Astroturfing

What it looks like

You are not being argued with. You are being shown a majority that does not exist, on the assumption that you will adjust toward it without noticing you have.

The commercial version is fake reviews. The organised version is many accounts posting near-identical wording in a narrow window. Both work on the same reflex: other people already decided, so this is probably fine.

Why it works

The most striking evidence is how little manipulation it takes. In a randomised experiment on a live news site, 101,281 comments were randomly given a single artificial up-vote, a single down-vote, or nothing. One fake up-vote raised the probability that the next real viewer would also up-vote by 32 percent, and raised the comment’s final average rating by 25 percent.1 Artificial negative votes, by contrast, were corrected by later readers. The bias runs one way.

The observable signature is coordination in time, not volume. Researchers using a platform’s own archive of disclosed campaigns as ground truth built detection from accounts posting and reposting within one-minute windows, and identified on average 74 percent of coordinated accounts at roughly a 1 percent false-positive rate — because ordinary users essentially never behave that way.2 Volume proves nothing. Simultaneity is the tell.

The commercial scale is large. An analysis of 73 million reviews across 127 business categories in 100 US cities classified 13.7 percent as inauthentic or highly suspicious.3 That study was published by a company selling fake-review detection, so we cite its sample size and its percentage and not its estimate of economic harm.

Documented cases

Several thousand reviews written in-house, 2026

In April 2026 the US Federal Trade Commission announced action against a supplement company over reviews on its own site. The complaint alleged that employees and vendors wrote several thousand five-star reviews, that consumers were given free or discounted product in exchange for five-star ratings, and that automated accounts impersonating real people posted comments on social platforms.5 A four million dollar judgment was entered, suspended on payment of 750,000 dollars.

Nothing here is exotic. It is a business buying the appearance of a satisfied customer base, and it is illegal for that reason.

Fake businesses with real-looking reviews, 2026

In May 2026 the FTC and the State of Illinois brought an action alleging the creation of thousands of listings for home-repair businesses that did not exist, together with fabricated five-star reviews used specifically to dilute genuine one-star reviews and raise the aggregate rating.6

These are allegations in pending litigation, not findings, and we say so because that is the standard this site holds other people to. The mechanism is worth knowing regardless: the target was not the individual review but the average.

What to do about it

• Read the clock, not the count. Many accounts posting near-identical wording inside a few minutes is the signature. Volume on its own means nothing.
• Open three of the accounts saying the same thing. Check creation dates, posting history, and whether they exist outside this one topic. Real accounts have unrelated pasts.
• On reviews, look at the shape rather than the score. Clusters dated within days of each other, generic phrasing, reviewers with exactly one review ever, and a distribution that is all fives and ones with nothing in between.
• Search a distinctive sentence in quotation marks. Copy-paste is the cheapest part of the operation and the easiest to catch.
• Form a view before you look at the score. A single fabricated signal moved the next person by about a third in a controlled experiment. Assume it has already worked on you.

Sources

Every figure on this page comes from one of these. Each entry names the publisher, the date and the sample size, so you can check it rather than take our word for it.

1. Muchnik, L., Aral, S. & Taylor, S. J. — "Social Influence Bias: A Randomized Experiment"Science 341(6146):647–651 · 9 August 2013 · n=101,281 commentshttps://doi.org/10.1126/science.1240466
2. Schoch, D., Keller, F. B., Stier, S. & Yang, J. — "Coordination patterns reveal online political astroturfing across the world"Scientific Reports 12:4572 · 17 March 2022 · 46 disclosed campaignshttps://doi.org/10.1038/s41598-022-08404-9
3. The Transparency Company — "The High Cost of Review Fraud"Published 5 December 2024 · 73 million reviews, 62 billion impressions · vendor-published researchhttps://askfortransparency.com/research/high-cost-of-review-fraud/
4. Pacheco, D. et al. — "Uncovering Coordinated Networks on Social Media: Methods and Case Studies"ICWSM 2021 · arXiv:2001.05658https://arxiv.org/abs/2001.05658
5. Federal Trade Commission — action over fabricated reviews and bot commentsFTC · announced 13 April 2026, final order 22 July 2026 · $4M judgment suspended on payment of $750,000https://www.ftc.gov/news-events/news/press-releases/2026/04/ftc-takes-action-against-truheight-deceptive-unsubstantiated-advertising-supposed-height-enhancing
6. Federal Trade Commission and the State of Illinois — action over fake business listings and reviewsFTC · 11 May 2026 · allegations in pending litigation, not findingshttps://www.ftc.gov/news-events/news/press-releases/2026/05/ftc-illinois-take-action-stop-deceptive-conduct-company-created-thousands-business-listings-fake
7. Google Threat Analysis Group — TAG Bulletin Q4 2025Google · published 29 January 2026 · ~18,200 YouTube channels terminated in the quarterhttps://blog.google/threat-analysis-group/tag-bulletin-q4-2025/
8. Meta — Adversarial Threat Report, Q2 and Q3 2025Meta Transparency Center · December 2025 · itemised network removals with asset countshttps://transparency.meta.com/sr/Q2-Q3-2025-Adversarial-threat-report/