In this article
A proper food photography shoot runs €150 to €300 per dish. A growing number of independent restaurants skip that cost and let ChatGPT, Midjourney or Canva's AI tools generate the photo instead — and nobody has ever priced what happens when that photo looks better than the dish that actually reaches the table.
It's an understandable choice. Hiring a photographer means blocking out half a shift around the light, someone rebuilding plates until they look flawless, and an invoice that runs into the thousands for a twenty-dish menu. An AI tool produces something that looks at least as good on a phone screen in two minutes, for free or a few euros a month. A growing number of restaurants already do this, and there's nothing wrong with the tool itself.
The problem is what an AI image is missing: physical constraints. A real plate has steam that dissipates, garnish that settles the moment it leaves the pass, a portion sized to what the kitchen can actually plate, and sauce swirls that are never perfectly symmetrical. A generated image has none of that — it shows the dish at its absolute best, glistening moment, under lighting that exists in no real kitchen. That's exactly why an AI photo can look better than the plate will ever be.
On a menu with a QR code, or in a delivery app, a guest orders based on precisely that photo. When the plate then looks less impressive than the picture promised, that's not just a disappointment — it's the start of a chain that can actually be priced: a redo cost, a risk to your rating, and in the worst case a problem with the platform you're listed on. Nothing on this site had ever put a number on that chain.
This article does, in four steps, each with its own separately-stacking cost. The calculator below lets you plug in your own dish count, your own order volume and your own complaint rate to see exactly what the gap between photo and plate costs you — and what it costs to close it.
What does the law say about this?
In most EU countries, misleadingly depicting what you're selling — including how a dish looks — falls under the national consumer-protection law implementing the EU's Unfair Commercial Practices Directive (2005/29/EC). That directive tests an image against one plain question: would an average guest have ordered differently if they'd seen the real plate first? For a photo that meaningfully diverges from what actually arrives, the answer is often yes.
This isn't a new rule written specifically for AI — it's the same test that has applied for years to any misleading product photo in any industry, now applied to a new tool for making that photo. The instrument used to create the image changes nothing about whether the image matches reality.
This is not legal advice: how this is transposed into national law, which authority enforces it, and what the penalties are, all differ by member state. Check your own country's rules if in doubt, or ask a lawyer — this article explains why the risk exists, not what would happen in your specific case.
The ultimate guide Digital & Data: Every Guide for Your Restaurant From AI applications to delivery apps: everything digitalisation can do for your restaurant, in one overview. Open the guideThe 4 numbers behind the risk
The four numbers below stack in order: the first is the cost of a single incident, the last is the choice that decides whether you ever encounter the first three at all. They're separately additive precisely because each prices a different moment in the same chain.
1. The redo cost of one order that's 'not as pictured'
A guest who orders through a delivery app based on a photo and then receives a disappointing plate does one of two things: asks for it to be remade, or files a complaint for a refund. Both cost money, and most restaurants only count the most visible cost — the refund itself — and miss the rest.
Run the example. A pasta dish with €4 of ingredients and fifteen minutes of prep time (roughly €6 in labour at an average loaded hourly wage) already costs you €10 just to redo. Add a full refund of the order value on top because the guest doesn't want it anymore, and you add those two figures together — it isn't one or the other.
On a delivery order there's often a third layer: the platform itself. When a guest opens a dispute because the dish didn't match the description or photo, the platform typically fronts the refund and then recovers all or part of it from your next payout. The exact mechanism and handling fee differ by platform and by case, but the effect is the same: you end up paying, often without ever getting a chance to defend the dish first.
That's the first, most direct cost — and the calculator further down this article starts exactly here: how many of these incidents do you have per week, and what does each one actually cost you?
The bigger the difference between what the photo promises and what arrives, the higher the cost that follows.
Illustrative, not a measured percentage: the bars show the relative size of the gap, not an exact measurement. The cost labels are real, though — they're the consequences described in numbers 1 through 3 above.
2. The revenue a lower rating costs you
One repeated incident is annoying. What makes it expensive is what comes after: a disappointed guest rarely writes a complaint to the kitchen — they write a review. And reviews stack up into an average rating, the single number every future visitor bases their decision on before they've seen a single dish.
Research by Michael Luca (Harvard Business School), based on Yelp data for independent restaurants, found that one additional star on a review platform can lift revenue by roughly 5 to 9%. That's a rule of thumb about a real, measured relationship between rating and revenue at independent restaurants — not an exact law that plays out identically for every business, but solid enough to take the reverse movement seriously: a structurally declining rating can take a comparable bite out of revenue.
On delivery platforms there's a second layer on top. Deliveroo, Uber Eats and Just Eat all state in their own partner documentation that a restaurant's rating factors into how visible it appears within a category search, without publishing an exact percentage. A falling rating isn't just a cosmetic problem on your profile — it partly decides how often you're even shown to a guest who hasn't chosen yet.
3. The risk of platform policy and reduced visibility
Major delivery platforms require in their partner terms that the photos on a listing give a realistic representation of what a guest actually receives. That's documented policy, not an unwritten rule — it's in the guidelines every restaurant accepts when it signs up.
