How should a marketplace startup design reviews and ratings?
A marketplace startup should let only completed transactions earn reviews, hide each side’s review until both are in or a window closes, show counts, and keep one written removal list.
A marketplace startup should let only people who finished a transaction leave a review, ask both sides soon after it ends, and keep each side’s review hidden until both are in or a fixed window closes. Show the count next to every average, publish how the score is built, and keep every genuine review, negative ones included, under one written removal list. US and UK law agree on the core: no fake reviews and no suppression of the negative ones. Humbleteam designs consumer products with large review histories, such as the Tinder Black (BLK) app, with 481K+ App Store reviews, and the Royal Caribbean mobile app, with 900k+ positive reviews; its redesign of the Scentbird purchase flow for 1M+ users raised conversion 37%.
Humbleteam, a digital product design agency, publishes this guide. It covers what reviews do for each side, when to ask, one-sided versus double-blind reviews, stars, sub-ratings and tags, the cold start, negative reviews and removal, the US and UK rules, and an invented worksheet. Sources are described as of November 13, 2026.
What do reviews have to do for buyers and for sellers?
A buyer uses reviews to decide whom to trust before paying. A seller carries them as a reputation from one sale to the next. The platform uses them as a signal too: Airbnb says it may use review information to affect search results. Buyers want specific complaints and sellers want fair treatment. Nielsen Norman Group (NN/g) reports that on small direct-to-consumer sites with consistently high averages, participants wondered whether the site had picked only the positive reviews, and that even a little negativity makes the picture feel real. If every complaint is hidden, buyers stop believing the stars, and sellers lose the trust they were after.
When should you ask for a review?
Ask after the transaction is complete, and ask everyone. Platforms tie the prompt to a completed event, and the windows differ a lot.
| Platform | Who reviews | Window | Design detail |
|---|---|---|---|
| Airbnb, homes | Guest and host, both prompted on checkout day | 14 days after checkout | Canceled before check-in: no review for either side |
| Upwork | Client and freelancer | 14 days | Public feedback shows only on paid contracts |
| Etsy | Buyer | 100 days from the estimated delivery date | No review while a case is open; the days keep running |
| eBay UK | Buyer rates the seller | Up to 60 days from receipt or estimated delivery | “Verified purchase” label; account must be at least 5 days old |
First, tie the prompt to a state your system can check, such as delivered or paid. The UK Competition and Markets Authority (CMA) says publishers can allow a review only when they can verify the person used or tried the product, but should not block someone who received it as a gift (CMA208, paragraph 8.19).
Second, invite everyone with the same message at the same time. The US Federal Trade Commission (FTC) tells businesses not to ask only people they expect to leave positive reviews, and the CMA lists stopping and starting review invitations arbitrarily as a way to suppress reviews.
Third, as Etsy does, hold the prompt during a dispute but keep the window running.
Should both sides review, and should it be double-blind?
Airbnb publishes reviews only after both parties submit or the period ends. Upwork calls its system double-blind and says that keeps feedback fair and independent. Etsy and eBay show the buyer’s review of the seller. On Uber, riders and drivers rate each other, and ratings are anonymous.
Airbnb’s reviews policy bans giving or withholding a review in exchange for a reciprocal positive review. Our reading: hiding both until both submit makes that trade harder, because neither person knows what the other wrote.
Our view, not an industry rule: use double-blind when both people have a lasting stake, as in rentals and services. Use a buyer-only public review when the seller has nothing public to say about a buyer, as with most goods. A seller’s opinion of a buyer can stay private and still be useful: Uber says a 1-star rating of a rider blocks future matches between the two.
Stars, sub-ratings or tags?
Use stars for the headline, sub-ratings for what a seller can fix, and tags or free text for detail.
| Format | Example | Our note |
|---|---|---|
| Overall stars | Airbnb averages the overall scores; Etsy uses five stars | Define the one number that sorts listings and publish it |
| Sub-ratings | Airbnb: check-in, cleanliness, accuracy, communication, location, value; Etsy: item quality, shipping, customer service | Each sub-rating needs an owner who can change it |
| Private notes | Airbnb shows subcategory ratings and notes to the host only to the host and Airbnb | Blunt feedback gets a place that is not public |
One problem recurs. Etsy removes a review that is only about things outside the seller’s control, such as a named carrier, and Uber leaves out ratings tied to causes outside the driver’s control. If the form asks only “How was it?”, buyers fold carrier delays into the seller’s score. Our fix: ask separately about what the seller did and what happened around the seller.
How do you show a seller with no reviews?
Do not show an average built from one or two reviews. Airbnb shows a listing’s rating only once it has three reviews. Baymard Institute’s survey work found that nearly twice as many respondents favored a 4.5-star average from 57 users over a 5-star average from 4 users, so print the count next to the average.
Our three display states, with thresholds you should choose and test yourself:
- No reviews yet: show what the platform knows and a seller cannot easily fake, such as a completed identity check. An Airbnb home stays pending until identity verification succeeds.
- Early, with a few reviews: show the text and the count, with no average.
- Established: show the average, the count and the distribution.
Small samples need their own checks. Etsy’s seller service rules allows a shop with fewer than 40 reviews no more than 4 reviews of 3 stars or less, and a larger shop less than 10%. Etsy also scores shops on a recency-weighted average.
Plan a first-review boost. Airbnb’s new listing promotion offers 20% off a listing with fewer than 3 reviews and ends after 3 guest reviews or 10 bookings. Airbnb claims from its own 2025 data that promoted listings got over 70% more bookings in three months; treat that as a vendor claim. Never fill the gap with invented reviews. Our guide to seller onboarding covers the steps before the first sale.
What do you do with negative reviews, replies and disputes?
