Responding to Negative Reviews with AI: A 5-Rule Playbook

Rule 1 — Never invent a specific

The single fastest way to make a negative review worse is to reply with something the reviewer knows is false. "We've spoken with the server about your experience" when no one spoke with the server, "we've changed the recipe" when nothing changed — both are fabrications the reviewer (and their friends) immediately spot.

AI drafts should stay general: acknowledge the specific complaint in the reviewer's words, apologize for the experience, and either invite them to share more privately or commit to looking into it. No specifics that aren't verifiably true.

Rule 2 — Match the severity

A 1-star review claiming food poisoning needs a different reply than a 3-star review complaining about portion size. AI draft templates should escalate severity automatically:

• 1-star with health claim (food poisoning, allergy, etc.) → no AI draft; operator writes from scratch, legal review if claim is specific
• 1-2 star general complaint → AI drafts a sincere apology + private follow-up invite (DM us, call us)
• 3 star mixed → AI drafts acknowledging the positives + addressing the specific complaint
• 4-5 star → AI drafts a thank-you, no apology phrasing

Rule 3 — Reply in their language

Replying in English to a Russian review is fine if the reviewer is fluent, but you lose the credibility signal native-language reply gives. The downstream readers of your TA / Google profile see a same-language reply as evidence you take that language community seriously — useful for tourist venues, essential for venues in multilingual cities.

Rule 4 — Avoid the "thank you for your feedback" trap

The single most-deployed AI reply opener is "Thank you for your feedback." Every operator uses it; reviewers recognize it instantly as boilerplate. A reply that opens with a specific reference to their experience ("We're so sorry the kebabs arrived cold — that's not how we send them out of the kitchen") reads as personally written even when AI-drafted.

The prompt config should explicitly forbid generic openers.

Rule 5 — Don't engage the specifics publicly when liability is real

If a review claims food poisoning, allergic reaction, or anything that could become a legal claim, the AI should NOT engage the specifics in the public reply. Instead, invite the reviewer to share details privately — "We take this very seriously and would like to look into it. Could you DM us with the date of your visit?" That keeps the dispute off the public profile while showing future readers you respond responsibly.

Some review-management tools let operators flip a "move to DM" toggle per venue that changes AI behavior on health/safety reviews. Use it.

The half-second test

Before approving any AI-drafted reply to a negative review, read it imagining the reviewer's best friend just read it too. Does it sound generic? Sincere? Will it make the reviewer feel heard, or feel handled? If "handled," rewrite.