In August 2024, the Federal Trade Commission published a rule that read, to anyone in e-commerce, like a declaration of war on fake reviews. The final Rule on the Use of Consumer Reviews and Testimonials banned inventing reviewers, buying stars and paying for sentiment, and it attached a price tag that made general counsel sit up: civil penalties for every knowing violation, adjusted annually, standing at $53,088 per violation in 2026. Every operator who read the announcement assumed the same target: marketplace sellers, review brokers, the five-star economy of the Amazon long tail.

Two enforcement cycles later, the ledger looks different, and it matters more now than it did in 2024. In December 2025 the FTC sent its first warning letters under the rule, to ten companies whose names the agency declined to release. A FOIA request pried them loose, and when the FTC posted the letters in April 2026 the list read: six property management companies, three personal injury law firms, and one accounting firm. In May 2026 the agency filed its first lawsuit under the rule, against a home-services company, jointly with the Illinois attorney general. As the law firm Benesch summarized the agency’s posture in April, the FTC is now prioritizing enforcement under the Consumer Review Rule. But the center of the fake review economy, the marketplace inventory where hundreds of millions of suspect reviews live, has been left to the platforms to police.

That division of labor might have been survivable when humans read reviews one listing at a time. It is not survivable in the holiday season that began on November 1. Adobe projects AI-referred traffic to US retail sites up 130 percent year over year through December 31, and a growing share of what those AI visitors read, and relay to shoppers, is review data of unverified provenance. A rule enforced after the fact, against the edges of the economy, cannot clean a corpus that machines now summarize before anyone can audit it.

What the rule actually bans

The rule, codified at 16 CFR Part 465 and effective October 21, 2024, is not vague. Drawing on the enforcement record as documented by Retail Reasoning, its core prohibitions are:

  • Fake or false reviews. Reviews from people who do not exist, including AI-generated ones, or people who never used the product. Creating them, selling them, buying them, or publishing them when you knew or should have known.
  • Incentives tied to sentiment. Rewarding anyone for a review expressing a particular opinion, positive or negative. The condition can be implied; it does not have to be in writing.
  • Insider reviews without disclosure. Officers, managers, employees and agents reviewing their own company’s products, including the relative-recruitment variant where staff ask family to post.
  • Company-controlled review sites. Running an “independent” comparison site that conveniently ranks your own products first.
  • Suppression. Using groundless legal threats to bury negative reviews.

Each knowing violation carries the per-violation penalty. The conduct in the December 2025 warning letters was almost identical across all ten recipients, according to the FOIA-released records: lease discounts, gift cards and other incentives handed to tenants, clients and employees in exchange for positive reviews, or for pushing friends and family to post them.

The enforcement ledger, read honestly

Here is the uncomfortable summary of two years of activity under the most explicit anti-fake-review law any major jurisdiction has passed:

  • December 2025: ten warning letters. Six property managers, three personal injury law firms, one accounting firm. Not a marketplace seller, not a consumer brand, not a single ASIN.
  • May 2026: first lawsuit under the rule, a home-services company, filed with the Illinois attorney general rather than alone.
  • Through 2026: DLA Piper’s review of the agency’s approach notes that Consumer Review Rule violations tend to surface as adjuncts to other FTC cases rather than as standalone marketplace sweeps.

There is a defensible logic here. Small-business enforcement builds precedent cheaply: local incentive schemes leave paper trails, tenants talk, and state attorneys general bring parallel authority. The FTC’s own Q&A acknowledges the rule is enforced through cases, and cases are expensive and slow. A regulator with finite litigation capacity rationally starts where provability is highest, not where volume is highest.

But note what this means for the shopper. “The rule everybody read as an Amazon rule got used on landlords,” as Retail Reasoning put it. The review corpus where Americans actually make purchase decisions, the marketplace listing page, remains policed almost entirely by the marketplace itself.

The deterrence math does not close

Set the penalty against the volume and the gap becomes arithmetic. Amazon reports having blocked more than 250 million suspected fake reviews in 2023 and more than 275 million in 2024. That is roughly 750,000 blocked suspect reviews per day, on one platform, before counting what slips through.

The FTC’s Consumer Review Rule docket, in its first two years, amounts to ten warning letters and one filed complaint. Even if the agency filed a new case every single day, it would touch a few hundred defendants a year against a phenomenon measured in hundreds of millions of reviews. The penalty is real, and the $53,088 figure has genuinely changed compliance conversations at mid-size firms. But deterrence scales with probability of enforcement, not severity on paper, and the probability that any given fake review produces an FTC action is effectively zero. Felukaa’s compliance analysis gets the mechanism exactly right: “The fix isn’t a legal memo. It’s whether a review request can ever be traced back to a real, completed sale.” Traceability is a systems property, not a litigation property.

