BRENNEN

Industries / Technology / Marketplaces and gig platforms

Trust and reviews

This is the work of letting strangers judge each other before they trade: ratings, written reviews, checks that reviews are real, and the rules against faking them.

6

Milestones achieved, from 2008 to 2024.

3

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 2008

    Workers review their clients

    On Mechanical Turk, businesses could reject a worker's submission, Amazon did not require them to answer workers' questions, and workers had no legal recourse against a requester who rejected work and then used it. In 2008 Lilly Irani and Six Silberman began Turkopticon, a website and browser add-on that shows workers other workers' reviews of each requester while they look for tasks. By their 2013 paper it had been installed more than 7,000 times and drew 100,000 page views a month.

    LessonWhen a platform rates only one side, the other side may build its own system without the platform.

    escholarship.orgwtf.tw

    Credit

    Lilly Irani and M. Six Silberman

  2. 2016

    Free-product reviews banned

    On 3 October 2016 Amazon changed its community guidelines to bar reviews written in exchange for free or discounted products, which it had allowed when disclosed, keeping only those from Amazon Vine, a program in which Amazon itself picks the reviewers. Sherry He, Brett Hollenbeck and Davide Proserpio later tracked about 1,500 Amazon products whose sellers bought fake reviews in private Facebook groups from October 2019 to June 2020. Ratings rose and fell back within 2 to 4 weeks, the share of one-star reviews rose once the buying stopped, and Amazon eventually deleted about a third of these products' reviews, on average more than 100 days after they were posted.

    LessonThe ban did not end paid reviews, and deletions that came months late left the short boost in place.

    techcrunch.commpra.ub.uni-muenchen.de

    Credit

    Amazon; Sherry He, Brett Hollenbeck and Davide Proserpio

  3. 2018

    Ratings drift upward

    On a large online labor market, the share of workers getting the top rating of 5 stars rose from 33% to 85% in six years, and four other marketplaces showed similar rises. When the platform also collected private feedback that workers never saw, private scores fell while public scores for the same jobs rose, and the authors estimate that more than half of the rise came from lower standards rather than better work.

    LessonRatings work because they can harm the person rated, and raters' reluctance to do that harm pushes ratings up until they stop telling sellers apart.

    nber.orgnber.org

    Credit

    Apostolos Filippas, John Horton and Joseph Golden

  4. 2019

    China bans faked sales

    China's E-Commerce Law, in force from 1 January 2019, bars online sellers from "fabricating transactions or user comments" to mislead buyers (Article 17). Article 39 requires platforms to run rating systems, publish their rules and let buyers comment, and bars platforms from deleting buyers' comments.

    subsites.chinadaily.com.cn

    Credit

    Standing Committee of the National People's Congress of China

  5. 2022

    Europe checks review pages

    In results published on 20 January 2022, consumer authorities in 26 EU countries, Iceland and Norway examined 223 major shops, marketplaces, booking and comparison sites, and could not confirm for 144 of them that reviews came from people who had used the product. From 28 May 2022, Directive (EU) 2019/2161 explicitly banned buying, selling and submitting false reviews and required traders to tell consumers how they handle reviews.

    ec.europa.eu

    Credit

    European Commission and the Consumer Protection Cooperation network of national authorities

  6. 2024

    US rule on fake reviews

    The Federal Trade Commission's rule on consumer reviews, 16 CFR Part 465, took effect on 21 October 2024. It bars businesses from writing, selling or knowingly buying reviews by people who do not exist or never used the product, which covers AI-written reviews, and from paying for reviews that express a set opinion; sites that merely host reviews are exempt. The FTC said the rule lets it seek civil penalties without first winning a separate administrative case. In December 2025 it sent ten warning letters, its first public use of the rule, and in May 2026 the Justice Department, acting for the FTC, sued a firm called Premium Home Service, seeking civil penalties under it.

    federalregister.govglobalpolicywatch.combeneschlaw.com

    Credit

    US Federal Trade Commission

Watch for

  1. Britain's first fake-review cases

    Since 6 April 2025 UK law has banned fake reviews outright, and the Competition and Markets Authority (CMA) can now decide a case itself and fine a business up to 10% of its global turnover. Amazon gave the CMA undertakings in June 2025 to detect fake reviews and "catalogue abuse," in which a seller moves the reviews of a well-rated product onto a different one. In July 2025 the CMA said 54 of more than 100 businesses it checked could be failing its guidance, and on 26 March 2026 it opened investigations into five, among them the food-delivery marketplace Just Eat and the review platform Feefo.

