How to ask customers for reviews: the errors that cost margin and the Masterestaurant method

Chapter 1 — The 90-minute window is the variable that decides everything else

Ask for the review within 90 minutes of the meal, because that is when the guest still remembers the texture of the dish and can write something concrete, which is exactly what a language model can quote later. The channel matters as much as the clock: 97% of SMS messages are read within 15 minutes of sending, according to Tabular (SMS Marketing Stats 2025), so a message sent at 9:40 p.m. to someone who closed their tab at 9:15 arrives while the memory still has detail. Compare that with the email that lands the following Tuesday, on a guest who no longer recalls whether the octopus was cooked right or overdone. The operation loses the social proof it manufactures every night, and it manufactures it anyway, asked for or not. An old review costs money because the diner reads it as expired information and the algorithm weights it lower, even if your average stays intact.

Chapter 2 — Why does an old review cost money even when the star rating holds?

Some 92% of diners read reviews before choosing where to eat, according to Restroworks (Google Restaurant Search Statistics 2024), and 71% read them specifically on Google before deciding, per BrightLocal (Local Consumer Review Survey 2024).

That reader does not compute averages: they check the date, they check whether anyone named the dish they want to order, and they decide in twenty seconds. With sector net margins running 3 to 9% according to Statista, losing two or three tables a night because the last review is eight months old is not an image problem, it is the entire margin of the shift. Reputation rusts on a calendar schedule. Below 500 thousand USD a year, the owner works the floor and can ask for the review personally: five well-made requests a night are enough, with no platform and no cost.

Chapter 3 — The effect by revenue band is not proportional, it comes in steps

Between 500 thousand and 1 million the first break appears, because the owner no longer covers every shift and needs the POS to fire the SMS; that is where the 97% read-within-15-minutes figure reported by Tabular (SMS Marketing Stats 2025) stops being trivia and becomes infrastructure. Above 1 million, delivery carries weight: in a global food delivery market worth 288.84 billion USD in 2024 according to Grand View Research, your rating is the entry price to the listing. And from 5 million upward, reviews stop being a floor tactic and become multi-unit brand governance. The celebrity-chef restaurant or the large-format themed venue above 5 million USD a year does not suffer from a shortage of reviews but from reviews that describe the room instead of the food —ceiling photos, the noise, the queue— and that generic mass feeds nobody. Its costs are different: a dedicated community manager, moderation in two or three languages, and a reply to every one-star review inside 24 hours before a publication picks it up.

Chapter 4 — Above 5 million the problem inverts: plenty of volume, no specificity

Past 10 million we are talking about a dashboard with reviews by venue, by shift and by platform, cross-referenced against the sales mix. Restaurants active on social channels reported 9.9% higher direct B2C revenue in 2024 according to Deloitte Digital, and in this band that single percentage point runs to hundreds of thousands of dollars. "Leave us your opinion" generates four-word reviews that no diner reads and that no AI quotes when someone asks it where to have dinner nearby. Swap the request for a closed question about one concrete element —the dish they ordered, the server who took care of them, the wait time— and the text that comes back carries usable nouns. That difference matters more every year: 57% of millennials decide where to eat based on what they see on social, per the TouchBistro 2025 Diner Trends Report, and what they see are sentences with proper names in them, not loose stars.

Chapter 5 — The generic question produces garbage; the specific question produces assets

Diego F. Parra and the Masterestaurant team treat the request as part of the shift's technical sheet, with a written script and an assigned owner, because a question improvised by a tired server returns exactly what it cost: nothing. Paying for reviews with a 10% discount violates Google and delivery platform policy, contaminates your sample, and teaches the guest that their opinion has a price, which is the worst commercial lesson available. For years I recommended the soft incentive —a complimentary coffee— until it became clear that it attracts the person who wants the coffee, not the one who had something to say. If you genuinely want to reward the relationship, do it through loyalty, which is legitimate: 32% of members use their membership several times a week and 47% several times a month, according to LoyaltyPass (Restaurant Loyalty Statistics 2026). There the benefit buys frequency, not testimony. Operators in the 90th percentile draw more than 37% of transactions from loyalty members, per Paytronix (Loyalty Trends Report 2024).

Chapter 6 — What happens if the system fails right in the 1.4 million band

Take an operator at 1.4 million USD who lets a delivery rating slide from 4.5 to 4.3 stars and watches the average ticket drop 11%: applied to off-premise sales, which run near 75% of traffic according to Circana, that decline eats several times over the 3 to 9% sector net margin Statista reports. Now remove the review from search as well and the damage doubles, since 71% consult Google before deciding per BrightLocal (Local Consumer Review Survey 2024). The bridge between two ideas that look opposed —asking for reviews is soft marketing, yet it is paid in cash— sits right there: reputation is the only restaurant asset produced free every night and lost whole when nobody collects it. Set up the automatic 90-minute trigger tomorrow with one written question. TIMING. The traditional approach asks whenever someone remembers; the system asks inside the 90 minutes after the meal, while the memory of the dish still has texture and the guest can write something concrete.

Chapter 7 — The five differences that move cash

That single variable, with nothing else changed, is what separates single-digit response rates from double-digit ones. CHANNEL. The QR on the check is a blind channel: it does not know who saw it, it never retries and it leaves no trace. The SMS or WhatsApp of a guest who already paid does leave a trace, and the channel evidence is blunt — 97% of SMS messages are read within 15 minutes of being sent, per Tabular (SMS Marketing Stats 2025). SPECIFICITY. «Leave us your opinion» produces four-word reviews nobody reads and no language model quotes; «how was the seven-hour lamb Andrés served you?» produces fifty-word reviews with a dish name and a person's name, which is what feeds AI recommendation shortlists. SYMMETRY. Filtering unhappy guests before the ask —review gating— is the practice that has sunk the most profiles in the last three years, because platforms catch it by statistical pattern and pull the entire block.

Chapter 8 — The five differences that move cash — in practice

You ask everyone equally and you manage criticism afterwards, in public, under a 24-hour SLA. GOVERNANCE. With no named owner, no weekly target and no dashboard, any review programme lasts six weeks. With those three pieces it lasts years, and reputation stops being a quarterly panic and becomes a marketing budget line with its own return.