In 2012, Google announced that Google Product Search, the free shopping engine it had run since 2002 under the name Froogle, was becoming Google Shopping. By that October, every free listing was gone. If you wanted your products in shopping results, you paid.

I remember how the industry talked about it at the time. Google said charging merchants would actually improve the product. A commercial relationship meant better data, they argued. Paying customers keep their feeds accurate, free riders let their listings rot. Plenty of smart people nodded along…and to be fair, the data quality argument wasn't entirely wrong. It was just beside the point.

The point was that Google spent ten years teaching an entire industry to hand over its product data for free, and then one morning that data became the raw material for an ads business. Nobody got a vote. The merchants who built their distribution on free listings woke up as advertisers with a monthly bill, and the ones who complained got told the same thing complainers always get told: nobody forced you to be here.

This is not a Google story. This is how platforms work, every time, and the only variable is speed.

Google took ten years. OpenAI is running the same sequence in eighteen months, and the carousels that showed up in ChatGPT this week are the tell.

I've spent close to my career on the paying side of this trade. I've sat in the quarterly reviews where the channel that used to be free quietly became the biggest line on the budget. I've heard every version of the pitch that precedes the toll, and it always sounds the same: give us your data, it makes the experience better for your customers. It does. Right up until the day that data gets an auction attached to it, and by then you've spent years making yourself easy to charge. So when a platform starts asking brands to upload structured product feeds to improve answer accuracy, I don't need to guess what comes next. I've watched this movie enough times to know the ending before the lights go down.

Run the timeline yourself:

April 2025. OpenAI adds shopping recommendations to ChatGPT search. Organic, unsponsored, all users. Brands are encouraged to clean up their product data so ChatGPT can recommend them accurately.

September 2025. Instant Checkout launches. Buy inside the chat, starting with Etsy sellers, a million Shopify merchants promised. Agentic commerce is the future, everyone writes the same post.

January 2026. OpenAI confirms it's testing ads. The pilot goes live February 9. Text ads at the bottom of responses for US users on the Free and $8 Go tiers. Clearly labeled, physically separated, and, per OpenAI, ads don't influence the answers.

March 2026. Instant Checkout dies. Roughly five months old. OpenAI's blog frames it as focusing "on product discovery" while merchants keep their own checkout.

May 2026. Product feed campaigns arrive, first reported by Digiday. Retailers connect a product catalog, pick which products are eligible, and the platform auto-generates ads from product names, images, and attributes. Criteo is the first ad tech partner. The system handles up to a million SKUs per advertiser. And here's the detail that should stop you cold: it largely accepts the same product file brands already send Google.

June 2026. Product feed ads go live in the Ads Manager beta. OpenAI tells advertisers feed ads were among the best performing formats in the test.

August 2026. This week. Digiday verifies screenshots of product carousels in ChatGPT, stacked side-by-side ad units pulled straight from retailer feeds. If you've ever looked at a Google Shopping results page, you already know exactly what this looks like.

Free organic listings. Feed uploads. Auction layer. Carousel. That's the whole Google Shopping arc, compressed from ten years into six quarters.

Be clear about what the feed actually is.

Brands uploaded product catalogs to ChatGPT so it could answer shopping questions accurately. Correct prices, real availability, current specs. That was the pitch, and it was a good one. Bad data in the answer hurts you, so you fix the data.

Then OpenAI connected that exact catalog to the auction. The same master product file now generates your ads automatically. Your organic presence and your paid presence run on one pipe, and OpenAI owns the pipe.

This is not really a complaint, it’s a mechanism you need to understand, because it collapses two disciplines into one. There is no longer a meaningful line between "AEO work" and "paid answer work." The product data you maintain for organic citations IS your ad creative. The team that owns the feed owns both games. And in every marketing org I've been inside, nobody owns the feed. It's an ops afterthought sitting between ecommerce, paid media, and whoever set up Merchant Center in 2019.

When OpenAI killed Instant Checkout in March, a lot of smart people read it as retreat. OpenAI stumbling on commerce, conversational shopping failing again like Messenger bots in 2016 (throwback, huh?).

The numbers tell a different story. Daniel Danker, Walmart's EVP of AI acceleration, told Wired that conversion rates were three times lower for items sold directly inside the chatbot than for items requiring a click out to walmart.com. And The Information reported that OpenAI's own staff concluded users were researching products in ChatGPT but weren't buying there.

ChatGPT is bad at closing and exceptional at discovering. So OpenAI stopped fighting for the transaction and monetized the thing it's actually good at: the moment a buyer asks "what should I get" and trusts the answer.

They don't want the checkout. They want the auction on the answer. Killing Instant Checkout wasn't retreat, it was focus. Truist analysts estimate OpenAI does under $1B in ad revenue this year and over $30B by 2030. Nobody builds a $30B ads business on transaction fees.

If you sell software instead of sweaters and you've read this far thinking "retail problem," look at the ad density data.

OtterlyAI found that 76.4% of shopping questions in ChatGPT now return a sponsored placement. By vertical, finance and insurance sit at 85.9% ad density, the highest measured. Healthcare sits at 42.3%, the lowest.

