Across 14,130 replies from 914 UK cafés, 8 per cent break at least one of the checks we apply to our own writing. Confine the count to the replies that answer one- and two-star reviews and the figure changes completely: 703 of 1421.
The gap is not about competence. It is about the moment. A one-star review arrives as an accusation, and the reply gets written in the ten minutes after reading it. What comes out is what a person says when they feel accused: an apology that goes further than the facts, an offer to make it right, a promise it will never happen again.
What breaks, in order
We expected the commonest failure to be conceding fault. It is not, and the gap is large. Counted across all 14,130 replies, the order is:
- moving the conversation off the page — 761 replies
- promising money back, a free drink or a voucher — 165 replies
- saying nothing at all — 149 replies
- conceding the allegation in writing — 57 replies
- administrative language no café actually uses — 45 replies
Moving the conversation off the page is not merely the commonest — it is commoner than everything else put together. 470 replies ask the reviewer to get in touch, and 617 print an email address or a phone number to do it with. The intention is good and the effect is not: every future reader sees a business that would rather have this conversation somewhere private.
Conceding the allegation outright is far rarer than we assumed — 57 replies in the whole corpus. When it happens it is the most expensive of the lot: "that is completely unacceptable" is a sentence the business would never say to an inspector, a landlord or an insurer, published under its own name and left there permanently. But it is not the everyday failure. The everyday failure is quieter, and it is the one nobody thinks of as a failure at all.
Guaranteeing the future — "this will never happen again" — is rarer still, and everyone reading it knows it might. A promise the reader does not believe costs more than it earns.
The one place nobody slips
There is a striking exception. 177 reviews in our corpus raise an allergy, an intolerance or a dietary requirement. The number of replies to them that concede fault is 0.
Nobody trained these businesses to do that. They simply know, without being told, that an allergen complaint is a different kind of document — that Natasha's Law made it a regulated matter and that a published admission is a published admission. The instinct is exactly right. It is also proof that the instinct exists: the same owner who will not concede on allergens will concede on a cold flat white ten minutes later, because that one does not feel legal.
It is not the same everywhere
We ran the identical test in six cities. The spread is wider than we expected for a test this narrow:
- Liverpool — 14% of 478 replies break a check
- Oxford — 13% of 529 replies break a check
- Birmingham — 12% of 1,099 replies break a check
- London — 10% of 2,108 replies break a check
- Nottingham — 10% of 790 replies break a check
- Cambridge — 9% of 535 replies break a check
- Leeds — 8% of 1,026 replies break a check
- Manchester — 8% of 918 replies break a check
- Edinburgh — 7% of 687 replies break a check
- Brighton — 6% of 832 replies break a check
- Bristol — 6% of 756 replies break a check
- Newcastle — 6% of 545 replies break a check
- Bath — 5% of 585 replies break a check
- Belfast — 5% of 408 replies break a check
- Sheffield — 5% of 791 replies break a check
- York — 5% of 599 replies break a check
- Cardiff — 4% of 682 replies break a check
- Glasgow — 4% of 762 replies break a check
The cities at the bottom of that list are not writing shorter replies. Median length barely moves between them. They are writing the same amount and conceding less.
The subjects that provoke it
Tagging every review by what it was about shows where the pressure comes from:
- slow service and long waits — 38% of replies break a check, across 344 reviews
- rude staff — 47% of replies break a check, across 307 reviews
- cold food and lukewarm drinks — 32% of replies break a check, across 212 reviews
- cleanliness and hygiene — 32% of replies break a check, across 117 reviews
- service charge and payment — 28% of replies break a check, across 102 reviews
Complaints about people provoke worse replies than complaints about things. A cold coffee is a fact; a rude server is a character judgement about someone the owner knows, and the reply gets written in their defence or at their expense. Both are mistakes, and the second is the one that ends up in writing.
What a reply can do instead
Not conceding is not the same as arguing, and it is not the same as saying nothing. The move is to state the standard rather than the verdict. Two lattes ordered together should match — that is not a matter of taste. A flat white should not arrive lukewarm. Neither sentence admits that yours did; both tell the reader you know what right looks like.
That is also what the next customer is reading for. They are not adjudicating between you and the reviewer. They are working out whether this is a business that knows its own standards.
The length trap
Replies to angry reviews are the longest in the corpus — a median of 67 words at one star and 62 at two, against 18 at five. That is the right instinct: a serious complaint deserves more than a thank-you.
But length is where the trouble enters. Every extra sentence is another chance to concede something, to offer something, or to move the conversation somewhere private. The replies that break a check are not the short ones.
This is the part that is genuinely hard, and it is worth being fair about. Nobody writes a bad reply on purpose. They write a long one because the complaint deserved a long one, and somewhere in the fourth sentence they give away more than they meant to.
What the reader is actually deciding
The reviewer has already had their experience and is unlikely to change their mind. The audience for the reply is the next person: someone with a free evening, three listings open, deciding where to go.
That person is not adjudicating the dispute. They are reading for something much simpler — whether this is a business that knows what it is doing. A reply that concedes everything answers no; a reply that argues answers no; a reply that states the standard and leaves it there answers yes.
How this was measured
14,130 review-and-reply pairs from 914 cafés in 18 UK cities, August 2026, read from live Google listings. A reply counts as breaking a check when it matches one of the patterns we apply to our own output before publishing. No café is named.