20 Nov 2026
4 Nov 2026
min read
In November 2022, Jake Moffatt asked the chatbot on Air Canada's website about bereavement fares after his grandmother died. The bot told him he could buy a regular ticket and apply for the bereavement rate within 90 days. Air Canada's real policy didn't allow claims after travel, so when Moffatt asked for the money back, the airline said no.
At the tribunal, Air Canada argued that the chatbot was a separate legal entity responsible for its own actions (yes, that was the argument). British Columbia's Civil Resolution Tribunal didn't buy it. In February 2024 it held Air Canada responsible for everything on its website, chatbot included, and ordered it to pay Moffatt about C$812 in damages, interest and fees.
C$812 won't sink anyone, but the reasoning reaches well beyond airlines: whatever appears on your website is your statement, and that includes the parts a widget writes for you. If you run a five-person bakery and add an AI chatbot or a live feed of Google reviews, you take on the same responsibility. You usually do it in an afternoon, and usually without touching a privacy policy or terms that were written years earlier.
So here's what to update, in roughly the order it makes sense to do it.
Website chatbots answer from the content you give them. An AI chatbot widget like Elfsight's, to take one example, learns from your site's pages, from files you upload (PDF, DOCX and others) and from question-and-answer pairs you write yourself. So if the refund policy on your site dates from 2023 and the PDF you uploaded is from 2025, the bot can quote either one. And when your content doesn't cover a question at all, many bots fall back on general knowledge, which is how a refund policy gets invented.
Before you switch a chatbot on:
Air Canada also argued that the correct policy was on another page of its site. The tribunal's answer was that customers shouldn't have to cross-check one part of a website against another. Keeping the bot and the policy pages in sync is a chore, but it's a cheaper one than explaining the difference later.
Since 2 August 2026, Article 50 of the EU AI Act has required AI systems that talk to people to make it clear they're AI, unless that's obvious from context. The duty sits mainly with the company that builds the chatbot (your vendor, in other words), but if you have customers in the EU, there's no reason not to say it yourself. In the US, California's BOT Act has made it unlawful since 2019 to talk to Californians online through a bot that misleads them about being a bot in order to push a sale.
You can cover this in one sentence. Put it in the welcome message – something like "Hi, I'm the AI assistant for Rosa's Bakery. I can help with orders and opening hours, or put you in touch with our team." – and add a short line in the chat footer that links to your privacy policy. Chatbot widgets usually have a field for both.
People type all sorts of things into chat windows – order numbers, phone numbers, health details, complaints about a neighbour. And if your chatbot has a lead form, it's collecting names and emails on purpose. If your privacy policy was written before you had a chat window, it probably doesn't mention any of this.
Add a section that covers:
If you're starting from scratch, DocPro's Privacy Policy (with GDPR) template gives you the structure, and you can add the chat section to it.
It's worth adding a sentence to your website terms and conditions saying that chatbot answers are general information and that the written policy on your site takes priority, because it sets expectations for anyone who reads it. It won't save you from a bot that misstates your own policy, though – Air Canada had the right policy on its own site and still lost. So write the disclaimer, but count on the first step (accurate source documents) to do the bulk of the protecting.
A reviews section looks like customers talking about you, but once it's on your own website it works as your advertising. The FTC's Rule on the Use of Consumer Reviews and Testimonials has applied to US businesses since 21 October 2024, and the FTC can seek civil penalties from knowing violators. Among other things, it bans:
The FTC does seem to be following up, too – on 22 December 2025 it announced warning letters to ten companies about possible violations.
The last item on that list is the one you can break without meaning to. Reviews widgets let you filter, and plenty of filters make sense: hiding spam, slurs, star-only ratings with no text, or a review that was clearly meant for another company. The trouble starts when a feed of hand-picked five-star reviews sits under a heading that implies it's everything customers said.
Two habits help:
The rule covers what your staff do as much as what your website shows, so it's worth putting your own rules in writing:
|
Document |
What to add |
|---|---|
|
Refund, shipping and cancellation policies |
One current, dated version of each, used as the bot's source |
|
Chatbot welcome message and footer |
A plain line saying it's an AI assistant, plus a privacy policy link |
|
Privacy policy |
Chat transcripts, lead-form data, the chat provider, retention, deletion requests |
|
Website terms |
Bot answers are general information; the written policy prevails |
|
Review policy (internal) |
Who asks for reviews, no rewards tied to ratings, staff disclosure, moderation rules |
|
Reviews section on the site |
The source's overall rating and count, and a label that matches the filter |
All of this is roughly an afternoon of work, which is about what it took to install the widgets in the first place. If you work in a regulated field like healthcare, finance or law, have a lawyer look at your chatbot setup before it goes live.
This article is general information, not legal advice.
Keywords:
Not the right document?
Don’t worry, we have thousands of documents for you to choose from: