Sock Puppets and Review Fraud

Why the falsity argument loses and the coordination argument wins

Read enough platform policy and one absence stands out: almost nobody has a defamation rule. Meta's Community Standards run to roughly twenty-five sections and none is about false statements; Twitch has five categories and none asks whether something was true. A business arguing "this is false" is answering a question nobody asked.

What those operators do have is rules about conduct. Who may post. How many accounts one person may use. Whether a reviewer has a financial interest. Whether anyone was paid. Whether an account exists to evade a ban. Those rules get enforced daily, and they do not require the platform to decide who is telling the truth about a business dispute.

That is why this page carries a different verdict from the rest of this group: establishing that a set of reviews or accounts is coordinated is a policy argument, and platform rules provide for removal on policy grounds.

Meta shows the shape of the move most clearly. Its Account Integrity policy reaches accounts "Owned by the same person or entity as an account that has been disabled," accounts "Created or repurposed to evade a previous account or entity removal," and "coordination within a network of accounts … that persistently or egregiously violate our policies" (Meta Transparency Center, Account Integrity, change log 28 May 2026, read 12 August 2026). One at a time, a determined attacker's posts may be protected opinion. Together, the accounts may be ban-evading. Reframing a matter from content to conduct is frequently the only route that goes anywhere.

Amazon's spending gate, and what it tells you about detection

Amazon operates the strongest structural anti-fraud rule in the review world, and most people arguing about fake Amazon reviews have never read it. It is not a detection system; it is a gate on who may post at all:

"You must have spent $50 on Amazon.com, using a credit or debit card, in the past 12 months, to: Create reviews (including star ratings), Answer customer questions, Submit helpful votes … (Note: Promotional discounts don't count towards the $50 minimum spending requirement.)"

— Amazon Community Guidelines, read 12 August 2026

The exclusion at the end is the interesting half: discounted spend does not count, which closes the obvious workaround of seeding accounts with heavily discounted purchases to unlock reviewing. To post at all, an account has to represent real card spend within the year.

For anyone assessing a suspicious wave of reviews, that is a concrete fact rather than a feeling. Every reviewing account cleared a spending threshold — so a ring is either running accounts with genuine purchase history or it broke a published rule to exist. Both are conduct, both are reportable in the operator's own language, and neither requires arguing about whether a review's contents were accurate.

The conduct Amazon names by name

Amazon's guidelines go further than most, and include one of the few operator uses of the word itself: under profanity and harassment, the prohibited list includes "Libel, defamation, or inflammatory content." More useful against a campaign are the rules aimed at rings:

• Multiple accounts and coordination. The guidelines prohibit "Drowning out opinions. Don't post from multiple accounts or coordinate with others" — the sock-puppet rule in a sentence.
• Competitors are barred outright. Amazon does not allow content about products or services offered by friends, relatives, employers, business associates or competitors, and removes reviews "posted by someone with financial interest in the product, or any other conflict of interest."
• Compensation, in either direction. Nobody may create, edit or remove a review in exchange for payment, refunds, discounts, products, gift cards or services — and the rule closes the loop a seller most often reaches for: "This policy applies even if a seller offers a refund to remove or change a negative review."
• Wrong channel. Reviews focused only on the seller, ordering, returns, shipping or packaging belong in seller feedback rather than product reviews — removable on their face.

The consequences make a report worth filing: removing content, limiting an account's use of community features, suspending or terminating accounts, and withholding payments. Amazon also reserves the right to act where someone violates state and federal law, "including the Federal Trade Commission Act."

What Yelp publishes over a business's own page

Yelp's response to coordinated activity is unusual because it is public. A Consumer Alert is a warning placed over the reviews on a business's page, and Yelp says that before placing one it investigates and will "provide the evidence we find whenever possible." Two of the six alert types describe it. A Compensated Activity Alert follows "evidence that someone has offered cash or other incentives in exchange for a review." A Suspicious Review Activity Alert is more specific:

"large numbers of reviews coming from a single IP address, or reviews from users who may be connected to a group that coordinates incentivized fake reviews"

— Yelp Trust & Safety, Consumer Alerts, read 12 August 2026

Note which way those alerts point: they land on the page of the business associated with the activity, so the mechanism reaches a competitor who bought reviews and equally a business that bought its own.

The quieter system underneath is the recommendation software. Yelp's stated reasons a review may not be recommended include reviews "originating from the same IP address" and reviews suggesting "an unfair bias (like the ones written by a friend of the business owner)" — the two signatures of an amateur ring, applied without anyone asking. Two things follow. The software is "completely automated, no Yelp employee can manually override the software," so nobody can filter reviews as a favor or a fee. And not-recommended is not removal: the review still exists, one click away.

