--- name: review_response_writing description: "Answer public reviews well: triage every review by rating and risk before drafting, write case-specific replies for happy, mixed, and angry customers, flag safety, legal, privacy, and discrimination claims for escalation instead of answering them in public, and run a weekly tracking routine that finds repeat problems. Use when replying to Google, Yelp, Tripadvisor, app store, or marketplace reviews, when a negative review needs a careful answer, when a review may be fake or policy breaking, or when setting up who replies, how fast, and what gets logged." --- # Review Response Writing ## Purpose Turn review replies from a chore done under stress into a steady system. Every review gets triaged before anyone writes a word, risky reviews leave the public thread and go to a named owner, and ordinary reviews get a short, specific reply that a future customer can trust. The public reply has two readers, the reviewer and everyone who finds that review later. Write for both, argue with neither. This skill covers answering reviews only. Getting more reviews, asking for reviews, and review gating are covered by the review generation skill. If a request mixes the ask and the answer, do the compliance check there first, then come back here for the reply. ## Workflow 1. **Collect before you reply.** Pull new reviews from every platform the business is listed on into one queue. Record platform, date, rating, reviewer name as shown, and the full text. Never reply from a screenshot someone forwarded without checking the live review, because edited or partial text changes the right answer. 2. **Triage by rating, then by risk.** Assign a lane first, then scan for risk flags that override the lane. Full matrix in `references/triage-and-risk.md`, and `bin/review_triage.py` scores a single review or a CSV export the same way. - Lane A, praise: 5 stars, or 4 to 5 stars with no complaint in the text. - Lane B, mixed: 3 stars, or a high rating with a real complaint inside it. - Lane C, unhappy: 1 to 2 stars, or any rating where the reviewer describes a failed service, a wrong order, a rude interaction, or money lost. - Lane D, hold for escalation: any review, at any rating, that mentions injury, illness, unsafe food or premises, discrimination, harassment, a child, theft, a threat, a lawyer, private personal data, or a demand for money to remove the review. Do not draft a public reply for Lane D until the owner named in the escalation step clears it. 3. **Check the record before drafting a Lane B or C reply.** Look up the visit, order, booking, or ticket the review describes. Write down what the record confirms, what it contradicts, and what it cannot settle. Google advises this check in its own response guidance. If there is no record at all, treat the review as unconfirmed, not as proven fake, and use the no record template. 4. **Pick the route the platform allows.** On Google, the reply is public and sits under the review. On Yelp, choose a public comment or a direct message, and for critical reviews Yelp advises a public comment first so readers can see the concern was addressed, then a direct message if more detail is needed. On Tripadvisor there is one management response per review, so coordinate who sends it and get it right the first time. Never paste the same reply into two routes on the same platform. 5. **Draft from the case template, then personalise it.** Start from the matching case in `references/case-templates.md`. Replace every placeholder with a fact from the review or the record. Use the reviewer's first name as shown on the platform, name one specific thing they mentioned, and sign with a first name or initials and a role. A reply with a placeholder left in, or with the wrong customer's detail copied in, is worse than a late reply. Delete it and start again. 6. **Keep the shape tight.** A working shape for most replies is four moves in order: thank them by name, name the specific point, own or explain the part that is yours, give one clear next step. Working lengths set by this skill, not platform rules: 25 to 60 words for praise, 50 to 110 words for mixed or unhappy reviews. Longer replies usually mean someone is relitigating the visit in public. Cut the defence, keep the next step. 7. **Run the never-argue check before posting.** Read the draft as a stranger would. Cut any line that disputes the reviewer's memory, explains why they are wrong, blames a named staff member, shares private details about their visit, offers a discount or free item in public, asks them to change or remove the review, or promises a fix with a date you cannot guarantee. If two drafts in a row come out defensive, stop, wait at least an hour, and draft again cold. 8. **Escalate Lane D in the same working session.