    Achieved whenthe CMA issues an infringement decision or a fine in one of these cases.

    gov.ukgov.uk

    Claimed by

    Competition and Markets Authority, March 2026

  2. AI summaries flatter products

    Two studies say the AI-written summaries that shopping and travel sites now show above reviews read more positive than the reviews themselves. Joanna Quan and Wenqi Shen of Virginia Tech compared the summaries for 905 hotels in 21 US cities with the 96,630 reviews posted from November 2024 to October 2025, and found sentiment raised by 0.23 on average on a scale from −1 to +1, and by up to 0.45 for poorly rated hotels. A UC San Diego team led by Abeer Alessa reported that language models changed the framing or sentiment of the source in 26% of summaries; of 70 participants, 84% of those who read summaries said they would buy the product, against 52% of those who read the original reviews.

    Achieved whenpeer-reviewed papers with their data appear, or another team finds the same gap in another site's summaries.

    search.ftc.govmediapost.com

    Claimed by

    Joanna Quan and Wenqi Shen, March 2026; Abeer Alessa and colleagues, February 2026

  3. Ratings that travel with workers

    GigCV, begun by the platform researcher Martijn Arets of the Amsterdam University of Applied Sciences, is a shared standard, a programming interface and a set of legal agreements under which platforms let workers download their reviews, skills and number of jobs done as a CV. Arets reported in September 2025 that, built with six platforms, it reached more than 100,000 workers in six countries in Europe and Africa, and that more than 30,000 CVs had been downloaded. An online experiment with 239 people by Diana Tran Nhat, Laura Thüter and Timm Teubner (2024) found that imported ratings raised demand for a worker about 35% as much as ratings earned on the site itself. From 2 December 2026 the EU's Platform Work Directive gives people who work through platforms the right to take ratings and reviews from a platform's automated systems to another platform in a machine-readable form (Article 9(6)).

    Achieved whenAn outside study reports how many workers used their GigCV with another platform or client, and whether it changed the work or pay they got.

    indl-8.sciencesconf.orggigcv.orglink.springer.compublications.europa.eu

    Claimed by

    Martijn Arets (GigCV), September 2025

Open

  1. 01

    Check review policies automatically

    An open tool that reads a shop's or marketplace's public pages and reports, with quotes, whether it tells buyers how reviews are collected and checked, how it prevents fake reviews, and whether it bans or labels reviews given in return for a reward. These are the points EU authorities checked in 2022 and the CMA checked in 2025. The tool reads policy pages only and collects no reviews and nothing about reviewers.

    The brief
    Credit · unclaimed
  2. 02

    A table of review laws

    A machine-readable dataset of fake-review rules in at least 25 countries and blocs: who is liable (the writer, the business that buys, the site that hosts), whether AI-written reviews are covered, whether paid reviews are banned or must be labeled, whether platforms may delete reviews, the penalty, the regulator and the date in force, with every cell linked to the text of the law.

    The brief
    Credit · unclaimed

Open 01 · the brief

Check review policies automatically

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An open tool that reads a shop's or marketplace's public pages and reports, with quotes, whether it tells buyers how reviews are collected and checked, how it prevents fake reviews, and whether it bans or labels reviews given in return for a reward. These are the points EU authorities checked in 2022 and the CMA checked in 2025. The tool reads policy pages only and collects no reviews and nothing about reviewers.

Why it is still open

Both checks were done by the authorities' own staff, on 223 sites and on more than 100, and both published only totals. A search in November 2026 found fake-review detectors and vendor tools, but no public tool that checks a site's review policies against these rules.

ec.europa.eugov.uk

Built with what exists

Browser Use 1.0 to open pages and follow links (Note: "Deleting 80% to start over"); Claude or Gemini to read each page and quote the evidence; Open Terms Archive for earlier versions of policies; Python.

  1. First step

    Turn the four points of the EU check and the CMA's guidance into a 10-item checklist and code 10 sites by hand.

  2. Done looks like

    Reports for 200 sites; on a sample of 40 sites coded by hand by two people, the tool agrees with them on at least 90% of checklist items; the code, the checklist and every report published.

Open 02 · the brief

A table of review laws

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A machine-readable dataset of fake-review rules in at least 25 countries and blocs: who is liable (the writer, the business that buys, the site that hosts), whether AI-written reviews are covered, whether paid reviews are banned or must be labeled, whether platforms may delete reviews, the penalty, the regulator and the date in force, with every cell linked to the text of the law.

Why it is still open

The rules differ in ways a seller cannot guess. The US rule exempts sites that merely host reviews, China's law bars platforms from deleting buyers' comments, and EU law requires traders to say how they check reviews. A search in November 2026 found law-firm guides written as prose, but no dataset.

federalregister.govsubsites.chinadaily.com.cn

Built with what exists

Official legal texts (the Federal Register, legislation.gov.uk, the EU and national gazettes); a language model to draft each cell from the text, checked by a person; CSV and JSON files on GitHub under an open license.

  1. First step

    Build the first five rows (the US, the UK, the EU, China and India) and publish the fields for comment.

  2. Done looks like

    25 jurisdictions, every cell quoting and linking its source and checked by a second reader, plus a published list of the points where the texts are unclear.