If Your AI Plan Is "We're Big on Amazon," You're Building Someone Else's Moat

On June 25th, Amazon bought an ad inside ChatGPT against the query “best dental chews for dogs.”

We know because we recorded it.

In that same category, over the same six weeks, we logged 146,401 citations — every source ChatGPT pulled from when it answered dog-supplement questions. Amazon appeared in three of them.

Three.

Amazon told it to stay out. There’s a public file on Amazon’s own site — amazon.com/robots.txt — that blocks every AI agent in the market from reading anything at all. GPTBot, ChatGPT-User, PerplexityBot, all of them. Ten seconds to check.

So Amazon locked the door, vanished from the answer, and then paid to stand next to it. Sounds like a mistake. It’s the most deliberate thing in this whole story, and if you sell on Amazon, you’re the one paying for it.

What we measured

Vektor10 records what ChatGPT cites and what it recommends across our clients’ categories. Between June 1 and July 16, 2026, we captured 756,565 citations across 38 brands, plus 173,523 product observations — each time ChatGPT put a product in front of a shopper and told them where to buy it.

Amazon is 0.21% of those citations. 1,603 out of 756,565.

Retailer	Citations	Brands it appeared for
Chewy	10,036	5
Walmart	9,093	36
Target	7,435	33
Amazon	1,603	10

Walmart is cited 5.7x more than Amazon. Chewy, a pet store, beats Amazon six to one.

The buying side is the same story. Of 173,523 times ChatGPT recommended a product and named a merchant, Amazon was the merchant 359 times. Also 0.21%. Amazon doesn’t crack the top 25 — that list is Target, Ulta, Walmart, Sephora, Wayfair, CVS, Chewy, Petco, Nordstrom. The best-performing Amazon variant ranks 108th.

Per brand, it’s starker. Our Place: zero. A protein brand: zero. A wall-art brand, a skincare line, a kids’ nail polish: all zero. The dog-supplement brand from earlier: three out of 146,401. The only brands where Amazon registers at all are the ones whose entire business runs through Amazon — a household cleaner at 3.16%, a skincare brand at 1.51%. If you don’t live on Amazon, Amazon doesn’t show up next to you.

Amazon shows up in the text of 2.78% of answers, and gets linked in 0.21%. ChatGPT will happily talk about Amazon. It just won’t send you there.

Why

None of this is mysterious. Amazon published the reason.

Go read amazon.com/robots.txt yourself. It’s a public file. Amazon names every AI agent in the market and tells each one the same thing:

GPTBot — Disallow: / OAI-SearchBot — Disallow: / ChatGPT-User — Disallow: / PerplexityBot — Disallow: / ClaudeBot — Disallow: / Google-Extended — Disallow: / GoogleAgent-Shopping — Disallow: /

And a dozen more. CCBot, meta-externalagent, Copilot, Bytespider, Devin.

Now go read Walmart’s robots.txt. Then Target’s. Neither one names a single AI crawler. The doors are wide open.

Same file, opposite decisions, and you can check both in about ten seconds. One group owns the AI shelf. The other has a locked door.

So the 0.21% is policy. Amazon wrote down that the agents can’t come in, and 756,565 citations are what that decision looks like from outside the wall.

What Amazon is actually protecting

Most takes read the wall as Amazon being scared, or slow, or fumbling the AI moment.

Amazon isn’t fumbling anything. They’re protecting the most valuable pile of commercial data on earth, on purpose.

Inside that wall: two decades of reviews. Q&A. Pricing history. Every listing, every image, every bullet. And underneath it, who bought what, after searching for what, at what price, and what they bought next.

All of that is why Amazon’s ad business did $68 billion last year. Plenty of companies have shoppers. Amazon has the ad, the decision, and the purchase all happening on one surface it owns, so it can prove to an advertiser exactly what their dollar did. That closed loop is the moat.

If Amazon opened the crawlers, ChatGPT would read every review, every price, every spec, free, and become the place where people decide. Amazon turns into the warehouse out back. The $68 billion doesn’t vanish overnight, but it stops being owned and starts being rented, a slice at a time, from a company that just ate the front door using Amazon’s own data as the filling.

Amazon ran that math and got the obvious answer. Keep the door shut.

The way they’re fighting it gives the game away. They sued Perplexity in November, and the claim is the Computer Fraud and Abuse Act — unauthorized access. Not copyright. Not trademark. Amazon isn’t arguing anything was stolen. They’re arguing about who controls the door. (The Ninth Circuit heard it June 11th and still hasn’t ruled, so that one stays open a while.)

Then there’s what Andy Jassy said on the Q3 2025 earnings call. Third-party agents “get a lot of the product details wrong. They get a lot of the pricing wrong.” He’s right — they can’t read Amazon’s details or Amazon’s pricing, because Amazon blocked the crawlers that would let them. He also said Amazon expects “to find ways to partner” with those agents over time.

That’s a landlord telling you the rent hasn’t been set yet.

So why buy the ad at all?

Because Amazon knows better than anyone what an ugly little ad unit turns into.