** Send the review text, the record check, and your draft notes to the owner, manager, or adviser named for that risk type in `references/triage-and-risk.md`. Log the handoff time. If the platform lets you post a holding reply before advice arrives, use only the short holding template from the case file, which thanks the reviewer, says the matter is being reviewed by a named role, and gives a direct contact. Nothing else goes public. 9. **Handle policy breaking reviews through the platform tool.** Flag a review that is fake, off topic, from a competitor or an insider, or that contains hate, threats, or private data, and cite the specific policy it breaks. While the flag is pending, post a calm unconfirmed review reply if the platform shows the review publicly, so readers are not left with silence. Do not threaten the reviewer, do not offer anything for removal, and do not post a second accusatory reply if the flag is denied. The denial closes that route. 10. **Log every reply and close the loop weekly.** Record lane, risk flags, template used, reply date, responder, escalation owner if any, and whether the reviewer replied, updated the review, or went quiet. Run the weekly routine in `references/weekly-routine-and-tracking.md` to count unanswered reviews, median reply time, repeat complaint themes, and cases still open with an owner. A theme that appears three times in a month becomes an operations fix with a named owner, not a better template. ## Output Contract When this skill answers reviews or sets up a response system, the deliverable includes: - **Triage table:** one row per review with platform, rating, lane, risk flags, record check result, and the route chosen, so anyone can see why a reply was written the way it was. - **Ready to post replies:** one draft per cleared review, personalised with the reviewer's name and one true detail, signed, inside the word range for its lane, with no placeholder text left in. - **Escalation list:** every Lane D review kept separate from the postable drafts, with the risk type, the owner it went to, the handoff time, and whether a holding reply was cleared. - **Platform actions:** any review flagged for removal, with the platform, the policy cited, and the date flagged. Flagged and answered are tracked as two different states. - **Weekly log update:** counts of new, answered, and still open reviews, median time to first reply, repeat themes, and the one operations fix the week produced, in the format in `references/weekly-routine-and-tracking.md`. ## Operating Rules 1. Never argue in public. No disputing memory, no sarcasm, no correcting grammar, no posting receipts or record details to prove a reviewer wrong. State your side once, briefly, only when a future reader needs a material fact, and move to the next step. 2. Never share private information in a reply. No visit times beyond what the reviewer stated, no order totals, no health details, no addresses, no staff member's full name. Google, Yelp, and Tripadvisor all bar personal data in responses. 3. Never offer compensation in a public reply, and never offer anything in exchange for editing or removing a review. Compensation, if a business chooses it at all, happens in a private conversation with no condition attached to the review. Tying an incentive to a review's sentiment breaks the FTC Rule on the Use of Consumer Reviews and Testimonials, 16 CFR Part 465, effective October 21, 2024, and incentivising a change to a negative review also breaks Google policy. 4. Never threaten, pressure, or invoke lawyers to get a review changed or removed. Review suppression, including groundless legal threats and intimidation, is banned by the same FTC rule. Real legal claims go to a lawyer privately, they never appear in a reply. 5. Apologise for the experience without inventing fault. Say you are sorry the visit fell short and own the parts the record confirms. Do not admit to events that did not happen, and do not apologise in a way that reads as dismissive, such as sorry you feel that way with nothing after it. 6. One voice, one responder per review. Decide who replies on each platform and keep that person consistent. On platforms that allow only one response, two staff members replying creates contradictions readers can see. 7. Reply to praise as reliably as to complaints. Thanking happy customers by name, with the detail they praised, is the cheapest retention work in the system, and skipping praise teaches reviewers that only anger gets attention. 8. Do not copy a template word for word twice in a row on the same listing. Readers scroll several replies at once, and identical text reads as automated. Templates set the shape, the details must come from the review in front of you. 9. Keep response targets as internal discipline, not public claims. A common working target is a first reply within one business day for unhappy reviews and two business days for praise. Report the time you actually hit, and never advertise a response promise the team has not kept for a full month. 10. If you cannot verify a fact, leave it out of the reply. Unconfirmed means the reply says the record could not be matched and invites direct contact. It never says the review is fake.