Your AI assistant is talking to your customers. We listen in.
We ask your store’s chatbot the questions real shoppers ask, check every answer against your own website, and tell you privately what it got wrong, before your customers act on it.
Your chatbot repeats whatever your website says. Including the parts that disagree.
Stores in our first testing rounds had at least one place where their own pages contradicted each other: two returns windows, two warranty lengths, or a sale that one page excludes from returns and another does not.
Wrong fact
Unpublished promise
Missing fact
Six things your customers ask. Six ways a bot can cost you.
Facts
Prices, delivery costs and times, stock and sizes.
Policies
Returns windows, refunds, warranties and cancellations.
Promises
Discount codes or offers your store never made.
Handoff
Whether a customer can reach a person when they need one.
AI disclosure
Whether it says it is an AI when asked.
Privacy
Whether it asks for details it does not need.
Quiet, careful, and fully evidenced.
We talk to your chatbot.
A small number of ordinary customer conversations on your public website, spread over at least two days. No logins, no tricks, no orders.
We check every answer.
Each reply is compared with your published pages. Anything that fails is asked again, and every finding is labelled Confirmed, Likely or Cannot verify.
You get a plain-English report.
For each finding: the transcript, a screenshot, your own page and a suggested fix. Contradictions on your website are listed separately, so you can fix the source.
Evidence you can act on in an afternoon.
- The most costly finding first, in one plain sentence.
- Every failure retested, with how often it failed.
- A grade on our own scale, so you can track it month to month.
- Facts and your own pages, never legal conclusions.
Example Store Ltd
10 conversations · 2 daysYour returns page says 30. Repeated in 2 of 3 retests.
Your delivery page lists it. Missed 2 of 3.
Your bot can only be as right as your pages.
We test like a customer. We behave better than one.
We always
- Use your public chat exactly as any visitor would.
- Ask ordinary shopper questions only.
- Keep volume low: at most 15 conversations, and no more than 10 on platforms sold with small monthly conversation allowances.
- Use our own test inbox if a chat asks for an email.
- Say it is a test and end the chat if one of your team joins.
- Keep your results private to your business.
We never
- Log in, access hidden pages or touch your systems.
- Use tricks or try to break the bot.
- Enter real personal details.
- Place orders, open tickets or request callbacks.
- Run automated tests on websites whose terms forbid it.
- Publish your name or your bot’s answers without your written consent.
Start free. Stay sharp monthly.
One round of ordinary conversations and a private report of what we found. No obligation.
Request a snapshotA fresh set of conversations every month, a report, and a check that last month’s fixes worked.
Talk to usFair questions.
Do you need access to our systems or chat platform?
No. We use the chat on your public website, the same way a shopper would.
Will our customers or staff notice?
No customer is involved. The conversations are few and ordinary. If a team member joins a chat, we tell them it is a test and end it.
Does a snapshot cost us anything?
The report is free. If your chat platform counts conversations towards a monthly allowance, our test chats count too, which is why we keep them to a small number.
Which chatbots can you test?
Any assistant on a public website. So far we have tested stores using Gorgias, Tidio (Lyro) and Intercom.
Is the report legal advice?
No. We describe what we observed and compare it with your published pages. Where a finding may raise a legal question, we suggest checking it with a qualified adviser.
Find out what your chatbot is telling your customers.
Email hello@quietshopper.com. Received a report from us? If anything in it is wrong, reply within 14 days and we will correct it. If you would rather not hear from us again, reply “no thanks” and we will not contact you again. Our reports describe what we observed; they are not legal advice.