Your Amazon Listing Can Change Overnight — and Amazon Won’t Always Tell You

You check your listing on a Tuesday. Everything looks fine. By Thursday, the title is different. The second bullet point says something you never wrote. Or — and this actually happened to a seller — your product title has been replaced with: “I’m sorry but I cannot fulfill this request.”

That’s not a hypothetical. That’s the output of a malfunctioning AI system applied to a live ASIN. The seller didn’t make the change. Amazon did.

Here’s what most sellers still don’t understand: Amazon is not just a marketplace anymore. It’s an active participant in your catalog. It is writing your listings, testing variations against your versions, and in some cases doing it without asking you first. The rules have changed. If you’re running a brand on Amazon in 2026 and you’re not monitoring your listing content, you’re flying blind.

This post is for brand sellers who want to understand exactly what’s happening, what’s confirmed, what’s reported, and what to do about it.

There are two distinct AI programs operating on Amazon’s catalog right now. They work differently. They have different levels of transparency. And they carry very different risks.

The first is Enhance My Listing. This is Amazon’s official, opt-in generative AI tool. When you’re in Seller Central, Amazon surfaces AI-generated suggestions for your title, bullets, and description. You see a recommendation, you approve or reject it, and it either goes live or it doesn’t. That’s the stated flow.

The second is what’s been reported as Project Starfish. This is a different animal entirely — and I’ll cover it in detail below.

The distinction matters because sellers are conflating these two programs. When a listing change happens and a seller doesn’t know why, they often blame one when the other may be responsible. Knowing which system touched your listing — and when — determines what your response should be.

Start with the confirmed facts, because they’re alarming enough on their own.

According to Amazon’s own reporting on aboutamazon.com, Enhance My Listing has reached more than 900,000 selling partners. Here’s the number that should make every brand seller pay attention: approximately 90% of sellers who receive AI-generated suggestions accept them with little or no editing.

Read that again. Nine out of ten sellers are publishing AI-written copy to their live listings with minimal review.

I’ve talked to sellers who do this because the suggestions look good at a glance. The AI generates something plausible, the seller is busy, and they hit accept. What they don’t always check: whether the new title contains a claim that violates Amazon’s policies. Whether a bullet contradicts something in their A+ content. Whether the description removes a differentiator that was driving conversions.

The tool itself isn’t the problem. Generative AI can write competent product copy. The problem is the behavior it’s producing — which is mass, low-scrutiny acceptance of machine-generated content on live listings.

And here’s the part Amazon doesn’t advertise: if you receive a listing change recommendation and don’t act on it within 14 days, it goes live automatically. That’s not a bug. That’s how the Review Listing Changes queue in Seller Central works. If you miss the notification — buried in a dashboard you check once a week, on a day when you’re dealing with a shipping issue — the AI’s version becomes your listing.

That’s the opt-in tool. Here’s what’s been reported happening without the opt-in.

Amazon has not publicly confirmed a program called Project Starfish. What I’m about to describe comes from Seller Labs’ reporting on internal Amazon documents — and that framing matters. These are not confirmed Amazon policy statements. They are descriptions from leaked internal materials, and sellers should understand that distinction.

With that said, here is what Seller Labs’ reporting describes:

According to Seller Labs’ reporting on internal documents, Project Starfish is an AI initiative designed to optimize Amazon’s catalog at scale. As reportedly described in those documents, the system scrapes approximately 200,000 brand websites, as well as content from manufacturers, distributors, and competitors. It then uses that data to auto-rewrite product titles, bullet points, and descriptions across millions of ASINs. The reported initiative also includes generating AI-produced images and video for listings.

The detail that has gotten the most attention: according to Seller Labs’ reporting, the system A/B-tests AI-generated versions of your listing against your seller-written version — often without notifying the seller that a test is running.

Internally, per that same reporting, the initiative was projected to drive $7.5 billion in GMV improvement.

What we don’t know: We don’t know the current operational scope of this system. We don’t know precisely which ASINs or categories have been affected. We don’t know Amazon’s internal timeline. Amazon has not confirmed these details publicly.

What we do know: Sellers are reporting listing changes they didn’t make. The ChatGPT error message that appeared in a live product title — that’s documented. Catalog edits that bypass Brand Registry protections on certain fields — that’s been observed. The mechanism behind those changes isn’t always clear, but the changes are real.

The reported initiative, if it operates as described, would represent a fundamental shift in who controls catalog content on Amazon. Not you. Amazon.

That’s the setup. Here’s what to do.

Here’s where it gets consequential.

Most sellers think about listing changes as a conversion problem. If the title is wrong, you lose clicks. If the bullets are weak, you lose conversions. That’s true. But it’s not the most dangerous outcome.

The most dangerous outcome is an AI-modified listing that now contains a prohibited claim, a compliance violation, or a policy-triggering phrase — and Amazon’s Account Health system flags it before you even know the change happened.

Amazon’s Account Health enforcement has become increasingly automated and pre-emptive. The system identifies patterns, enforces first, and routes to human review after. That sequence is important. By the time a human looks at your case, your ASIN may already be suppressed.