The 2026 rule that changed what a fake review is

Yelp's content guidelines now carry a provision that did not exist when most published advice on fake reviews was written, and it is broader than its headline:

"you shouldn't use third-party AI tools and chatbots to create reviews or other content, including using such tools to draft or revise content"

— Yelp Content Guidelines, yelp.com/guidelines, read 12 August 2026

The words "or revise" do a great deal of work. A review composed by a person and cleaned up by a chatbot is inside the rule, and so is a business response written the same way — a policy hook where a wave of reviews reads like generated text, and a risk if your own marketing drafts replies with the same tools.

It also changes the value of an old signal. "These reviews are suspiciously similar" worked when fraud meant one person copying and pasting. Generated text produces varied prose on demand, which has largely destroyed template-matching as evidence — while creating a fresh rule violation in its place.

The rule that cuts against your own review drive

Every operator that bans incentivized reviews bans them in both directions, and this is where a business trying to fix a rating becomes the party in breach. Glassdoor removes content where there is "evidence that users were incentivized to or coerced into leaving the content," applies that to positive reviews where employees were compensated or coerced, and does not allow employers to ask workers to show confirmation of posting (Glassdoor Community Guidelines, updated 15 April 2025, read 12 August 2026).

Yelp goes further and says businesses should never ask customers to write reviews at all. Amazon's rule covers a refund offered to remove or change a negative review. Read together, the standard advice a business gets after a bad review — run a review drive, offer the unhappy customer something, ask staff to balance it out — describes conduct three major operators publish rules against. If you intend to argue that a set of reviews is coordinated and incentivized, your own solicitation history is part of the same picture. I would rather say that before a report is filed than after a platform notices.

What a pattern proves, and what it does not

The observable signals are worth documenting: accounts created days before posting, a step change in volume against the business's own baseline, a sudden split of one-star and five-star ratings with nothing between, reviewers whose other activity is all in one distant city, reciprocal reviewing across competitors, and the same complaint appearing on four platforms in a week in the same words. That material is the substance of a platform report.

What it is not is an identification, and this is where the market oversells.

• There is no denominator. Nobody publishes the rate at which genuine reviews look like this, and without a base rate "this looks like fraud" is an impression with a chart attached.
• The strongest pattern evidence shows a campaign, not an author. Establishing that six reviews came from one hand is a different and easier problem than establishing whose hand it was.
• Platform detection is proprietary and does not explain itself. Device fingerprints and behavioral telemetry are signals no outside analyst has. A decision to remove a review is not a finding that a named person wrote it, and a decision to keep one is not a finding that it is genuine.
• Your own evidence is discounted for being yours. Glassdoor states it plainly: "We generally do not consider evidence offered by someone with a vested interest in removing a review, as its reliability cannot be verified."

The enforcement record makes the same point from the other end. The Federal Trade Commission's canonical sock-puppet matter, its 2019 complaint against a skincare company, rested on internal instructions quoted in the complaint — "Create a new persona. Choose their name, city, skin type" and "Connect to the internet ONLY using the VPN." The company was defeating the very signals an outside analyst would rely on. These cases are made with internal communications, subpoenaed records and testimony, not by reading reviews.

The version of this that gets the business penalized

There is an aggressive version of this, and three separate systems now penalize it.

Yelp maintains a Questionable Legal Threats Alert, placed on a business's page where there is "evidence that a business may be abusing the legal system to intimidate or silence a reviewer." Glassdoor says that where it judges a suit is primarily intended to suppress speech, it may publicly highlight the situation. And since 21 October 2024, the FTC's Trade Regulation Rule on the Use of Consumer Reviews and Testimonials, 16 CFR part 465, has made it an unfair or deceptive act to use "an unfounded or groundless legal threat, a physical threat, intimidation" to stop a review being written or to force one down — 16 CFR § 465.7. Announcing warning letters to ten companies in December 2025, FTC staff wrote that such companies "may be subject to FTC enforcement actions and civil penalties," of up to $53,088 per violation.

Two further parts matter. § 465.4 covers buying negative reviews, so paying for an attack on a competitor sits inside it. And the FTC's guidance states that ordinary consumers cannot be liable under the rule for what they say in reviews, while businesses, agencies, review brokers and reputation management companies can be.

The limits, plainly. Reporting a ring is a request under a platform's rules, decided by that platform, on no published timetable and with no obligation to explain the outcome. The FTC does not adjudicate individual disputes: it will not remove a review or identify a poster, and a report goes into a database that informs enforcement at the Commission's discretion. Filing is worth doing and is not a remedy. Accusing a named competitor of running a ring is its own exposure — the pattern that convinced you is not proof. Policies change, too: everything quoted here was read on 12 August 2026.

Frequently Asked Questions

Published 2026-08-12