65% of AI shoppers click the name they already know. Now that ad talks back.

The findingThe ad stopped being an asset.

On September 16 OpenAI began testing Sponsored Agents, and the mechanic is worth stating precisely. Click the call to action and you don't get a landing page. You get a labelled, walled-off conversation with the advertiser's own AI agent, running inside ChatGPT. Wayfair has confirmed it's in the pilot, alongside Best Buy, Lowe's and Angi.

That is a smaller change than it sounds and a considerably larger one than it looks. Every signature ad format before this was a fixed asset that somebody approved in a deck. Google had the text link, Facebook had the feed unit, TikTok had the in-feed video. This one is a live representative, answering questions nobody wrote down in advance.

Speaking at Goldman Sachs on September 8, OpenAI CFO Sarah Friar framed the current card format as a placeholder rather than a destination: "This is all before we've really launched a format that feels truly endemic to AI." A roadmap, not a hedge.

The interpretationRecognition decides the click. The conversation decides the rest.

The more useful way to read that news is beside the other finding published the same week. Bazaarvoice and retail media analyst Kiri Masters went looking for what shoppers actually do when an AI tool puts several options in front of them. Sixty-five percent click the brand they already know. The labelled best match takes 24 percent, and price takes 11.

So the media buys the conversation. Your name decides whether it starts.

For a category team that lands on two levels at once. At brand level, your best-known SKU is about to be represented by an agent that fields questions on ingredients, sizes and substitutions. It builds those answers from whatever happens to be published about you. At category level, a format that everyone assumed would flatten brand equity turns out to reward precisely the equity you were told was eroding.

You can approve an ad. You cannot approve an answer. That gap is why we built Echo, our machine reputation intelligence product. — Ringer Sciences POV, 2026

OpenAI has already run the other version of this experiment and published what it learned, which is more than most platforms do. Instant Checkout opened in September 2025 and closed in March 2026, after Walmart data showed in-chat conversion running three times below click-out. Catalogue-to-checkout became catalogue-to-ad. The platform has been candid about the limits of buying inside the chat, and brands would do well to be as candid about the limits of talking inside it.

The strategic implicationTwo things do not come back to you.

The announcement is silent on what an advertiser receives from the transcript, and equally silent on what the reportable outcome of a conversation is supposed to be. An impression fits a column and a click fits a column. Turns don't.

That gap should feel familiar, because this category has lived through it before under a different name. It was called retail media, and the lesson ran the same way: when the platform owns the exchange, you need an independent read of it. Only the clock has changed.

What to doFour moves, and where our team picks them up.

1. Audit the unpaid answer first. That's Echo. Before you fund a format that lets a machine speak for you, establish what it says today with no media behind it. Echo runs that sweep at scale, and our analysts then score the results SKU by SKU, so you can see which products get described accurately and which quietly get substituted away.
2. Turn Wayfair's guardrails into a scored checklist. That's Readiness. Their paid media director described testing "at limited scale with guardrails around product accuracy, transparency and service handoffs," which is exactly the right instinct. Our Launch Readiness Index turns that instinct into something you can hand to legal and to your retail partners: what's present, what's missing, and what to fix before the format goes live.
3. Define the outcome before the platform does. That's Signal. Decide now what a good conversation looks like, whether you measure it in turns, handoffs or carts, and instrument it while the standard is still unwritten. Signal maps the journey once it leaves your own channels, which is precisely where this format now sits.
4. Fund the source material. That's GEO Strategy and Spark. Agents answer from reviews, editorial and creator coverage, and 94 percent of shoppers go and verify a recommendation elsewhere regardless (Bazaarvoice, 2026). GEO Strategy shifts what the machines say about you, and Spark produces the on-brand content that gives them something to say it from. The two work as a pair.

Here is the reassuring part, and it is worth holding onto while the format settles. Recognition still decides the click, and it remains the one asset no platform can reprice on you. Our analysts can put the current machine read of your top SKUs in front of you inside a week, which is a better place to begin than a media plan. Start there, then decide whether you want a machine speaking on your behalf.