Should I let AI write all my review responses?

Let it draft all of them. Auto-send only the four- and five-star replies. Hold every one-, two-, and three-star reply for a human before it goes out. In our cohort about one in five AI drafts to negative reviews was too defensive or conceded something the operator didn't want to offer. Those need a person.

Won't guests notice the replies are written by AI?

On a five-star thank-you, no, and it doesn't matter — nobody re-reads a 'thanks, see you soon.' On a complaint, yes, and it matters a lot. A generic AI apology to an angry guest reads as a brush-off and makes the thread worse in public. That's exactly why the negative ones get a human.

What's the actual win if a human still has to touch the bad ones?

Speed and coverage. Reply time across the cohort went from about three days to under five hours, and response rate went from roughly 40% to 100%. The AI clears the easy 80% in seconds so your manager spends their time only on the handful of replies that can actually cost you a guest.

AI can answer your five-star reviews. Keep it away from your one-stars.

We let AI answer 30 days of guest reviews across the cohort.

On the five-star reviews, it was the best hire we made all quarter. On a one-star, it nearly cost a venue a fifteen-year regular.

Both things are true. The play is knowing the difference.

What we tested

Forty-one venues. Every Google and Yelp review for 30 days, drafted by AI the moment it posted. The operator saw each draft in a queue and could send, edit, or kill it.

Going in, the fear was the obvious one → robotic replies, guests feeling brushed off, a brand voice that sounds like a help desk.

What actually happened was more useful than that, and more dangerous.

The five-stars: send them all

Nobody agonizes over a five-star reply, but everybody skips them.

That's the real problem. Across the cohort, before AI, the average venue replied to 38% of its reviews. The five-stars got ignored because the staff was busy and a happy guest didn't feel urgent.

But replying to a five-star review is not for that guest. It's for the next 200 people reading the page. A wall of glowing reviews with zero owner replies reads as absentee. The same wall with a warm, specific reply under each one reads as a place that's paying attention.

AI is perfect for this. It pulls the guest's name, references the dish or the server they mentioned, keeps it short, and sounds human. Reply time went from three days to under five hours. Response rate went from 38% to 100%.

That's the whole win, and it's a big one. Auto-send the four- and five-stars. Don't even read them.

The one-stars: this is where it bites

Here's the near-miss.

A guest left a one-star after a botched anniversary dinner. Real complaint, fair, emotional. The AI drafted a reply in four seconds. It was polite. It apologized. And it offered a free dinner to make it right.

The operator caught it in the queue. That guest was a regular — fifteen years, knew the owner by name. A canned "so sorry, here's a free meal" from a bot would have read as exactly what it was → a machine processing him like a ticket. He didn't want a comp. He wanted the owner to call him. Which the owner did.

That reply, auto-sent, doesn't lose one guest. It loses the guest and it loses every reader who watches a real complaint get handled by a robot in public.

Across the cohort, roughly 1 in 5 AI drafts to one- and two-star reviews had a problem → too defensive, conceding a refund nobody authorized, or missing context only the operator had. One in five is not rare. One in five is a coin flip you can't afford to lose in public.

The three-stars: the ones worth your time

The middle review is the one everyone forgets.

A three-star is a guest who almost liked you. They came, something was off, and they're telling you in public instead of walking away quietly. That's a gift. It's also the only review category where a thoughtful owner reply can flip the rating — we watched it happen 30-some times across the month.

AI can draft the three-star reply. A human should read it before it sends. This is the cheapest conversion in the building, and it's worth ten seconds of a manager's eyes.

The play

Draft everything. Gate by rating.

→ Four and five stars: auto-send, no human needed. → Three stars: hold for a glance — these are the ones you can actually convert. → One and two stars: a human writes the final, every time. AI can draft a starting point, but a person owns what goes out.

The negative review is the one a hundred future guests will read most closely. It is the single most expensive sentence your venue publishes all month. That is not a place to save four seconds.

What this is really about

AI in hospitality isn't a question of on or off. It's a question of where the human goes.

Put the human where the stakes are. The five-star reply has no downside, so automate it completely and reclaim the hours. The one-star reply can cost you a regular and a reputation, so that's where your manager's attention belongs — not on typing "thanks, glad you enjoyed it" 40 times a week.

Most operators have it backwards. They hand-write the easy ones when they get to them, and the hard ones sit for three days while everyone hopes they'll disappear.

Flip it. Let the machine handle the gratitude. Keep the apology human.

— Damon