Amazon Publisher Services is in active discussions with websites and external firms about tools to monetize ads inside AI chatbots. Per The Information this week, Amazon has already talked to at least two parties about its plans.
Amazon's Publisher Services division — APS — already runs header bidding and programmatic ad auctions for thousands of publisher websites. Transparent Ad Marketplace, Unified Ad Marketplace, a Prebid adapter that opened to beta in January. Mature, scaled infrastructure. What Amazon is exploring now is extending all of that into chatbot interfaces — powering the ad layer for AI properties Amazon doesn't own.
Think AdSense. When Google launched AdSense in 2003, it didn't just monetize Google search results. It monetized everyone else's websites too. One of the most durable revenue models in tech history. Amazon has been on the outside of that model for two decades, running a $68.6 billion ad business that flows mostly through its own surfaces — Sponsored Products, DSP, Prime Video. APS extending into chatbots would be the first real move toward an Amazon equivalent.
A lot is happening in chatbot ads right now, and it's worth seeing the full picture before zooming in on Amazon's play.
Criteo announced March 2nd that it's the first ad tech partner inside OpenAI's ChatGPT ad pilot — connecting 17,000 advertisers at a $60 CPM and a $200,000 minimum commitment. That CPM is comparable to NFL broadcast inventory. Early data from 500 Criteo retailers in February: users referred from LLM platforms convert at roughly 1.5x the rate of other referral channels. One month of data, grain of salt — but directionally it tracks with what I wrote in Instant Checkout: users arriving via AI interface are further along in the decision because the AI already did the consideration work.
Target is routing retail media spend from Roundel into ChatGPT as an off-site inventory slot, same as it would treat any display exchange. They're not testing whether chatbot ads work. They're treating them as an established channel.
And then there's Perplexity, which pulled the plug on advertising entirely last month. They'd been testing ads since 2024, stopped accepting new advertisers in October 2025, and the head of advertising sales left in August with no replacement. Their reasoning: "The challenge with ads is that a user would just start doubting everything."
I think Perplexity is going to regret that.
The "users won't tolerate ads" argument sounds good in a press release, but it doesn't hold up against history. Amazon had zero ads for years. Then they introduced them slowly, and sellers complained. Then they added more. And more. Amazon's ad business is now $68.6 billion and growing 22% year over year. Their marketplace is stronger than ever. Google went through the same arc — people grumbled about sponsored results and kept using Google because the product still worked.
Consumers say they hate ads. What they actually hate is not finding what they need. If the product still delivers, they adapt. There's a small slice of users who genuinely won't touch an ad-supported product on principle, and they're vocal about it online, but they're a rounding error in the actual usage numbers.
Perplexity didn't give it enough time, and they're also not Amazon. Amazon has the advertiser relationships, the purchase data for closed-loop attribution, and the infrastructure to make ads perform well enough that they're useful rather than just noisy. Perplexity had none of that. Walking away from ads when you don't have the machinery to make them work isn't a principled stand — it's a lack of options dressed up as philosophy.
OpenAI's approach is structural separation — ads below the fold, never influencing the response content. Amazon's answer inside Rufus is different: they call it Sponsored Prompts, AI-generated questions derived from product listing content. "Does this product provide thermal protection?" The brand didn't write that question. Amazon's AI did, based on the paid product's content. No visible ad label in the traditional sense. Less transparent than OpenAI's approach, arguably. But Amazon has twenty years of evidence that consumers adjust.
The Amazon infrastructure play makes more sense through what I've been calling Block, Take, Sell.
A few days ago I wrote about Amazon's Agent Policy update — the one that requires every AI tool operating on Amazon to self-identify in HTTP headers, accept shutdown on demand, and respond truthfully when asked if they're a bot. The policy is real, the enforcement risk is real, and it gives Amazon a kill switch over every external agent touching their marketplace.
At the same time, Amazon's Buy for Me agent scrapes 500,000+ third-party retailer sites — pulling product data from merchants who never agreed to it, with none of the self-identification requirements Amazon demands from competitors. And Amazon is actively suing Perplexity for doing, on Amazon listings, roughly what Buy for Me does to everyone else.
The chatbot ad infrastructure play extends that same logic.
Block: Competitors' AI agents cannot freely access Amazon's marketplace. 47+ crawlers are blocked in robots.txt. The BSA kill switch is now policy.
Take: Amazon's own agents scrape the open web at will, surfacing competitor products inside Amazon's interface.
Sell: Amazon wants to power the ad infrastructure inside the same chatbots it blocks from accessing its data.
So: your bot can't reach our marketplace, but we'll run the commerce recommendations inside your bot — for a fee.
The data angle is where this gets really interesting.
Google has intent data from search queries. Meta has interest data from social graphs. Amazon has transaction data from purchases. Nobody has all three. Whoever assembles that combination first holds it for a long time.
If Amazon powers ads inside external chatbots, those conversations become visible to Amazon. Not just click-and-buy data. Amazon would start learning what people ask about before they have purchase intent, how they compare products across brands, what they reject and why. Chatbot conversations are almost entirely consideration and research — the top-of-funnel that Amazon has historically been blind to.
That changes what Amazon knows in a fundamental way. Not incremental. Qualitative.
For sellers, this is early. But it's directional.
Amazon already tested the ad format inside Rufus — Sponsored Prompts, the AI-generated question format tied to product listing content. Brands see impressions, clicks, and purchase data per prompt in the Ad Console. They can't bid on individual placements or allocate dedicated budget; Amazon determines eligibility based on listing quality. Which tells you something: the brands with the most complete, well-structured product data are going to have better outcomes here. Same principle as organic search ranking, applied one layer up.
If APS extends into external chatbots, your Amazon ad campaigns could eventually reach consumers inside ChatGPT or other AI surfaces without building a separate integration for each one. Existing product data and campaign structure carry over. Useful reach expansion — assuming Amazon doesn't turn it into another margin extraction lever. Which, honestly, it might.
That 1.5x conversion rate from Criteo's early data is the number sellers will hear in every pitch for the next six months. Worth noting: $200,000 minimums make this a brand-budget play right now. When pricing normalizes — and it will — that changes.
The bigger pattern here is re-intermediation.
The popular version of the AI-in-commerce story is that agents remove middlemen. They research, decide, buy — skipping the friction layers in between. Partially true. But every major platform shift has replaced old intermediaries with new ones. The web killed travel agents and created Expedia. Mobile killed paper maps and created Google Maps. Search intermediated retail and then retail media monetized it back.
AI is going to create new intermediaries. Invisible ones. You don't think about Visa when you tap your card. You don't think about AWS when you load a website. If Amazon becomes the ad infrastructure for chatbots, most consumers will never know that the product recommendation from their AI assistant ran through Amazon's ad marketplace.
Some people will argue the trust layer can't survive ads. Perplexity made that bet when they walked away. I think they're wrong. The trust layer survives ads when the underlying product still works — when people still find what they need, still get good answers, still feel like the tool is useful. Amazon proved that. Google proved that. The pattern is consistent enough that I'd bet on it here too.
The companies that figure out how to run ads inside AI without degrading the product are going to build enormous businesses. The companies that sit out on principle are going to watch from the sideline while the economics get locked up without them.