Black Friday 2026 starts in one week, and for the first time in the event’s six-decade history, the deal itself is under surveillance. Two releases in the past five days, one from a watchdog and one from a logistics giant, describe the same market from opposite ends. Groundwork Collaborative published Deal or Decoy? Amazon’s Deceptive Discounts, an audit of more than 350 products Amazon itself promoted as the best deals of its recent Black Friday and Cyber Monday events, and found that nearly one in three of them saved the shopper nothing at all. DHL eCommerce published its 2026 E-Commerce Trends Report, a survey of 29,000 shoppers and 5,800 businesses across 29 countries, and found that nearly nine in ten Black Friday shoppers have used or plan to use AI to compare prices.
Put those two facts in the same sentence and you have the theme of this holiday season: the fake discount is dying at the exact moment the machines capable of killing it reached scale. What has not been built, and what this piece argues is now the binding constraint on AI-assisted commerce, is the equivalent audit for the product itself. The price history of that robot vacuum is becoming a queryable fact. Whether its 4.8-star average reflects 4.8 stars of human experience is not. Black Friday’s trust problem did not disappear this year. It moved.
The decoys, quantified
Groundwork Collaborative’s methodology is refreshingly direct. The researchers took the products Amazon itself curated and advertised as top deals during its Black Friday Week and Cyber Monday events, then reconstructed each product’s price history over the 30 and 90 days before the sale, and asked the only question that matters: could you have paid less earlier?
The findings, in the report’s own numbers:
- 29.1 percent of the promoted deals were priced the same or higher during the sale event than in the month leading up to it. Extend the lookback to 90 days and the failure rate climbs to nearly 60 percent.
- 9.4 percent of “deals” were actually more expensive during the sale than in the prior month. Against the 90-day window, that share nearly triples to 27 percent.
- Half of the no-savings items were available at or below the “deal” price on at least eight of the 30 days before the sale. The discount badge was certifying a price that had been ordinary all month.
This is not a story about random price wobble. These were the deals Amazon chose to feature, the ones with the struck-through reference price, the red percentage badge, and the “Black Friday Deal” label. The report quotes Doug Herrington, CEO of Worldwide Amazon Stores, announcing “incredible Black Friday Week and Cyber Monday deals across a vast selection of products with deep discounts across more than 35 categories and some of [Amazon’s] lowest prices of the year.”
The commercial stakes are enormous, which is why the deception, intentional or negligent, is structurally motivated. Amazon runs nearly 40 percent of US e-commerce, concentrates roughly a third of its revenue in the fourth quarter, and has kept stretching the event: Black Friday Week ran 12 days in 2025, and the expansion of Prime Day to four days that year drove $24 billion in total US online spending. Major sale events drove nearly $70 billion in online spending in 2025. Consumer appetite keeps rising: 70 percent of US customers plan to shop the Black Friday and Cyber Monday window this year, up from 64 percent in 2025, and many delay purchases for weeks specifically to catch the deals. A shopper who waits five weeks to buy a blender and pays the November 5 price for the privilege has been harmed in the most literal way the word supports.
The auditors arrived
Here is where the DHL data transforms this from a consumer-protection story into a market-structure story. The 2026 E-Commerce Trends Report found:
- Nearly nine in ten surveyed Black Friday shoppers have used AI to compare prices or plan to do so. The bargain hunt, the historical core ritual of Black Friday, is now majority-delegated to machines.
- Nearly half have already used AI chat or virtual assistants to browse or buy.
- AI-assisted shoppers close: 50 percent of them purchased during Black Friday, versus 42 percent of traditional deal seekers, and they are more likely to increase their buying in future years.
- More than a third of e-commerce businesses already let shoppers browse or buy through AI-powered chat or virtual assistants.
Price comparison is the oldest form of shopping software, but the DHL numbers show something qualitatively new: it is no longer a niche of spreadsheet people and browser extensions. It is the default posture of the Black Friday shopper, and it is bundled into conversational surfaces where asking “was this actually cheaper last month?” costs one sentence.
Vendor research shows the same shift from the supply side. AI shopping agent Deal.ai, in a November 17 study covering 24,394 identical products across 16 major US retailers, found that the price at one randomly chosen store averaged 15.6 percent above the lowest listed price, and that checking just three stores cut the gap to 3.7 percent. About one in three products at Walmart and Amazon, and half of those at Target, were listed for less at another major retailer at the same moment. The company has frozen a basket of 972 popular products and is repricing it daily through Cyber Monday, and its early data shows how fast the “deal” evaporates: a Shark Stratos vacuum that sat at $264.99, half price, at Target, Best Buy, and Kohl’s three weeks before Black Friday was back to $529.99 at all three within three days. “The exact same product can cost hundreds of dollars more at one store than another, at the same moment,” said Alex Mehr, CEO of Famous Labs, the company behind Deal.ai. (The study is vendor-published and Deal.ai earns commissions on referred sales, so read it with that lens, but the price gaps it measures are mechanically verifiable by anyone with a browser.)
Groundwork’s audit ran on hindsight. Deal.ai’s tracker runs in real time. Both now exist in a market where, per DHL, roughly nine in ten Black Friday shoppers have the tooling to ask.
What happens to a fake discount when the reference price is queryable
Reference-price manipulation works by asymmetry of memory. The struck-through “list price” or “typical price” only functions as an anchor if the shopper cannot cheaply recall what the item cost on November 3. For sixty years, that asymmetry held because human memory is short and retailer price histories were private. In 2026 the asymmetry inverted: the machine remembers everything, the price history is a lookup, and the shopper delegates recall by default.
