ChatGPT Just Dropped Nearly 3,000 Stores. See If Yours Survived.

First, the honest part. If you sell service calls, this post is not about you. Go read the one about local AI answers instead.

This one is for the people with a catalog. Retail, parts, equipment, anything with products and prices.

Because something happened to you in July and I doubt anybody told you.

On July 10, 2026, the way ChatGPT Shopping picks which products to show changed. Not gradually. On one day.

Before that day, most of its picks came from searching the web. After it, most came off a list of connected product feeds.

Stores inside the list kept showing up. A lot of the stores outside it stopped.

Nobody announced it. Nobody emailed the stores that fell out. Their orders just got quieter.

TL;DR (the short answer)

On July 10, 2026, ChatGPT Shopping picks coming from connected product feeds jumped from 8.26% to 61.54%. In one day. Unique merchants referenced fell from 13,524 to 10,607. That's roughly 2,900 stores. Of 687 stores tracked, 450 lost at least a third of their Shopping visibility. All of that is Profound's tracking across 1,757,723 prompt runs, reported by Search Engine Journal on September 9, 2026. Sell on Shopify or Etsy and your catalog is already connected. Everywhere else, the feed application is waitlisted.

3 Things to Remember

Nearly 3,000 Got Dropped.

Here are the actual numbers.

Roughly 2,900 stores stopped being referenced.

And it bunched up hard. The top ten stores went from 22.5% of all references to 41.8%.

The pool got smaller. The big end of it got bigger. Same day.

Two out of every three tracked stores lost a third or more.

That's not a slow slide you'd catch in a monthly report. That's a cliff on a Thursday.

Most Stores Never Saw It.

Here's why this one slipped past everybody.

ChatGPT Shopping is not in your analytics the way Google is. No Search Console for it. No graph. No email telling you what dropped on July 10.

So the owner never sees a number move. He sees a slower month. He blames the season. Or the economy. Or the fact that he raised prices in June.

Same shape as my September 4 post, when Google ran a spam update and people assumed they'd been hacked. Nobody hacked them. Nobody wrote to them either. The traffic just left.

That's the modern version of getting dropped. It's silent.

You have to go looking for it on purpose. That's what the last section of this post is for.

The mistake: assuming a bad month is seasonal because nothing told you otherwise.

The fix: go check where you stand instead of waiting to be notified.

The payoff: you find out whether July was the season or the feed.

Your Listing Isn't Your Store.

Most people have this backwards. Let me say it plain.

A product feed is not your website.

A feed is a plain file that updates itself. It lists every product you sell. Price, stock, images, details. In a format a machine reads without guessing.

Google Shopping has worked this way for years. It's an old idea.

Your website is a page built for a person. Your feed is a list built for a machine.

Here's the change in one line.

It went from mostly reading the web to mostly reading the list, in a day.

So you can have a gorgeous store that loads fast and reads well. And still not be in the room where the picking happens.

Your listing is what's being shopped. The entry in that feed. Not your store.

And here's the part that stings. It has nothing to do with how good you are at your job.

Sell on Shopify or Etsy and your catalog got connected for you. You didn't apply. You didn't earn it. You were on the right platform when the music stopped.

Sell anywhere else and the direct feed application is waitlisted. You ask to be let in. Then you wait.

That's not fair and I won't dress it up. It's also where we are.

Bad Feeds Get Cut First.

Let me be careful here. I won't stretch a number past what it says.

Profound's data shows which stores were in the feed list and which were not. It does not measure whether anybody's feed was any good.

So I can't tell you sloppy feeds were the ones cut in July. I won't pretend otherwise.

What I can tell you is what a useless feed looks like. I've seen plenty of them. A broken feed is the same as no feed at all.

• It's stale. Prices from March, stock counts from last year. If it says in stock and it isn't, you get picked and then you disappoint somebody.
• It's thin. A title, a price, and nothing else. No size, no material, no model number. A machine cannot match a thin listing to a specific question.
• The titles are written for you, not the buyer. Internal SKU names. Cute product names nobody searches for.
• It quietly stopped updating. The most common one. Somebody set it up three years ago, something broke, and no alarm went off.

So here's the honest version. Being in the feed is what got you through July. Having a feed worth reading is what decides if that's worth anything.

One of those you control today. Go look at yours.

This Data Has Limits and I'm Going to Name Them.

Every number in this post comes from one company's tracking. Here's what that means before you go make decisions with it.

• It's observational. Profound watched outputs and measured what changed. Nobody at OpenAI confirmed what happened or why.
• It's tracked prompts, not real shoppers. Those 1,757,723 runs are Profound's own prompts. Not real customers with real credit cards. The shape is real. The size may not match your sales one to one.
• The GPT 5.6 link is an inference. Profound connects the July 10 shift to a model change. OpenAI has not confirmed that. It's a reasonable read, not a fact.

I'm telling you because the change is worth acting on even with those limits. And because you should know which parts are solid.

What's solid: the picking mechanism moved toward feeds, and stores outside the feed list lost ground.

What's inferred: exactly why, and exactly how much it cost any individual store.

Act on the first one. Don't quote the second one at your accountant.

See If Yours Survived.

This takes about twenty minutes and costs nothing.

• Ask it directly. Open ChatGPT and shop your own products the way a customer would. Best waterproof work boots under $200. Replacement pump for a specific model. Use buyer words, not your product names. Write down who comes back.
• Run it five or six times, different wordings. One result tells you nothing. A pattern tells you where you stand.
• Find out if you have a feed at all. On Shopify or Etsy, you're likely already connected. On anything else, ask whoever runs your store one question. Do we have a product feed, and where does it go?
• Open the feed and read ten rows. Right prices? Right stock? Enough detail to answer a real question? If you can't answer that, that's your project.
• Get on the waitlist. If you're outside Shopify and Etsy, apply for direct feed access. It costs you nothing to be in the line.

Then compare it to July. Pull your own numbers for the weeks either side of July 10. See whether your quiet month started on a Thursday.

Here's the thing I keep coming back to.

Every one of these shifts rewards the same thing. A clean, current, machine readable version of what you actually sell.

Feeds today. Something else next year.

The stores that keep winning are the ones whose information is right and easy to read everywhere. Without a person patching it by hand.

That's not a trick. That's having your house in order while everybody else finds out in September what happened in July.

Let me show you what that looks like for your catalog.

Take a Test Drive →

You get a real look at your own market. No obligation, and nothing to cancel.

Want the whole playbook first? Plan your attack. Balls Out Marketing.

FAQ