twimco.Let’s talk ↗︎
Menu
Field notes / Commerce

A marketplace review should come from a finished order

On a marketplace, stars are the thing buyers trust most and question least. If anyone can leave one, sellers will find a way to fill them in. Tie each review to a real, completed transaction and most of the hard problems get smaller.

Editorial engraving of a sunny apartment hallway where a buyer kneels beside an opened box with a glowing vintage lamp and rates the order on her phone.

Let people review only what they actually bought or booked, after it was delivered or completed. Moderate by written rules that apply the same way to five stars and one star. Let sellers reply but not remove, ask every buyer rather than only happy ones, and never reward a particular rating. The FTC's 2024 rule on fake reviews sets a floor for the businesses involved; a review system built on real orders makes it easier to stay above it.

The argument

On a marketplace, the review is the product page's most persuasive part and its least checked. A buyer can't inspect the seller. They read the stars and a few lines from strangers and decide.

So the rule for who gets to leave a review matters more than any ranking formula. Our argument is simple: a review should come from a finished order. Someone bought this item, or booked this service, through your marketplace, it was delivered or completed, and now they get to say how it went. Not before. Not without an order. Not twice for the same purchase.

A star rating is only as honest as the rule for who gets to leave one.

Yes, you'll have fewer reviews at first. A new listing will sit at zero for longer, and some sellers will complain. That's the price, and it's worth paying. Twenty reviews from people who actually bought the thing are worth more to a buyer than two hundred from people who might not have, and buyers can tell the difference faster than most operators expect.

What open reviews invite

When anyone with an account can review anything, a few things happen, slowly and then all at once.

Sellers create accounts and review themselves. Friends and family help out. A competitor leaves one-star reviews on a rival's best product. Someone sells review packages. A seller offers a discount "for an honest five-star review." Each of these looks, on the page, exactly like a real buyer.

Your team ends up playing detective: matching IP addresses, guessing which accounts are related, arguing with sellers about whether a review is real. That's expensive and never finished.

There's also the law. The FTC's rule on consumer reviews, in effect since October 21, 2024, prohibits businesses from writing, creating or selling reviews from people who don't exist, didn't use the product, or misrepresent their experience. It prohibits buying such reviews, and offering rewards conditioned on a particular sentiment, positive or negative. Officers and managers reviewing their own business need to disclose the relationship. How the rule applies to you as the operator, versus your sellers, is a question for your lawyer. But a system where reviews can only come from real transactions makes most of those violations harder to commit on your platform in the first place.

Decision guide

Open reviews or order-based reviews

ConsiderationAnyone can reviewReviews from finished orders
WhoAny account, including the seller's friends.The buyer on a completed order.
WhenAny time, even before delivery.After delivery, within a window.
WhatWhatever listing they choose.The item or service on that order.
Team effortDetective work on suspicious accounts.Rules checked against order data.
Decision framework; assess against your own marketplace and counsel's advice. Not measured results.
A star rating is only as honest as the rule for who gets to leave one.

What a finished order gives you

Tie the review to the order record and a lot of decisions become data, not detective work.

Who: the buyer on the order, and only them.

When: after delivery or completion, plus whatever window makes sense, so people review the experience rather than the checkout. Close the window after a while, too; a review two years later is about something else.

What: the specific item, variant or service on that order. That stops the classic trick of moving a well-reviewed listing's reviews to a different product.

How many: one per order line. Repeat purchases can earn repeat reviews, which is honest; one purchase earns one.

What about returns: a buyer who returned the item still had an experience worth reviewing, often the most useful one. An order cancelled before it shipped didn't, so it shouldn't unlock a review.

What label to show: a plain "verified order" marker that means exactly that, and nothing more.

Moderation that doesn't hide the bad ones

Even with verified orders, you'll need to remove some reviews. Write the rules down and apply them the same way to every rating.

The FTC rule draws a useful line. It treats it as deceptive for a business to suggest that the reviews it displays are most or all of the reviews it received when it's suppressing some because of their rating or negative sentiment. It lists grounds for withholding that aren't about sentiment, applied equally to all reviews: content that's defamatory, harassing, abusive or obscene; someone else's personal information; discriminatory content; trade secrets; content that's clearly false; reviews the seller reasonably believes are fake; and reviews wholly unrelated to the product.

That's a good starting policy whether or not the rule applies to you directly. Publish it. Log every removal with the reason. And check, once a month, whether the reviews you removed skew low. If they do, something in your process is judging sentiment, even if nobody meant it to. A common culprit is an automated filter for "abusive" language that trips mostly on angry customers, or a moderator who gives five-star reviews the benefit of the doubt and one-star reviews a second look. Sample both piles side by side and the bias usually shows up quickly.

How the work moves

From order to published review

  1. Completed

    Order delivered or service finished.

  2. Invite

    Every buyer asked the same way.

  3. Review

    One per order line, within the window.

  4. Moderate

    Published rules applied to every rating; removals logged.

  5. Reply

    Seller may respond, never remove.

Illustrative workflow, not a client screenshot; moderation grounds drawn from 16 CFR 465.7 as a source, not legal advice.

Sellers, incentives and the hard cases

Seller replies. Let sellers respond publicly to any review. Don't let them edit, hide or remove one. A calm reply to a bad review often does more for a seller than a dozen good ones.

Disputes. Give sellers a way to report a review they think breaks the rules, and decide it against the published policy. "This hurts my rating" is not a reason.

Asking for reviews. Ask every buyer after every completed order, the same way. Sellers who message only their happy customers are curating. Rewards tied to a particular rating aren't allowed under the FTC rule; if you offer anything for reviews at all, make it the same for every review, whatever it says, and ask your lawyer first.

When buyer-only is too strict. Some marketplaces arrange the introduction and the transaction happens elsewhere, like a contractor booked through you and paid in person. You can still tie reviews to a confirmed booking or a completed job record. Build that record; it's useful for much more than reviews.

Put this into practice

Pull your last hundred reviews. For each, can you point to the order it came from? Count the ones you can't. Then pull your last hundred removed reviews and look at their star ratings. Those two numbers will tell you where to start.

We build marketplaces where the transaction record is the backbone, so reviews, disputes and payouts all trace back to something real. If your reviews have become a moderation problem, talk to us. For the dispute side, see the evidence pack sellers need before the clock runs out.

Related work: Paddle8.

twimco.

We build alongside the people doing the work.
Bring us the problem you’re thinking about.

Related work

Paddle8 ↗︎