Amazon is the de facto regulator, and that is the problem

With the public regulator working the edges, the platform has become the operating regulator of review integrity, and to its credit it has escalated. The public record through 2025 and 2026 includes:

This is genuine enforcement, and it is also a structural anomaly. As Lawsuit Watch observes, most fake-review litigation in the United States is brought by the platform, not by the state. The company that profits from engagement on its listings is also the company that decides what engagement is fraudulent, with detection models it does not publish and thresholds it sets unilaterally. Sometimes the incentives align with shoppers. Sponsored placements and the sheer volume of third-party inventory give the platform its own reasons for tolerance at the margin. A shopper cannot audit any of it, and neither, critically, can an AI agent.

The variable nobody priced in 2024: the reader became a machine

The Consumer Review Rule was written for a world where a human scans a listing page and weighs 4.6 stars against a suspicious wall of superlatives. That world is ending mid-enforcement.

This holiday season is the first with mass-market agent surfaces live, and the numbers describe a rapid handoff:

  • Adobe’s forecast, built on more than one trillion visits and 100 million SKUs, projects AI-referred US retail traffic up 130 percent year over year, peaking at 159 percent growth on Thanksgiving.
  • A Forter/Harris survey of 2,017 US adults fielded September 15 to 16, 2026 finds 67 percent of Americans plan to engage in AI holiday shopping, 78 percent of holiday shoppers are comfortable letting AI handle at least one part of the journey, 47 percent are comfortable letting AI find deals and 44 percent letting it run product search or comparison. Only 20 percent are comfortable letting AI select the product itself, and the top comfort conditions are human final approval (26 percent), spending limits (24 percent) and guaranteed refunds (22 percent).
  • FirstPageSage’s October dataset puts 61.4 percent of shoppers researching in AI, 39.2 percent reading agent-written review summaries, and 6.8 percent approving an agent checkout. The handoff is happening at the research layer, exactly where reviews live.
  • Actual end-to-end agent purchasing remains the exception: an April 2026 L.E.K. survey of 2,650 US consumers found 8 percent of AI users had completed a purchase entirely through an agent.

Read those two datasets together and the risk surface is precise. Shoppers delegate reading, not deciding: 39 percent of them now consume reviews pre-digested by a model, while checkout approval stays with the human. That means the fake review’s job description has changed. It no longer needs to fool a skimming shopper. It needs only to survive ingestion by a summarizer that treats every review in the corpus as equally legitimate input. A hundred coordinated five-star posts, written by people who never opened the box, become one fluent sentence: “Reviewers consistently praise the build quality.” The laundering happens in the summarization step, and no subpoena runs at the speed of inference.

The FTC’s penalty attaches after the fake review is created, sold, bought or published, and after an investigation that takes months. The agent’s summary is generated in under a second, this afternoon, for a shopper holding a phone in a store aisle. These two clocks do not meet.

Verification has to move upstream of the summary

If enforcement is post hoc and summarization is real time, then the only place the fake review problem can actually be solved is before the corpus reaches the model. The question that matters is not “can this review be proven fake in court” but “has this review been traced to a real, completed sale, and does the corpus around this product deserve its rating.” That is a filtering and scoring problem, and it is solvable today.

This is the layer GoBuy builds. Instead of counting stars, the Smart Score grades products 0 to 100 on the quality and authenticity of the review evidence: fake and low-veracity reviews are filtered out before they count, authentic ones are weighted up, and the shopper sees a ranked shortlist of seven products rather than a thousand listings sorted by whatever the platform’s incentives favor that day. Products that hold a score above 80 across 90 days earn a GoBuy Verified badge, which is evidence accumulated over time rather than a snapshot that can be gamed in a weekend. For humans, the Chrome extension injects this trust panel directly onto the Amazon page. For agents, the same signals are exposed through the MCP server at gobuy.ai/api/mcp, so a shopping agent can consult verified review-quality data before it writes a single sentence of summary. Agent developers can integrate it in an afternoon via the docs at gobuy.ai/agent-docs.

The FTC deserves credit for what it did build: a rule that makes every practice in the fake review supply chain specifically illegal, with a penalty no small business wants to explain to its insurer. Two years of enforcement history simply show what a courtroom can and cannot do. It can punish the landlord who traded gift cards for stars. It cannot audit 750,000 reviews a day, and it was never going to.

The audit has to happen in the decision path, at machine speed, before the summary. The platforms will not publish their detection models, the regulator will never litigate at corpus scale, and this holiday season the first reader of most product reviews will be a language model. The trust layer that decides what those models are allowed to trust is the missing infrastructure of agentic commerce. Try it at gobuy.ai before your next purchase, and if you build agents, wire them to the trust data first.