Most advice about positive reviews stops at "reply and say thank you". The more interesting thing a five-star review contains is a noun.
Guests who liked you name things. Not categories — not "the food was lovely" — but the short rib, the burnt cheesecake, the flat white. That noun is a guest voluntarily doing your menu merchandising for you, and it costs nothing to count.
The useful move is not to read those mentions. It is to rank them, then rank your sales mix, and look at the gap between the two positions. Four things can be true of any dish, and three of them require you to do something different.
The two lists
List one: what gets named. Take your last sixty five-star reviews. For each one, write down every menu item mentioned by name. One review can contribute three. Ignore adjectives entirely; you want the noun count. Twenty minutes with a notepad.
List two: what sells. Pull ninety days of item-level sales from the till. Units sold, not revenue — you are measuring how many guests chose it, not how much it cost.
Now the step that matters: convert mentions into a rate. Raw mention counts tell you almost nothing, because a dish you sell four hundred times a month will accumulate more mentions than a special you sold thirty times even if the special is the better dish. Divide mentions by hundreds of units sold.
Mentions per 100 sold = (named in reviews ÷ units sold) × 100
That single division is what makes the comparison honest, and it is the step almost everyone skips.
The worked example
A sixty-cover neighbourhood restaurant, ninety days, sixty five-star reviews:
| Dish | Units sold | Mentions | Mentions per 100 | Sales rank | Mention rank |
|---|---|---|---|---|---|
| Buttermilk chicken | 812 | 9 | 1.1 | 1 | 6 |
| Beef short rib | 164 | 21 | 12.8 | 7 | 1 |
| Burnt cheesecake | 243 | 17 | 7.0 | 5 | 2 |
| Cacio e pepe | 388 | 6 | 1.5 | 3 | 5 |
| Halloumi salad | 301 | 1 | 0.3 | 4 | 8 |
| Lamb flatbread | 97 | 6 | 6.2 | 8 | 3 |
| Fish of the day | 205 | 8 | 3.9 | 6 | 4 |
| Chips | 1,140 | 0 | 0.0 | 2 | 9 |
The short rib is mentioned at roughly twelve times the rate of the bestseller and sells a fifth as often. That is not a small discrepancy. That is a dish persuading people to write about you while sitting seventh on the menu.
The four quadrants, and what each one asks of you
Named often, sells often — your signature. You have one job here: do not break it. Lock the spec in writing, keep the garnish consistent when the usual chef is off, and never let it come off the menu for a supply wobble. In the example above nothing lives here, which is itself the finding.
Named often, sells rarely — the buried asset. This is where the money is. The short rib, the cheesecake and the flatbread all land here. Guests who order it tell other people about it; most guests never order it. That is a placement and a prompt problem, not a kitchen problem. Move it to the top-right of its section, cut the sentence describing it down to eight words, and give servers one line to say when a table hesitates. Expect the mention rate to fall as volume rises — it should, and the absolute number of advocates still goes up.
Named rarely, sells often — the default. Chips and the chicken. Nothing wrong with them; they are what people pick when they are not choosing. The trap is mistaking volume for affection and building your identity on them. A dish nobody writes about is not a reason to make the journey.
Named rarely, sells rarely — the cut list. The halloumi salad. Before you remove it, check whether it is load-bearing: the only vegetarian main, the only thing under £12, the dish one regular orders every week. Plenty of low-performing dishes are there to make a group booking possible. If it is not doing that job, it is costing you prep, fridge space and a line of menu attention.
What to do with the buried asset this week
One change, measured. Not a menu redesign.
- Pick the single dish with the widest gap between mention rank and sales rank.
- Change exactly one thing: its position on the page, or its description, or the server prompt. One, so you can tell which worked.
- Record its share of units sold for the fortnight before and the fortnight after.
- Watch the mention count too. If volume rises and mentions stay flat in absolute terms, you have found the ceiling — the dish was being ordered by the people most likely to love it, and you have now reached people who merely like it.
In the example, the restaurant moved the short rib to the head of the mains and gave it a seven-word description. Units went from 164 a quarter to 291. Mentions went to 26. The chicken lost about forty units, which is the honest cost and worth knowing.
Four ways this analysis lies to you
Mentions are biased towards the unusual. Anything photogenic, oddly named, on fire, or shareable gets named more than it deserves. A dish called burnt cheesecake has an advantage over chocolate pot that has nothing to do with how it tastes.
Sides and drinks are structurally invisible. Chips scored zero on 1,140 units. That does not mean the chips are bad; it means nobody writes a review about chips. Compare dishes to dishes and sides to sides, never across.
Sixty reviews is thin. With ninety days of five-star reviews you are working with counts in single digits for most items, and single-digit counts move around. Treat a gap of one or two rank positions as noise. The short rib moved six positions, which is why it is worth acting on.
Specials distort everything. A dish available for three weeks can post a spectacular mention rate off a tiny denominator. Exclude anything that was not on the menu for the whole period, or you will promote a dish you cannot reliably produce.
This is not the same as reading your four-star reviews. Praise tells you what to amplify. The caveat at the end of a four-star review tells you what quietly costs you returning guests, and that reads completely differently — the clause after "though" is where that lives. If ordering friction is what your reviews keep describing, the menu itself is the thing to look at: "we weren't sure what to order" is a revenue problem.
The honest reason this gets done once
The counting is tedious, the arithmetic is trivial, and the answer changes slowly. Which means it is a perfect candidate for doing once, getting a genuinely useful result, feeling pleased, and never doing again — at which point the menu changes, the chef changes, and the list is stale.
The version that keeps working is a standing count rather than a project. OMMU extracts dish nouns from reviews as they arrive and tracks mentions per hundred sold against your sales mix, so the comparison is a number you glance at rather than an afternoon you have to find. A spreadsheet and a disciplined Monday does the same job.