I wrote about this in June, when eMarketer forecast that OpenAI misses its $100 billion ad target by 2030 — and I took the other side of that bet. On the numbers in front of us right now, eMarketer is right. The ads are bad. No price, no button, no reviews, parked in a footer under the answer where your eye already left.

Amazon’s ad units were bad too, once. CNBC called Amazon “the sleeping ad giant” in 2013, which was a polite way of saying nobody thought the giant would wake up. Amazon didn’t break advertising out as its own line until 2018, and when they did it turned out to be a $10 billion business that had been hiding inside a segment labeled “Other.” By 2025 it was $68 billion.

Amazon lived that story. They know what the floor looks like because they were the floor.

So we catch Amazon running paid ads inside ChatGPT across seven categories and 81 different shopping queries, served from aes.us-east.ono.axp.amazon-adsystem.com — Amazon’s own ad system, delivering into OpenAI’s product. That’s a company who watched an ugly ad unit turn into the third-largest ad platform on earth, buying a cheap option in case it happens twice.

It’s not much money. The queries are what got me:

“best affordable probiotic for dogs” “best dental chews for dogs” “best supplement for dog teeth and gums” “best probiotic for dogs with sensitive stomachs” “best vet recommended dental chews for dogs”

Every one of those is a query where Amazon has effectively no organic presence, in a category where it was cited three times out of 146,401.

During the Prime Day window we captured 207,463 citations. Amazon appeared in 3. Its mentions in answer text actually fell — 0.64%, against 3.69% across the rest of our window. Zero organic presence, paid presence, same week.

The creative says where they are in the process. “Best Pet Gear of 2026.” “Prime Deals: Pets.” “Top-Rated Kitchen Gear.” “Buy in Bulk.” Category-level, every time. Never a specific product. You don’t advertise a category when you’re trying to sell a thing — you advertise a category when you’re still figuring out whether the channel works.

Protect the moat. Buy the option. It’s a good strategy.

One more thing, and I almost missed it. While Amazon blocks ChatGPT from reading amazon.com, its own third-party sellers are buying ChatGPT ads that point straight at their Amazon listings — and they’re outspending Amazon on the channel roughly five to one. Philips Norelco, Cuisinart, First Aid Beauty. Amazon won’t let the agents in. Its sellers are buying them tickets. Amazon.com is the ninth most common landing page across every ad we recorded.

Which brings us to you

Amazon’s wall is working. It’s protecting Amazon’s data, Amazon’s closed loop, and Amazon’s $68 billion.

It is doing nothing for you.

Every review you’ve earned over there, every piece of A+ content you paid an agency to write, every Q&A you seeded, every ranking you fought for — all of it is inside that wall. It’s an asset. It’s Amazon’s asset. You built it, Amazon monetizes it, and the moment a shopper asks ChatGPT instead of typing into a search bar, none of it does a thing for you.

You’ve spent a decade making Amazon’s moat deeper. That was a fine trade when Amazon was where the deciding happened. The deciding is moving somewhere Amazon has locked itself out of on purpose, and your equity doesn’t come with you.

Walmart and Target left the door open, and they’re getting cited 5-6x more than Amazon for it. Nobody had to be clever. They just didn’t lock it.

So what earns the AI shelf? Not retailers. 86.7% of every citation we recorded came from editorial sources — the roundups, the reviews, the comparisons, the “best of” lists. That’s the shelf now, and it gets earned in public, off Amazon, on sites that let the crawlers in.

The honest part

I could be wrong about all of this, and here’s how.

Amazon’s bet is rational. It might also be the same bet newspapers made about Google. They were right that the content was the asset and wrong about what came next, because Google routed around them, and the moat ended up protecting a building nobody walked into.

Amazon’s difference is real — they own the transaction, so they can wait a lot longer than a newspaper could. If the AI shelf stays a novelty, Amazon looks brilliant for never giving away the crown jewels. If it doesn’t, they own the best shelf in an empty aisle.

Nobody knows which way that goes. Not eMarketer, not me, and judging by that ad buy, not Amazon either. That’s what an option is for.

You don’t have to settle that to act, though, because your side of it is already settled. Whatever happens to Amazon, your Amazon equity stays inside Amazon’s wall. It was never going to travel.

Go ask ChatGPT what the best product in your category is. See who it names. See where it sends them. That’s the assignment for this week, and it takes about four minutes.

---

Vektor10 measures the recommendation shelf — what AI actually cites, recommends, and sends buyers to. If you want to know where you stand on it, come talk to us.

---

Notes and sources. Our data covers 756,565 citations and 173,523 product observations across 38 brands, June 1 to July 16, 2026. Client brands other than Our Place are described by category rather than named. robots.txt files fetched July 16, 2026, and are publicly verifiable. Jassy quotes from Amazon’s Q3 2025 earnings call, October 30, 2025. Amazon v. Perplexity is No. 3:25-cv-09514-MMC (N.D. Cal.), argued before the Ninth Circuit June 11, 2026, undecided as of publication. Amazon’s FY2025 advertising revenue of $68B+ is from company earnings releases. Amazon’s ChatGPT ad buy was first reported by Modern Retail; our data independently confirms it.

Amazon declined to comment to other outlets on its ChatGPT advertising. We didn’t ask — we just counted.