← October 2026
App Idea Cards 2026-10-20

TurfTrace

Reruns a product's star rating after pulling out the reviews that read like a bot or a paid account wrote them, and shows you which ones it threw out.

TurfTrace

TurfTrace

Reruns a product's star rating after pulling out the reviews that read like a bot or a paid account wrote them, and shows you which ones it threw out.

Problem

You're comparing baby monitors at midnight. One has 4.6 stars across 3,000 reviews, most tagged Verified Purchase. That badge is supposed to mean something, but a May 2026 study by Pangram Labs found that 93% of the AI-generated reviews it detected on front-page Amazon listings carried the Verified Purchase badge anyway, and the AI-written share of reviews climbs to around 5% in categories like baby gear, supplements, and beauty — exactly where a wrong purchase costs the most. Amazon says it blocks hundreds of millions of fake reviews a year, but that happens upstream, before you ever see the listing. Nothing tells you, on the page, in the ten seconds before you click buy, what got through.

Target user

Shoppers making a purchase they can't easily unwind — supplements, baby products, skincare, anything health-adjacent or hard to return — who read reviews carefully but have no way to tell a planted one from a real one. Also small resellers vetting suppliers before they commit to a product line, where a wrong call costs inventory, not just a bad week.

MVP scope

  • Manifest V3 extension that activates on amazon.com product pages
  • Reads the review text and metadata already rendered on the page — no login, no API key, no server round-trip for the base tier
  • Flags individual reviews a lightweight AI-text classifier scores as likely machine-written, with a one-line reason next to each
  • Flags review bursts: clusters of reviews posted the same week from accounts created in the same window, the pattern the FTC cited as evidence in its 2026 TruHeight and Premium Home Service cases
  • Shows a recalculated star rating next to Amazon's native one, with flagged reviews excluded
  • Caches the score per product for seven days so a repeat visit doesn't re-score from scratch

Monetization

Freemium. Free tier: the adjusted rating and a flag count. Pro at $4.99/mo: the full per-review breakdown, support for Walmart and Sephora listings, and a watch-list alert if a tracked product's review mix shifts sharply overnight — the kind of spike a paid campaign tends to leave behind.

Why now

The FTC's Consumer Review Rule moved from warning letters in late 2025 to actual complaints in 2026, including a $4 million judgment against TruHeight in July and a pending case against Premium Home Service for the same incentivized-review pattern. Pangram's study gives the first hard number on how much of this is AI-written rather than just human astroturfing, and it's the Verified Purchase badge specifically that's compromised. Mozilla's Fakespot, the closest thing to this tool, shut down in July 2025 after Mozilla said it couldn't make the product sustainable as a standalone service — which argues for building this as a lean extension with a direct subscription, not a venture-funded platform play. Chrome's own Shopping extension category is one of the few where install growth is currently outpacing new entrants (+37.4% installs vs. +26.3% new extensions over 90 days as of May 2026), which suggests the audience is actively looking.

Risks & open questions

  • AI-text classifiers misfire on terse writers and non-native English speakers; flagging a genuine review as fake erodes trust in the tool faster than letting a fake one through does.
  • Amazon's review DOM changes often enough that scraping the rendered page is an ongoing maintenance cost, not a one-time build.
  • Will anyone pay for this? Amazon already markets itself as handling fake reviews, and Fakespot's shutdown suggests the standalone-checker model struggles for retention even without much competition.
  • Amazon's terms restrict automated handling of its pages; even read-only, client-side scraping inside a user's own session could draw a takedown notice against the extension listing.
  • The burst-detection signal (same-week reviews, same-age accounts) is inferred from public FTC complaints, not from testing against a labeled dataset — real accuracy is unverified until built.

Next step

Score 50 real Amazon listings against Pangram's published methodology and check agreement rate before committing to the extension UI.

Sources

More from App Idea Cards