This isn’t random, it’s a heat map of customer lifetime value. High-LTV categories with aggressive bidders got monetized first. Now ask yourself where B2B software sits on that map. A buyer asking "best CRM for a 12-person services firm that hates Salesforce" is one of the highest-intent, highest-LTV queries on the internet.

B2B doesn't have a feed spec yet. But you already have a feed, you just haven't been forced to structure it. Your pricing page, your feature matrix, your integrations list, your security docs. That's the B2B catalog waiting for a schema. My prediction, and I'll put a date on it: OpenAI starts accepting feeds for services and software within 12 months. When it happens, the sponsored layer arrives with it, and every playbook in this piece applies to your category overnight.

Here's why the organic work doesn't die when the auction shows up. It gets priced in.

First, a definition, because this is a metric I use with clients and you won't find it in any tool yet. Answer Capture Rate is the share of high-intent prompts in your category where the engine names you as the answer. Not ranked, not mentioned in passing, named as the answer. Take fifty prompts a real buyer would type, count how many times you're the recommendation, that's your ACR. Rankings measure whether you showed up on a page. ACR measures whether you won the conversation, and in a world where the conversation is the whole store, it's the only organic number that matters.

Now the parallel. Google's auction never let you buy your way past bad relevance. Quality Score meant a trusted, relevant advertiser paid less per click than a garbage one. The answer engine version of Quality Score is how much the engine already trusts your brand, and that trust is built by exactly the work AEO people have been doing: consistent brand facts, clean product data, citations across the sources these engines actually pull from.

So the sequence isn't "organic answers, then paid answers replace them." It's "organic trust becomes the multiplier on your paid efficiency." And there's already data proving these are separate games: Seer Interactive analyzed 3,697 ChatGPT ads and found the advertiser appeared in the organic answer above their own ad just 5.4% of the time, and got cited as a source only 3.3% of the time. The ad slot is currently where the answer-losers go. Buying the placement does not buy you the answer. Your ACR today is your auction advantage tomorrow. The brands treating AEO as a content hobby are going to bid cold into auctions where the engine already believes their competitor.

One honest caveat, because the Google parallel cuts both ways. Google eventually let free listings back into Shopping in 2020, after Amazon had eaten enough of the market that Google needed inventory more than it needed the toll. With Gemini, Perplexity, Copilot, and Claude all fighting for the same queries, the organic layer in AI answers probably never fully closes. But "probably never fully closes" is a hope, not a strategy.

Three things, in order. This is the same sequence I walk advisory clients through, and the order matters.

1. Assign feed ownership to a single person, this week. Not a committee, not "marketing ops broadly." One name. That person owns the single source of truth for what you sell, what it costs, and what makes it different, across every place that pulls it in. Their first task is a consistency audit: pull your product names, prices, and top three claims from your website, your Merchant Center file (or pricing page if you're B2B), your G2 or marketplace listings, and your LinkedIn page, and put them side by side. Every mismatch is a place where an answer engine has to guess which version of you is true. Fix the mismatches before you spend a dollar anywhere else. If you're retail, your ChatGPT catalog is now part of this audit. If you're B2B, start building the structured version of your pricing and feature data now, before a spec forces you to do it badly in a weekend.

2. Baseline your Answer Capture Rate before the sponsored layer eats your category. You cannot measure encroachment without a starting point, and this takes one afternoon. Build a list of 50 prompts in three buckets: 10 branded ("is [your product] good for X"), 20 category ("best [category] for [specific buyer situation]"), and 20 competitive ("[you] vs [competitor] for [use case]"). Write them the way a buyer actually talks, full sentences with context, not keywords. Run them in ChatGPT, Gemini, and Perplexity. Score each answer four ways: are you named, are you recommended, are you cited as a source, is a sponsored placement present. Log it in a spreadsheet with the date. Rerun monthly, same prompts. When sponsored placements arrive in your vertical, you'll know exactly what they took and exactly what your organic trust is worth in the auction.

3. Treat paid answer placements as a hedge, not a strategy. The auction is coming whether you bid or not, and it isn't cheap: OpenAI has reportedly priced early placements around $60 per 1,000 impressions, roughly three times Meta's average CPM. But the compounding asset is the trust layer underneath it. Every dollar of paid answer spend on top of weak brand trust is rented visibility at full price. Every dollar on top of strong organic citations is subsidized.

Don't take my word for how far along this is, and don't take OtterlyAI's either. Open ChatGPT on a free account, logged in, and run ten commercial prompts in your own category. Real buyer phrasing: "best [your category] for a [size] team that's frustrated with [incumbent]." Count how many responses come back with a sponsored placement underneath, and note whether the advertiser also appears in the organic answer above it.

Then reply to this email with your number and your category. I'll compile every response and publish the breakdown in an upcoming issue: sponsored rates by vertical, from actual operators running actual prompts, not a vendor study. If the heat map theory in this piece is right, B2B software categories should be lighting up in the next two quarters. Let's find out together, on the record, with a timestamp.

Google taught everyone this lesson in 2012 and it took a decade to land. OpenAI just taught it again in 18 months. The next platform won't take that long.

Structure your truth now (while it's still cheap).