ReviewVoice
← What we measured

Nearly half of café review replies do not mention the review

31 August 2026 · 6 min

We read 14,130 Google review replies published by 914 cafés across 18 UK cities. Then we did something simple: for each reply, we checked whether it contained any of the distinctive words from the review sitting directly above it.

6,768 of them contained none. Not one. That is 51 per cent — very nearly half of every reply we read.

14,130
replies read
51%
share no distinctive word with the review
3,018
are eight words or fewer

What "distinctive" means here

Every word in every reply was weighted by how rare it is across the corpus. "Food", "service" and "thank" appear in almost every reply, so they count for almost nothing. "Brisket", "shakshuka" and "Natasha's Law" appear in one review and nowhere else, so they count for a great deal. A reply scores zero when it shares none of the rare words — when nothing in it could only have been written about that one review.

This is a deliberately generous test. A reply does not have to be good to pass it; it only has to be about something. "Sorry the brisket was dry" passes. "Thank you for your feedback" does not.

The shape of the problem

3,018 replies — 21 per cent — are eight words or fewer. At that length there is no room to be about anything. But length is not the whole story: plenty of the zero-scoring replies run to forty or fifty words. They are long and still say nothing, which is worse, because the reader can see the effort and still learns nothing from it.

Thank you so much for your kind review, we really appreciate you taking the time and hope to welcome you back again soon!

Thirty-one words, scores zero. Published under a review that named a specific dish and a specific member of staff.

Why it costs something

A reply is the only place on your listing where you get to speak. The person reading it has already read the review; they are looking to see whether anyone was listening. A reply that could sit under any review at any business answers that question in the wrong direction.

There is a second cost that is easier to measure. 664 sentences in our corpus appear word for word at two or more separate businesses, and 122 of them appear at five or more. Generic replies are not just empty — they are shared. A customer scrolling three listings on a Tuesday evening has read your closing line twice already.

The band where it is worst

Five-star reviews attract the shortest replies and the most repetition. The median five-star reply in our corpus is 18 words, against 67 for one star. That is not wrong in itself — a happy review does not need an essay. What is wrong is that the twenty words are almost always the same twenty words.

The same failure, subject by subject

The complaint being answered barely moves the needle. We tagged every review by what it was actually about and ran the same test inside each group:

Two things stand out. Complaints about staff and about waiting attract the worst replies, and complaints about coffee attract the most replies overall — which makes sense, because coffee is the subject a café is allowed to know more about than the reviewer, and almost nobody uses that.

What a reply that scores looks like

It does not require length or eloquence. It requires naming the thing. A reply that says the grind is set for the milk it goes into, or that two lattes ordered together should match, is anchored — because those sentences could not have been written under any other review.

That is the whole test. Not whether the reply is warm, or apologetic, or well punctuated. Whether a stranger reading it could tell which review it belongs to.

Why it is not laziness

It is worth saying plainly: the people writing these replies are not lazy. They are running a café. The reply gets written between the morning rush and a delivery, on a phone, and the safest sentence is the one that cannot possibly cause trouble.

The trouble is that the safest sentence is also the one that does nothing. The reply was the one place on the listing where the business could speak, and it was spent on a formula. That is not a character problem. It is a time problem, and time problems have tools.

The sentences everyone writes

If a reply says nothing, it usually says nothing in a sentence someone else has already used. These are the ten most repeated sentences in the corpus, with the number of separate businesses publishing each one:

None of these is a bad sentence. Each is a perfectly reasonable thing to write once. The problem is arithmetic: a customer comparing three cafés on a Tuesday evening reads all three listings, and the same closing line greets them at each.

How this was measured

14,130 review-and-reply pairs, read from live Google listings at 914 cafés in 18 UK cities across 90 neighbourhoods, August 2026. Reviews not written in English were excluded from this particular count, because a word-overlap test cannot be run honestly across two languages — that exclusion is itself a finding, and we wrote about it separately.

No café is named. Nothing here is estimated, modelled or rounded up from a sample. Every figure is a count.

See where your own café stands

Free, about twenty seconds, and we write one reply for you before you pay anything.

Find my café →