about their home and expect a person to follow up.

landlords

Want to know who sees their details and when they'll hear back.

To Google, Bing and AI

Evidence: Official guidance

Chat widgets are usually loaded by script, so AI crawlers reading your page code won't see what the bot says (lesson 3.3): never rely on a chatbot to carry facts that aren't on your pages. A bot that quotes a different fee from your fees page creates exactly the mismatch customers and AI both notice. On responsibility: under the DMCC Act a trader answers for statements made on its behalf, and a Canadian tribunal held Air Canada liable for its chatbot's wrong answer in 2024. The ICO's own chatbot record shows good practice: it says it's AI, needs no personal data, hands over to a human, and has its content reviewed quarterly. For phone AI, the ICO's rules on automated marketing calls (PECR) apply.

An agent example

A lettings agency installs an AI chat widget trained on its whole website, including an old PDF with outdated fees. A visitor asks about an admin fee and gets a figure the agency no longer charges. The agency removes the PDF, restricts the bot to its fact sheet and current fees page, blocks valuation questions (the bot now offers to book one instead), adds a clear "AI assistant" label and a "talk to a person" button, and reads the transcripts weekly.

03Your turn

Check yours, then fix it

  1. 1Ask five questions about fees, availability and areas you know the answers to.
  2. 2Ask one it should refuse: "How much is my house worth?"
  3. 3Record every wrong or invented answer.

  1. 1Can you reach a person in one click? Does anyone see it promptly?
  2. 2Does the widget say it's AI?

  1. 1Read ten recent transcripts or call logs.
  2. 2Look for wrong answers and unhappy visitors.

Copy this prompt

Here is our fees page and a chat transcript from our website assistant. List any statement in the transcript that doesn't match the fees page.

04Keep it up

Weekly transcript read and test while it’s new; monthly once it’s settled.

~15 min/week

That's lesson 9.3

Worked through it? Mark it done and we’ll keep track on this device.

← Back to 9.2 Speed-to-Lead: Valuation and Enquiry Journeys

Up next · 10.1Measuring Honestly: The Monthly Report, and What Can’t Be Measured