How to ask British customers for reviews, and what you should not offer them

Most review programmes fail on timing and wording rather than on effort, and a surprising number of British businesses stop asking because one awkward request produced a complaint. The process below gets replies, and the second half of this page is a measurement that should change where you spend the effort afterwards.

The six step process

1. Ask at the moment of relief, not the moment of invoice. The right moment is when the customer's problem is solved and they can feel it. For a service business that is usually the day after delivery, not the day the invoice clears.
2. Ask by the channel they already use with you. If the relationship happened by email, ask by email. If it happened by phone, ask on the phone and send the link afterwards. A channel switch halves the response rate.
3. One person asks, by name. A request from the person who did the work gets answered. A request from a no reply address does not.
4. Make it one click and name the platform. Send the direct link. Asking somebody to search for your listing is asking them to do your work.
5. Suggest what to mention, without scripting it. Something like a note about the part that mattered most to them. This improves the usefulness of the review without putting words in their mouth, and it is the line between helpful and wrong.
6. Ask once, then stop. One reminder after a week. A third request is where irritation starts and it is the most common cause of the complaint you are worried about.

What not to offer, and why this needs checking rather than guessing

Do not offer anything in exchange for a review, and do not offer anything in exchange for a positive one. Beyond the obvious problem that an incentivised review stops being evidence of anything, the United Kingdom's consumer protection rules in this area were tightened recently and the enforcement position has moved, so what is permissible has changed within the last couple of years.

We are not going to quote statutory wording here, because we could not read the primary text on the day we wrote this. Two things happened worth recording. The authoritative legislative host cannot be fetched from our machine at all. And the guidance pages we tried on gov.uk returned a styled, fully rendered 404 of exactly 52,491 bytes, which looks like a successful read to any automated check, so we did not treat it as one. A 404 records what a server did on a date and is not evidence about the rules.

So the practical instruction is this. Before you offer a customer anything connected to a review, including a prize draw entry, a discount or a free month, get the current position from your solicitor or from the regulator's own current guidance. It is a short question with a definite answer, and the cost of getting it wrong now is higher than it was.

What is always safe

• Asking, politely, once, with a reminder.
• Making it easy with a direct link.
• Replying to every review, including the critical ones, in public.
• Fixing the thing a critical review identified and saying so.

The measured part, which should change where you spend the effort

On 22 October 2026 we put a 198 question set of British buying questions to ChatGPT and Gemini, with the market fixed inside the prompt because the gl, hl and location parameters on the search API are accepted, return 200 and answer from somewhere else. That produced 64 blind readings, 32 per engine. We counted every host cited.

Set against that, one single British agency website, whito.co.uk, was cited in 19 of the 64 readings and in 18 of ChatGPT's 32.

Read that carefully, because it is a finding about this dataset rather than about the value of reviews. In these British buying questions, the engines leaned on official hosts and on individual company websites, and barely touched the consumer review platforms at all. So if your reason for building a review programme is to get named inside AI answers, this evidence does not support it. If your reason is that a human buyer checks reviews before calling you, which they do, then the programme is worth running and the measured conclusion simply tells you not to expect a second benefit from it.

The thing that did earn citations in this set is a company publishing its own specific material. That is where the AI visibility effort belongs, and it is a different workstream from the review programme rather than a competing one.

What British businesses ask about collecting reviews

Which platform should we use? The one your buyers actually check in your category, which differs between trades, professional services and retail. Ask five recent customers where they looked before contacting you.

Should we reply to bad reviews? Always, briefly, factually, and without arguing. The reply is read by the next buyer rather than by the complainant.

Can we ask only the happy customers? Selecting who you ask on the basis of expected rating is the version of this that gets businesses into trouble, and it also produces a suspiciously perfect profile that buyers discount. Ask everybody.

How many reviews are enough? Enough to look like a real trading business with recent activity. Recency matters more than volume in most categories.

Do reviews help our search visibility? They help the listing they sit on and they influence human choice. Our measurement above is about AI answers specifically, and there they were weaker than expected.

What to do with the half of the effort this frees up

If the review programme is for human buyers and not for AI answers, the AI side needs its own work and its own measurement. That means a fixed set of the questions your buyers ask, read monthly, with two things recorded separately: whether your pages were cited and whether your business was named. The whito finding above is the clearest possible argument for keeping those apart, since the same site was cited in 18 ChatGPT readings and named in 4. AI Knows Us keeps that record across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews, and for a British buyer the plan that reads all five engines is 299 dollars a month, the price for the United States and the United Kingdom, read from aiknowsus.com on 22 October 2026.

The full directory measurement, host by host, is in our page on do AI assistants read British directories.