She didn't know it was a bomb.

Sarah had been selling her turmeric supplement on Amazon for 14 months. Reviews were stacking. Ad spend was dialed. She was doing $38K a month and climbing.

Then on a Tuesday morning, her listing vanished.

No email. No warning. Just ... gone.

She spent three days in Seller Support hell before she got a one-line response: "Your listing contains a claim that is not substantiated."

The word? "Relief."

One word. Buried in her third bullet point. "Provides natural joint relief." She'd written it herself. It sounded like good marketing. It was also, according to Amazon's automated compliance filters, a medical claim.

Sarah isn't alone. This is happening to thousands of sellers right now, and the words getting flagged aren't the ones you'd expect.

"Eco-friendly." Flagged. Unless you have a government-recognized certification to back it up.

"Guaranteed." Flagged. Warranties have to be filed with Amazon for approval before you can even hint at one.

"Anti-bacterial." Flagged. Falls under both claim-related and pesticide-related restricted terms.

"Detox." Flagged. Skincare, supplements, general listings... doesn't matter. Amazon's filters treat it the same.

"Best seller." Flagged. It's a subjective claim AND a promotional term. Double violation.

Even "free" gets pulled. So does "gift." So does "new." So does "today."

Most sellers find out the hard way. A suppression notice drops with no explanation. Revenue stops. The appeal process is a black hole. And the whole time, the listing that took you months to optimize is sitting in the dark.

Here's what makes it worse: Amazon's compliance enforcement isn't just targeting the obvious stuff like drug claims or illegal ingredients. It's sweeping ordinary marketing language against a policy framework that most sellers have never read and Amazon never surfaces clearly.

Words that feel like smart copywriting ... "safe," "proven," "results," "high quality," "stress relief" ... are getting caught by the same automated layer that flags "CBD" and "coronavirus."

The gap between what you think is safe and what Amazon actually flags is wide. And it's getting wider every quarter.

This is especially brutal in supplements, skincare, and anything pesticide-adjacent, where the tolerance for unsupported claims is effectively zero. But it bleeds into every category.

Holiday words like "Christmas" and "Valentine's Day" can trigger promotional keyword filters. "Military" and "tactical" are restricted. Even "bulk" is flagged as a promotional term.

Sarah eventually got her listing back. It took 11 days and $9,200 in lost revenue. She rewrote every bullet, every description, every A+ content module with a restricted keyword list open in the next tab.

She said "I had no idea I was sitting on a minefield."

Most sellers don't. Until they step on one.

The fix isn't complicated. But you need the map.

Dream 100 member Vanessa Hung and her team at Online Seller Solutions put together a full restricted keywords guide that covers exactly this. It's a free PDF download, and here's what's inside:

It breaks down Amazon's title requirements (200 character max, no promotional phrases, restricted symbols, no word repetition, title case formatting rules).

Then it gets into the real minefield: hundreds of restricted and high-risk keywords organized by category. Guarantee keywords like "approved," "certified," "tested," and "validated" that require formal filing with Amazon.

Claim-related terms covering everything from "antibacterial" to "mold" to "high quality." Prohibited ingredients you can't even mention in copy. Drug keywords. Skincare terms that are off-limits without FDA certification. Disease claim keywords (this list alone is massive... everything from "cancer" to "flu" to "anxiety" to "seizures").

Eco-friendly terms that require government certification. Promotional and time-sensitive words. Subjective claims. Supplement and medical terms. Pesticide claims.

Even a section on the FTC Green Guides for eco-product sellers and the full Enhanced Brand Content prohibited actions list.

It's not a surface-level overview. It's the actual compliance map.

Download it here

Audit your listings before Amazon does it for you.

Ecom Mastery AI speaker Jo Lambadjeiva says five months ago it was perhaps the biggest e-commerce shakeup ever.

But now direct checkout inside ChatGPT is gone. Buy buttons. Removed. In-chat purchases. Over.

From now on you get redirected to the retailer's website. Or their app.

What Went Wrong

Users were happy to research products in ChatGPT. But when it came to actually buying? They bounced.

And OpenAI couldn't fix it.

The numbers tell the story:

Facebook did the same thing with Facebook Shops. Built it. Hyped it. Killed it.

Google did the same with Buy with Google. Built it. Hyped it. Killed it.

Three tech giants and nobody has cracked in-platform commerce. Not yet.

What This Means for Brands

ChatGPT isn't leaving commerce. It's repositioning.

Instead of being the store, it becomes the concierge. The recommender. The top of the funnel.

The question is no longer "Can people buy from you on ChatGPT?"

It's "Does ChatGPT even mention your product?"

The Discovery Shift Changes Everything

Think about how product discovery works today on Amazon. A consumer types in a search term, scrolls through dozens of results, filters by price or ratings, and picks one. Sponsored ads influence placement. Reviews influence trust. Keywords influence visibility.

Now replace that with a conversation.

"What's the best portable blender under $50?"

ChatGPT doesn't return 47 results. It returns 3. Maybe 4. With reasoning behind each recommendation.

Just a direct recommendation based on whatever data the model has access to.

If your product is one of those 3, you capture outsized attention. If it's not, you effectively don't exist in that purchase journey.

The Pattern Repeating Itself

Every major platform goes through the same cycle with commerce:

Phase 1: "We're going to let people buy right here on our platform."

Phase 2: Merchants slowly integrate. Adoption stays low. Technical complexity escalates. Tax compliance creates friction.

