Christian Umbach at Autopilot says rank tracking made sense in 2016. But in 2026, it's measuring the wrong thing.

Why Rank Tracking Is Dead

There is no "the rank." Amazon personalizes the SERP by browsing history, purchase history, location, device, and time of day. Millions of ranks, one per shopper. A scraper from a data center isn't seeing what your customer sees.

Mobile rewrote the page. Over half of Amazon traffic is mobile. Most rank trackers still scrape desktop. Your "position 8" might not exist where your customers actually shop.

Amazon shuffles every product 4-to-12-to-6 to test placements. That noise stabilizes in 72-96 hours. Teams and tools react to it daily and tank their own bids.

The SERP isn't where the business lives anymore. Browse nodes, Subscribe & Save, Deals, DSP, Sponsored Display, external traffic, Rufus. Rank tracking measures a shrinking slice with growing precision.

Rank tells you nothing about outcomes. You can rank #1 on a zero-volume keyword. A competitor can eat your share while you sit at position 2. Rank is a proxy for visibility, visibility is a proxy for traffic, traffic is a proxy for sales. Three layers from what matters.

The Real Tool: Search Query Performance

If you have Brand Registry, SQP is sitting in your Brand Analytics tab. Most brands barely touch it.

Every week, for every ASIN, on its top 100 queries, Amazon hands you four numbers as first-party data:

No scraping. No proxies. No 72-hour shuffle artifacts. Goes back two years. Available via SP-API.

The Insight That Buries Rank Tracking

SQP doesn't just give you share. It gives you Total Count (the market's absolute volume) and Brand Count (your absolute volume).

Now you can answer the only question that matters: is it us, or is it the market?

Rank dashboards show you a flat green line in all five.

Branded vs Unbranded

One more cut. Pull your queries apart.

Winning purchase share on "YourBrand water bottle" is loyalty harvest.

Winning purchase share on "stainless steel water bottle" is category conquest.

Only one builds a durable business. Only SQP separates them cleanly.

Why This Gets More Important, Not Less

Rufus is starting to surface. Amazon is giving brands early access to Rufus prompt data. Conversational queries fragment into longer, higher-intent phrases. "Stainless steel water bottle" becomes "an insulated bottle that keeps coffee hot for an 8-hour shift and fits a car cup holder."

Rank means even less in that world. The queries fragment. The underlying markets persist. Whoever measures share of demand stays oriented.

Categories spike. Halloween barely exists for 11 months, then becomes one of the largest categories on Amazon. Back-to-school compresses months into three weeks. A rank tracker can't tell you your category is 10x normal this week. SQP's Total Count will, and your Brand Count tells you if you caught the wave.

Your Move

Stop opening the rank dashboard tomorrow morning.

Pull SQP. Pick your top 20 queries. Run the funnel diagnostic. Separate branded from unbranded. Track share and market size weekly.

You're doing 2016's work with 2026's stakes. The data is already in your account.

Amazon just turned Rufus into a profile-matching machine.

A new feature called "Tell us about you" lets shoppers type free-text descriptions of themselves into Rufus. Style. Hobbies. Kids. Pets. Apartment size. Whatever they want to share. Amazon saves it as a persistent identity profile tied to the account.

That profile follows the shopper everywhere. Every search. Every Alexa request. Every Rufus chat. Every device. The query stays the same. The identity layer underneath it changes the results.

Two shoppers type "storage bins." One has told Rufus she has three kids and a small apartment. The other said he runs a minimalist home office. They get different ASINs. Same keyword. Different stacks.

Vanessa Hung surfaced the rollout on April 22 after Ritu Java spotted it first.

Why this matters for sellers

The old game was keywords, reviews, and conversion rate. Those signals still count. But there is now a new layer sitting between the search bar and your ASIN: who Rufus thinks the shopper is.

Your bullets, A+ content, and backend keywords now do two jobs at once. They feed the keyword index like always. They also get read against a saved shopper profile.

Listings that name the person win. "Designed for families with young kids." "Built for small home offices." "Perfect for pet owners in apartments." Those phrases are anchors. Rufus maps them to real profiles on file.

Generic spec sheets do not anchor to anyone. Dimensions, materials, feature lists with no human in them are invisible to the profile-matching layer. Doesn't matter how well you rank on the keyword.

The scale

300 million people used Rufus in 2025. $12 billion in incremental sales. Monthly users up 149% year over year. Interactions up 210%.

Shoppers who use Rufus convert 60% higher than shoppers who don't.

This is not a side feature. This is the front door.

Build the buyer persona first. Then write the bullets. Then the A+. Then the backend.

Identity language is the new keyword stuffing. Except it actually has to be true and specific.

The bigger picture

Amazon's ad layer is getting more AI-mediated by the month. AI shopping prompts went paid on March 25. You can win the auction and still lose the match if your listing doesn't speak to the shopper Rufus has on file.

Paid placement gets you in the room. Identity language gets you the sale.

The shift is permanent. Rufus is not going to forget what shoppers told it. Sellers who write to a person beat sellers who write to a search engine.

Start now.

Here’s an important tip from BDSS Dream 100 member Vanessa Hung:

What happened: A seller hit #1, then their US listing vanished the same day. The attack came through Mexico, not the US.

Hijackers uploaded a flat file in the Mexico marketplace flagging the ASIN as an adult product. Amazon's catalog applied it upstream and suppressed the US listing.

Why you're exposed: In 2022, Amazon auto-enrolled every active US FBA seller in NARF (North America Remote Fulfillment). Your ASINs got duplicated in Canada and Mexico automatically. Those duplicate listings accept external flat file contributions, which is the open door.

Action steps:

30 seconds to check. Worth doing before the next ranking win, not after.

Now everybody say, “Thanks Vanessa” and be sure to visit her website

Anthropic just ran a one-week experiment called Project Deal. They turned their San Francisco office into a Craigslist-style marketplace, but with a twist: AI agents did all the negotiating.

Here's how it worked.

69 employees signed up. Each got a $100 budget. Claude interviewed every participant about what they wanted to sell, what they wanted to buy, and how they wanted their agent to negotiate.

Those answers became custom system prompts. Then the agents went to work, posting listings, haggling with each other, and closing deals on behalf of their humans.

The numbers:

Now here's the part sellers should pay attention to.

Anthropic ran a secret parallel test. They quietly assigned some participants their top model (Claude Opus 4.5 at the time) and others their smallest model (Claude Haiku 4.5). The result? People with the smarter model got measurably better outcomes. People with the weaker model got worse deals and didn't notice.

Read that again. The losers didn't know they were losing.

This is the agent-to-agent commerce future everyone's been theorizing about. And it's already showing the same dynamic that's about to hit Amazon, TikTok Shop, and every other channel: the quality of the AI representing the buyer (or the seller) determines who wins. Cheaper model, worse outcome. And the human on the losing end has no idea.

If shoppers start using agents to buy on Amazon, the seller whose listing, pricing, and offer are optimized for agent evaluation wins. The seller still optimizing for human eyeballs loses. Quietly.

Project Deal was a pilot with a self-selected pool of AI nerds. But the signal is loud. Agent-to-agent commerce is closer than it looks, and the model gap is going to matter.

✌🏼 See you again Thursday …

The answer to today’s STUMP BEZOS is
ChatGPT’s test generated $100 million in annualized revenue