What exactly happens after a violation differs by platform and escalates with repetition: a single complaint typically triggers nothing visible, but a pattern of 'not as described' reports can lead to an internal review of the listing, a requirement to replace certain photos, or in persistent cases a restriction on how or how often the listing is shown. No platform publishes a fixed number of complaints that triggers this — it's a rising risk, not a fixed threshold.
You price that cost with your own numbers, not with a published percentage. On 120 orders a week, one day of reduced visibility already means roughly seventeen missed orders — run that against your own average order value and it quickly becomes clear why platforms take this seriously, and why you should too.
The chain from one disappointing photo to fewer orders
Each step costs a share of the last. This is a chain, not a fixed probability — not every step follows automatically from the one before.
Number 1: the direct redo cost per incident — see above, and calculate your own count in the tool.
Number 2: research by Michael Luca (Harvard Business School) links one star to roughly 5-9% revenue at independent restaurants.
Number 3: delivery platforms explicitly name rating as a visibility factor, without publishing a fixed percentage.
The three annotated steps carry a concrete figure or a documented source; the rest of the chain is a sequence, not a measured conversion rate.
4. The real alternative: what three options actually cost
There are three ways to get a photo, and they differ in more than price alone. The first is professional food photography: €150 to €300 per dish, one-time, with a photographer who controls light, styling and composition so the plate looks its best without ever lying about what actually reaches the table.
The second is a correctly lit smartphone photo, and that costs almost nothing. Four habits make the difference: natural window light instead of the restaurant's own ceiling lights, a matte plate or a neutral, non-reflective background, the guest's eye-level angle at the table rather than shooting straight down for a hot dish (a flatlay works for cold, flat food — not for something meant to steam), and never the built-in flash, which produces flat, unnatural light.
The third is an AI-generated image, and that also costs almost nothing — but it carries the full risk of numbers 1 through 3 above, because it isn't a photo of your dish. It's a guess at what your dish might look like.
The conclusion isn't a ban, it's a boundary: AI images are excellent for mood, decor and marketing backgrounds — an empty table setting, a sunset terrace shot, a social media banner. They don't belong on the specific dish a guest orders and then sets next to the real plate to compare. That exact moment of comparison is precisely what sets numbers 1 through 3 in motion.
Run your own risk numbers
Fill in the six fields below with your own situation: how many dishes on your menu or delivery listing carry an AI-generated or heavily retouched photo, how many delivery orders you get per week, your average order value, your estimated 'not as pictured' complaint rate, what an incident typically costs you to fix, and your current average rating.
The calculator shows three things: the estimated annual cost of the complaints themselves, an indicative range for the revenue at risk from rating erosion, and what it would cost to replace your AI photos with professional ones. Everything runs in your own browser — nothing is sent or stored.
What does the photo-to-plate gap cost you a year?
Six fields, your own numbers. Everything runs in your browser — nothing is sent or stored.
—
This is not legal or financial advice — it's an arithmetic exercise with your own numbers. The revenue-risk figure applies a cautious, explicitly stated assumption on top of a real research finding (see above); everything runs locally in your browser.
The revenue-at-risk figure is deliberately a range, not a precise number. Nobody — not even Luca's research — has measured exactly how many stars are lost to a given number of 'not as pictured' complaints; that varies by restaurant, by audience and by how visible the complaint becomes. What the calculator does is apply a cautious, explicitly stated assumption on top of Luca's measured 5–9% rule of thumb, so you get a realistic order of magnitude rather than a falsely precise figure.
The direct cost of the complaints themselves isn't up for the same kind of debate: that's arithmetic on your own numbers, not a research finding. If that cost alone pays back professional photos within a few months, the decision is easy. If it doesn't, the smartphone approach from number 4 above is your cheapest first step.
What to actually do today
This doesn't have to become a big project. Most restaurants don't have twenty problematic photos — they have a handful of dishes where the gap between photo and plate is widest, and those are often exactly the dishes ordered most, because an appealing photo simply drives more orders.
Start there. Go through your menu and your delivery listings, mark which photos are AI-generated or heavily retouched, and replace the photos of your best-selling dishes first with a correctly lit, honest photo. That's the shortest path to fewer complaints, not the most expensive one.
And keep the AI tools for what they're genuinely good at: mood shots, decor, a banner for your socials — anything that doesn't show a specific dish a guest will later set next to the real plate.
- Go through your menu and delivery listings and mark every photo that's AI-generated or heavily retouched.
- Start with the dishes that sell the most — that's where the gap between photo and plate costs you the most.
- Take a new photo of those dishes with natural window light, a matte plate or a neutral background, from the guest's eye level — no flatlay and no flash for a hot dish.
- Keep AI images for mood, decor and marketing backgrounds — never for the specific dish being ordered.
- Read the photo policy of every delivery platform you're listed on and compare it to what's currently online.
- Re-run your own numbers in the calculator above at least once a quarter — your photos, your order volume and your rating all change.