BDSS Dream 100 member Vanessa Hung (founder/CEO, Online Seller Solutions) says you need to get your listing on the right shelf. Then figure out how Amazon decides whether it lets you keep it there.

Part one: fix your categorization before it quietly kills your traffic

In the AI era, your data quality is your product quality. Great title, killer bullets, beautiful A+ content. None of it matters if Amazon has you filed on the wrong shelf. Alexa for Shopping, A9, or whatever routes the customer can't find a product it can't categorize.

Three attributes decide where your listing lives:

Steal her analogy. The browse node is the highway. The prototype is the avenue. Your product is the little house. Alexa is the Uber driver. If the driver doesn't know what street you're on, you never get the ride. Doesn't matter how nice the house is.

Why it's urgent: this past February 26th, Amazon restructured the browse tree guide through a flat file change. It quietly reshuffled prototypes and item type keywords.

Plenty of listings fell out of alignment and sellers have no clue. They just watch traffic slide.

Her example: a product showed "air mattresses" in the front-end path, but the backend item type keyword said "inflatable beds." Close, not fatal, not aligned.

Rule of thumb: the last word of your category path should match your item type keyword. When it doesn't, you leak traffic.

The gotcha most sellers miss: Amazon's policy now says your title, bullets, and description are key attributes used for product classification. You used to pick your category. Now Amazon reads your copy and picks for you. So the fix isn't just a backend flip. You may have to rewrite copy to push the algorithm to the right category.

The steps:

Part two: Vanessa doesn't stop there. Now make sure your edits win.

Here's the catch. You can fix all of this and still get overwritten. It only makes sense once you know who Amazon lets make changes.

The Amazon catalog runs on a contribution hierarchy. Every contributor gets a score from 0 to 100. Highest score wins, every time.

The breakdown:

Every attribute has one authoritative owner. Title, brand name, bullets, images. Each one.

So when two contributors submit different values for the same field, Amazon doesn't negotiate. It compares scores, keeps the higher one, rejects the other automatically.

Think of your listing as a leaderboard. If you're not the top score on a field, you're locked out of it. Someone else owns it.

That's why the jump matters. Brand Registry lifts you from 30 to 52, and that climb puts your edits above the open catalog. The points you earn are the points that protect your business.

Audit your backend item type keyword against your front-end category path. If they don't match, your copy is routing you to the wrong shelf, and no amount of PPC fixes a listing the algorithm can't place. Then check your contribution score. Fixing your listing only counts if you outrank everyone else trying to change it.

Amazon just added a new disclosure step. When you upload images to A+ Content or Brand Stories, you now have to tell Amazon two things. Is the image AI-generated? And does it show a photorealistic AI person?

Here's what's actually happening:

So does this change anything today? Not really. Nobody's getting suppressed over it right now.

But pay attention to the direction. Amazon is building a paper trail on AI content. Once they have the data, they have options. Ranking signals. Compliance rules. Labels shoppers can see. None of that exists today, but the plumbing is going in.

If you're leaning on AI for your creative, keep track of what you flagged and where. The disclosure is easy now. The rules that come later might not be.

A tool called Higgsfield Supercomputer is making the rounds, and one of its three demo workflows is aimed straight at you.

The pitch: one chat, one prompt, a finished Amazon brand.

The operator types "suggest three unbranded Amazon products worth selling, include price, margin, and competition." The agent browses bestseller lists in real time, runs FBA fee math, reads reviews to find gaps, and hands back three picks. It lands on dog socks. $19 retail, a buck to make, 71% margin, 68k monthly searches.

Then it names the brand, builds a product sheet, generates listing images, studies winning ads on Meta, TikTok, and Shorts, and writes 15 ad concepts with video to match. All in one window. Slick demo.

Where the hype falls apart

Watch the product research step again. The agent "finds" dog socks. $19 sell price, $1 cost, 68k searches, "none with real branding, none with video ads." A virgin market hiding in plain sight.

Think about that for one second. If an AI can surface that opportunity in 12 seconds from a generic prompt, so can the 50,000 other people who watched this same video and ran the same prompt this week. The "low competition" it's bragging about evaporates the moment the tool exists. You are not early. You are the demo.

This is the AI hype trap in a nutshell. The flashier the part of the workflow, the less edge it actually gives you, because flashy means easy, and easy means everyone has it.

Product research, niche picks, "winning product" finders. That's the casino floor. The house already knows the dog socks gold rush is coming.

Where it earns its keep

Now the unsexy part nobody clips for the highlight reel. Asset production.

Listing image sets. Infographics. Comparison cards. Lifestyle shots. Fifteen ad variants in a sitting. This is the work that actually eats your weeks and drains your agency retainer, and it's the work AI genuinely compresses from days to minutes.

One real edge here: it watches competitor video ads with the audio, instead of reading a text summary of them like most tools. For modeling what's working in your niche, that's worth something.

The pattern holds across every AI tool you'll evaluate this year. The boring middle of the workflow is where the money is. The magic-button bookends are where the hype is.

Two things it won't save you from

The output still has AI tells. Hands clipping through products, items morphing between cuts. Nothing ships unreviewed.

And none of it is Amazon-compliant out of the box. Main image on white, no badges or text on the hero, claims you can actually back up. The agent doesn't know your category rules. That's still on you, and a suspension doesn't care that a robot made the asset.

Under the hood it just routes between Claude, GPT, and Gemini per task. The model isn't the product. The wrapper is.

Bottom line: lousy oracle, useful factory. Use AI to build faster, not to think for you.

✌🏼 See you again Thursday …

The answer to today’s STUMP BEZOS is
Only a small 2% of apps launched in 2023 are gone.