So the scenario plays out like this: an AI system rewrites your title to include a health claim that crosses a policy line. Your listing goes live with that change. Amazon’s compliance system flags the claim. Your listing gets suppressed — meaning there is no Buy Box on that ASIN, no organic placement, no ad serving. You get a notification. You log in to investigate. And now you’re filing an appeal for a change you didn’t make, on a listing that’s already dark.

That’s not hypothetical. That’s a realistic sequence of events given how both systems currently operate.

The Buy Box distinction matters here. A Buy Box lost means a competitor holds your Buy Box — you’re eligible, but someone else is winning. A Buy Box suppressed means the Buy Box doesn’t exist on the listing at all. An Account Health enforcement event from a content change produces a suppressed Buy Box, not a lost one. These are different problems with different remedies, and sellers who conflate them will chase the wrong fix.

If you’re in Brand Registry, you have more protection than an ungated seller. But you don’t have the protection most sellers think you have.

Here’s how the content authority hierarchy actually works.

Brand Registry A+ content gives you the highest authority tier for the content it covers. If you have A+ content live on an ASIN, you have meaningful catalog authority. But A+ content protects your enhanced content module — not everything.

The BrandCatalogLock mechanism, available to brand-registered sellers, does protect your title, hero image, bullets, and description from unauthorized catalog edits. I’ve seen sellers — particularly those who’ve dealt with unauthorized catalog edits from bad actors — activate this after the fact, when the damage was already done.

Here’s what BrandCatalogLock does not protect: your A+ content itself (ironically), your backend keywords, your variation structure, and your category node assignments.

Category node changes are particularly dangerous and underappreciated. If your ASIN gets moved to the wrong category, your Buy Box eligibility criteria change, your ranking signals shift, and your search placement deteriorates — all without a single word of your listing copy changing. Sellers focused only on title and bullets often miss this entirely.

The practical implication: Brand Registry is necessary but not sufficient. You need to know when any of these fields change, not just the ones you think are protected.

I’ve been watching this play out across a lot of catalogs. Here’s my honest take on what matters.

First: audit your listings for changes you didn’t make. Pull your current live titles, bullets, and descriptions and compare them to what you originally submitted. If you haven’t done this in the last 90 days, there’s a non-trivial chance something has changed. Do this today.

Second: stop treating the Enhance My Listing queue as low-priority. That 14-day auto-apply window is a real risk surface. Set a recurring calendar reminder to check it weekly. If you have a large catalog, assign someone to own this. The 90% acceptance rate across the platform means sellers are not doing this review seriously — that’s an opportunity if you are.

Third: don’t assume Brand Registry protects everything. Map out which fields you care about most. Title, bullets, hero image — the BrandCatalogLock covers these. Backend keywords, category nodes, variation relationships — these are exposed. Know where your gaps are.

Fourth: connect listing content changes to your Buy Box data. A content change that doesn’t affect your Buy Box wins right away may still affect them within days — once the algorithm re-evaluates the updated content, or once Account Health flags something. If you’re tracking Buy Box wins in isolation from listing content, you’ll miss the causation. You’ll think it’s a pricing issue when it’s actually a content flag.

Fifth: understand this is a structural shift, not a temporary experiment. The structural shift in how Amazon manages catalog content is not going to reverse. Amazon has hundreds of millions of ASINs and a clear financial incentive to optimize listing quality at scale using AI. That is not going away. The sellers who build monitoring into their operations now will be ahead of the ones who discover this the hard way.

I’ve seen brands with clean, well-maintained listings still get caught off guard by a field change they didn’t notice for two weeks. By then, the conversion drop is real and the diagnostic is complicated. The single biggest operational change a brand seller can make right now is moving from periodic listing audits to continuous listing monitoring.

The competitive surface on Amazon has expanded. It used to be: have a good product, price it right, run ads, win the Buy Box. That’s still the foundation. But there’s a new layer now.

Your listing is no longer static infrastructure. It’s a live document that multiple systems — some under your control, some not — can modify. The content of that listing determines your policy compliance, your conversion rate, your Buy Box eligibility, and your ability to appear on AI-driven shopping surfaces that are increasingly mediating the path between Amazon’s search and the purchase.

If you’re investing in Amazon’s AI tools for sellers — ads automation, demand forecasting, campaign optimization — and you’re not monitoring your listing content with the same rigor, you’re optimizing one part of the funnel while leaving the foundation exposed.

The sellers who are going to win the next phase of Amazon aren’t the ones who fought AI. They’re the ones who built operational infrastructure to detect change fast, respond intelligently, and stay in control of what’s on their page.

That’s not optional anymore. That’s the job.

SentryKit monitors your Buy Box and listing status in real time — so you see content changes, suppression events, and Buy Box shifts before they become a crisis. The Content Changed alert fires the moment your listing fields are modified. The Listing Suppressed alert catches ASIN dark events before they drain your revenue.

If you’re managing a catalog with real brand equity, you need to know when your listing changes. Not two weeks later. Now.

Start monitoring your listings with SentryKit →