That inversion has three consequences worth spelling out.
First, fake urgency decays into a computed fact. A “46 percent off” badge on an item that sat at the sale price for eight of the last thirty days is not a discount, it is a fact-check waiting to happen, and the fact-check now happens before checkout, in the chat window where the purchase conversation is already occurring. Regulators have started moving on the practice in parallel: European authorities have been targeting fake urgency claims this season under existing unfair-commercial-practices law, and the FTC’s own consumer advice treats inflated reference prices as a recurring enforcement theme.
Second, the FTC’s holiday guidance has quietly become an AI prompt. The Commission’s November 19 consumer alert on holiday search scams tells shoppers to “scroll past the top results,” because “the organic results below the paid ads are ranked by relevance, not by who paid the most,” and to “check out the seller” by searching the company name “plus the words ‘complaint’ or ‘scam.’” Every one of those steps is a lookup a shopping agent can execute natively. The advice only ever failed as consumer education because humans do not do diligence at 2 a.m. over a $34 gadget. Agents do it every time, at zero marginal cost. The FTC wrote the algorithm by writing the advice.
Third, pricing discipline gets enforced in milliseconds rather than fiscal years. A deceptive-practices lawsuit takes years; a price-tracker comparison takes a second, and nine in ten shoppers now carry one. Retailers will still run reference-price theater as long as it works, but its working lifespan is collapsing. The rational response is not more honest badges, it is pricing that survives the lookup, which is a materially different discipline than pricing that survives a glance.
The audit stops at the price tag
Now run the same audit logic up the funnel and watch it fail. The machine can verify that the vacuum’s price history justifies the badge. Can it verify that the vacuum deserves its rating?
Not with what it is given. FirstPageSage’s November 2026 study of 5,212 US shoppers found review summaries are now the third most common AI shopping behavior in America at 40.1 percent of online shoppers, behind only research and price comparison. That means the single most consulted trust signal in AI-assisted shopping is a machine-generated summary of a review corpus that is itself under industrial-scale attack. Amazon’s own reporting says it proactively blocked hundreds of millions of suspected fake reviews in 2025, on top of more than 275 million blocked in 2024, and the FTC’s Consumer Review Rule now carries civil penalties of $53,088 per fake review, numbers that describe a corpus no honest summarizer should treat as ground truth.
The gap becomes stark when you put the two audits side by side. Question one: was this product cheaper last month? Answerable, increasingly answered, and monetized by a growing field of price agents. Question two: were this product’s reviews written by people who owned and used it, and does the rating reflect their experience rather than a farm’s output schedule? Unanswerable from the listing itself, because nothing in the review payload distinguishes a purchased five-star from an earned one. The price check is a lookup; the trust check is an investigation, and almost no consumer-facing system performs it.
DHL’s own CEO put his finger exactly on this seam. Pablo Ciano, CEO of DHL eCommerce: “AI can find the bargain, compare the options and increasingly influence the buying decision, but it can’t replace the trust that wins the customer. As shopping becomes more AI-powered, that human connection becomes more, not less, important.” DHL sells delivery trust, so the sentence continues into logistics, but the first clause is a category-wide confession: the bargain-finding is solved, the trust is not. DHL’s survey also found 77 percent of shoppers would abandon a purchase if they did not trust the counterparty behind it. Price verification without product verification optimizes how much you pay for a thing you cannot yet trust.
Building the missing half of the audit
If 2026 made price machine-verifiable, 2027 is when product evidence gets the same treatment, or it should be. The requirements are already visible in what failed this season:
- Filter before you summarize. Any trust signal computed on the raw review corpus inherits the farm’s work product. Fake reviews need to be detected and removed, and authentic ones weighted up, before any score or summary is generated.
- Score quality, not volume. A 4.8 built on thousands of purchased reviews and a 4.3 built on verified, detailed, durable reviews are not nearly the same product, and a single 0-to-100 score computed from review quality rather than review count is the only honest way to collapse that distinction for a human or an agent.
- Short lists beat shelves. An audited recommendation is defensible only if every slot in it is defensible. Seven verified survivors, not thousands of ranked listings.
- Verification needs a track record. A score that can be gamed the week before Black Friday is not a trust signal, it is a campaign. A badge that requires a sustained 80-plus over 90 days cannot be sprinted into existence for the sale.
- Agents need a place to call. Price agents have endpoints; trust agents need one too.
That is the specification GoBuy was built against, and it is why the two halves of this season’s story both point the same way. GoBuy filters fake reviews out before scoring and weights authentic ones up, computes Smart Score from 0 to 100 on review quality rather than quantity, surfaces only the top 7 products per category, and awards GoBuy Verified to products holding 80-plus over 90 days. The MCP server at gobuy.ai/api/mcp exists so that the same shopping agents now auditing prices across 16 retailers can audit the product before checkout, and the Chrome extension puts the trust panel directly on the Amazon page for the shopper doing the last check themselves.
Groundwork showed the badges lie about a third of the time. DHL showed the machines arrived to check them. The part nobody shipped yet is the product audit, and until it exists, every “best deal” recommendation this holiday season is a price-verified box around an unverified thing. Verify the deal, then verify the product. Start at gobuy.ai, and if you build agents, wire trust into the pipeline at gobuy.ai/agent-docs.