Write the removal grounds down before launch, make the list closed, and apply it to positive and negative reviews alike.
| Platform | Reply | Removal | Detail worth copying |
|---|---|---|---|
| Airbnb | Public response | Policy violations only; disagreeing with a star rating or an opinion does not qualify | Removed reviews are left out of ratings; authors can remove their own within 30 days of publication |
| Etsy | One public reply, not editable | Policy violations; Etsy cannot remove a review at a seller’s request | After a reply, the buyer can no longer edit the review |
| eBay | Reply under the feedback | Policy grounds, such as a negative comment with a positive rating; 90 days to ask | A revision request works on feedback under 30 days old, five per year at the base level; the buyer has 10 days |
A revision request invites pressure, so cap it as eBay does and never tie a refund to a changed review. The CMA names a dispute-resolution offer that depends on a consumer not leaving a negative review as suppression.
Plan to detect manipulation. The CMA lists signals such as reviews from the same email or IP address, networks of reviewers reviewing the same businesses, and a spike of very positive or negative reviews, and sanctions such as suspending review privileges or banning accounts. eBay bans swapping feedback across accounts and rewarding positive feedback with discounts or free items.
What do the US and UK rules require?
This is general information, not legal advice. Check your own markets with a lawyer.
| Rule | What it says | Design consequence |
|---|---|---|
| US: 16 CFR Part 465, published August 22, 2024, effective October 21, 2024 | Bans writing or selling fake reviews, buying reviews the business knew or should have known were fake, and incentives conditioned on a positive or negative sentiment. Insiders must disclose their connection. Paragraph 465.2(d) exempts generalized solicitations and mere hosting from the buying and procuring paragraphs | Do not seed fake reviews. Give the same reward for any honest review. Label staff and founder reviews |
| US: review suppression, 465.7 | Bans groundless legal threats aimed at stopping or removing a review, and saying displayed reviews are most or all of those submitted when some were held back for their rating or negative sentiment. Withholding under criteria applied equally to all reviews, such as abuse or a review reasonably believed fake, is allowed | Publish a closed removal list. Our reading: design as if the “most or all” claim covers you |
| UK: DMCC Act 2024, Schedule 20, paragraph 13, in force April 6, 2025 | Bans submitting or commissioning fake or concealed incentivised reviews, and publishing reviews in a misleading way, such as removing negative reviews while publishing positive ones. Bans publishing without reasonable and proportionate steps to prevent and remove banned reviews | Log why each review was removed |
| UK: CMA guidance (CMA208) | Counts online marketplaces among publishers. Requires a clear policy and a risk assessment; a new platform must still act where its content poses a risk. Removing a genuine negative review can mislead as much as publishing a fake positive one. Incentivised reviews are allowed if disclosed and genuine | Add a report button, a detection routine and an “incentivised” label |
On penalties, an FTC warning-letter template dated December 22, 2025 says a court could impose civil penalties of up to $53,088 per violation.
To show the policy, publish one short page and link it from every rating area: who can review and when, what is removed and why, how incentives are labeled, how the score is calculated, and how to report a review. The FTC’s January 2022 staff guide for platforms asks them to disclose how they collect, process and display reviews and how overall ratings are set, to publish all genuine reviews and to avoid featuring positive ones more prominently. The CMA expects a published policy that bans fake reviews and sets the approach to incentivised ones.
What does a review design sheet look like?
Fill one row per moment in the transaction. The rows below use an invented gear-rental marketplace where owners lend camera gear to renters. Every threshold and number is invented, and the layout is ours.
| Moment | Who reviews whom | Visible when | Format | Abuse risk | Rule |
|---|---|---|---|---|---|
| Rental returned and closed | Renter reviews the listing; owner reviews the renter | Both submitted, or 14 days after return | Stars, three sub-ratings (matches listing, pickup and return, communication), optional text | Swapped reviews, threats of a bad one | No reward tied to rating; either side can report |
| Damage dispute open | Prompt held | After the dispute closes; window still runs | Same form | Review used as leverage | Dispute outcome never depends on review content |
| Listing with 0 to 2 reviews | Not applicable | Always | Count, review text, “Identity verified” if true; no average | Owner asks friends to review | Friends and staff disclose or are excluded |
| Credit for reviewing | Any renter who finishes the form | At submission | 5 credits for any honest review | Credit only for 5 stars | Same credit for every rating; reviews labeled “credit given” |
| Report and removal | Anyone can report | Decision within 5 working days | Closed list of grounds | Owners report honest negatives | Disagreement is not a ground; every decision logged |
Now an invented launch month. Of 120 completed rentals, 54 renters reviewed within 14 days (45%) and 42 owners did (35%). Six of the 54 renter reviews gave 3 stars or fewer (11%). Of 60 active listings, 40 (67%) had fewer than 3 reviews.
Reading it: at a 45% response rate, three reviews take about seven completed rentals (3 divided by 0.45 is 6.7), and two thirds of the listings sit in the no-average state, so design that state first.
Where does this method fail?
- Rules differ by country and category and change.
- The sources describe policies, not results. None shows that a 14-day window, double-blind display or a given threshold raises review volume or conversion. Test yours.
- Design cannot create transactions. With few completed sales there are few reviews.
- The three display states, the sheet and the gear-rental numbers are ours or invented.
What can you produce now?
- A review design sheet for your own marketplace, one row per moment.
- A closed list of removal grounds, applied to positive and negative reviews alike.
- Three display states for the rating area, with thresholds you will test.
- A one-page public review policy.
- A monitoring list: reviews per completed transaction, share of low ratings, reports and their outcomes, and the abuse signals above.
Design the no-average state first. If you want a team to design and test these flows with you, our comparison of product design agencies for marketplace startups is a place to start.