Phase 3: "We've decided to focus on discovery and partner with retailers for checkout."

Phase 4: The platform becomes the most influential product recommendation engine in its category. And controls who gets visibility.

Facebook went through it. Google went through it. Now OpenAI is going through it.

Phase 4 is where the leverage is. The brands that figure out how to show up in that recommendation layer capture disproportionate market share.

What Sellers Should Be Doing Now

1. Clean up your product data.

AI systems rely on structured data feeds like Schema and LLMS.txt files. Plus, if your product titles are still keyword-stuffed from 2023, AI models can't properly categorize or recommend you. Clean titles. Accurate descriptions. Proper categorization. This is table stakes now.

2. Build brand signals outside of Amazon.

ChatGPT pulls from the open web. If your product only exists as an Amazon listing, you're largely invisible to AI recommendation engines. Third-party reviews, press coverage, YouTube reviews, Reddit discussions, blog mentions ... these are the signals AI models use to determine authority and relevance.

3. Take AEO (AI Engine Optimization) seriously.

This is the new SEO. How do you show up when a consumer asks an AI assistant for a product recommendation? It's not about keyword density anymore.

It's about being the answer. Having the most complete product information. The most credible third-party validation. The clearest value proposition. ( DragonFish can help you with this ).

4. Pay attention to Google's Universal Commerce Protocol.

While OpenAI pulled back from checkout, Google is advancing its own commerce standard called the Universal Commerce Protocol (UCP). It's designed to let AI systems access structured product data from merchants without handling the transaction.

If you're already submitting product feeds through Google Merchant Center, you have a foundation to build on. If you're not, that's a gap worth closing.

5. Watch Amazon's next move.

Amazon has the product data. The reviews. The fulfillment network. The purchase history. If Amazon becomes the default checkout destination for AI-recommended products, every brand on the platform benefits from that traffic flow.

But Amazon also gets to decide who surfaces in those recommendations. The dynamics are familiar, just on a new playing field.

ChatGPT shopping isn't dead. It relocated.

The buy button is gone. But the recommendation engine is just getting started.

And that recommendation engine matters more than any checkout button ever did. Because the checkout was the last step. The recommendation is the first step. And whoever controls the first step controls the sale.

If you're not optimizing for AI discovery right now, you're focused on the wrong part of the funnel.

For those of you in Las Vegas this week for Prosper, be sure to stop by the Dragonfish/Ecom Mastery booth (BDSS on the map) and play Murderbot for a chance to win $8,000.00 in prizes and more!

And Tuesday night, join us at Atomic Golf

And don’t forget to your unofficial 2026 Prosper shirt. You can reserve one now.

While at Prosper, swing by the Ecom Mastery AI booth and Kevin King will be around signing shirts. Once you’ve got the signatures, you can toss your name in for giveaways.

They’re giving away up to $5k in prizes, including a Gold Ticket to Ecom Mastery AI featuring BDSS in Nashville (April 8–11).

Remember the aggregator gold rush?

Back in 2020-2021, Wall Street couldn't throw money at Amazon roll-ups fast enough. Over 100 aggregators raised $12.3 billion (75% of it debt) to buy up third-party FBA brands.

The pitch was simple: buy fragmented Amazon brands cheap, cut costs, stack them under one roof, and ride the pandemic e-commerce wave forever.

Prosper 2021 in Las Vegas was full of expensive satellite events, some costing hundreds of thousands of dollars, with most sponsored by Amazon aggregators who were rolling in cash.

Thrasio raised $3B+. Razor Group pulled in $800M from BlackRock and Victory Park. SellerX landed a $400M+ loan from BlackRock. Infinite Commerce, Berlin's Branded, Heroes, and dozens more piled in.

The problem? As Thrasio co-founder John Hefter later admitted, competition drove acquisition multiples from 2x EBITDA to 7x for what he called "Chinese vaporware garbage." The aggregators destroyed the very economics that made the model work.

Then the music stopped.

Post-2021, e-commerce growth normalized. Interest rates spiked. These highly leveraged roll-ups started suffocating under mountains of debt and unsold inventory. The model that worked at 0% rates flat out didn't work at 5%.

Thrasio filed Chapter 11 in February 2024 with $495M in debt. SellerX defaulted on BlackRock in 2023. Razor Group absorbed distressed rival Perch in 2024 (converting roughly $400M of Perch debt into equity), then merged with Infinite Commerce in August 2025.

Now BlackRock's TCP Capital Corp. just wrote down a $25M loan to Infinite Commerce from 100 cents on the dollar to zero. The loan was marked at par just three months ago. This is their second abrupt zeroing in recent months.

BlackRock TCP's Q4 2025 was ugly. NAV dropped 19% in a single quarter, from $8.71 to about $7.07 per share. Most of that pain came from aggregator exposure.

The Razor Group position got zeroed in Q3 2025 for a $72.6M loss. SellerX was partially written down after a 2024 debt-for-equity swap made BlackRock its largest shareholder. And now Infinite Commerce joins the zero club.

Victory Park Capital also fully wrote off its Infinite Commerce position, blaming weak demand and tariff-driven inventory costs. BlackRock waived a third of its own management fee.

It's a bloodbath.

When easy money meets impossible valuations and no operational discipline, everyone loses. The smart operators in this space always knew you can't just financial-engineer your way to success on Amazon. You actually have to know how to run the brands.

✌🏼 See you again Thursday …

The answer to today’s STUMP BEZOS is
Amazon loses about $1.2M per minute when it is down