Amazon's AI Chatbot Has 250 Million Users and Delivered One Brand 88 Ad Clicks
Amazon's internal pitch deck for Rufus advertising cites 250 million active users and claims chatbot shoppers are 60% more likely to complete a purchase. Consumer goods brand Paladone, one of the early testers, saw 88 ad clicks from Rufus in Q1. Their regular Amazon advertising delivered 500,000 over the same period.
Both numbers are real, and somehow both numbers are the point.
The chatbot ad market is a genuinely strange place right now. Four major platforms sell ads inside AI conversations, each using a completely different pricing model, and the gap between what the pitch decks promise and what the dashboards show is unlike anything in the digital ad industry's recent memory. Emarketer puts the total AI-influenced commerce market at $20.57 billion in 2026, roughly 1.5% of U.S. retail ecommerce and nearly quadruple the 2025 figure. The money is moving. The measurement hasn't caught up.
88 clicks, 250 million users, and a pitch deck doing heavy lifting
Amazon launched Rufus sponsored prompts to general availability in the U.S. on March 25. The format works differently from standard sponsored products: brands create keyword-triggered questions that start a conversation with the chatbot, and the ad shows up as a suggested prompt rather than a product listing. According to a leaked pitch deck obtained by Adweek, Rufus has 250 million active users and shoppers who interact with it are 60% more likely to complete a purchase than non-Rufus shoppers.
The pitch sounds great on paper. And sometimes it is.
But the early click data tells a different story. Paladone, a consumer goods brand running ads in Rufus, reported 88 clicks from Rufus ads in Q1, compared to 500,000 clicks from the rest of their Amazon advertising. As AdExchanger reported, other advertisers have described similar ratios. Sponsored prompts currently make up less than 1% of clicks across sponsored product campaigns. The format is live, but the volume is still closer to a beta test than a media channel.
The saving grace, honestly, is the pricing. Rufus CPCs are running at less than half the cost of standard Amazon ads. So brands aren't paying much for those 88 clicks. The question is whether you're buying clicks or buying time in a channel that will matter in six months. I think most brands buying Rufus ads right now are betting on the latter, whether they'd say it that way or not.
Four platforms, four completely different advertising models
There is no "chatbot ads" playbook right now because every platform has built something fundamentally different.
Amazon Rufus sells sponsored prompts on a cost-per-click basis. The ad triggers a conversation. Advertisers pay when someone clicks, but clicks are scarce because most Rufus users are mid-task and don't want a detour from their product search. It's closer to search intent than display, but the volume makes display look like a fire hose by comparison.
OpenAI's ChatGPT launched ads in February 2026 with impression-based CPM pricing at roughly $60 per thousand impressions, a $200,000 minimum buy, and clearly labeled ad units that sit below the organic answer. The ads don't influence the response, according to OpenAI. This is the premium play: you're buying presence next to an answer that 100+ million monthly users are reading. But the format is so deliberately separated from the conversational content that click-through behavior looks like nothing you've seen in Meta or Google. You're paying for brand proximity to a trusted answer, not for direct response clicks.
Google AI Mode has 75 million daily active users and ads now appear in 25.5% of AI Overview results, a 394% increase year over year. But 93% of AI Mode queries end with zero external clicks. Google says they're monetizing at the same rate as traditional search. Those two facts shouldn't coexist, and yet. What it means in practice is that Google is selling attention that doesn't translate into site visits in any way your analytics can currently track.
Perplexity went the opposite direction entirely. After testing sponsored follow-up questions through 2024 and 2025, they pulled all advertising in February 2026, citing trust concerns. Their argument: if users believe answers might be influenced by ad dollars, the product stops working. They're betting on subscriptions instead, targeting $500 million in annualized subscription revenue. One AI search engine believes ads are the business model. Another believes ads are the thing that kills the business model.
So you've got CPC, CPM at $60, contextual search ads with a 93% zero-click rate, and a platform that decided ads are fundamentally incompatible with the format.
Good luck writing a unified strategy brief.
The attribution gap is the part that should worry you
The click and impression data is almost beside the point. The bigger problem is that none of these platforms share enough data for advertisers to connect chatbot interactions to actual purchases through their existing measurement stack.
Albertsons ran ChatGPT ads and couldn't see their own performance data. They kept spending anyway. They see it as a calculated bet: being early in an unmeasurable channel beats being late in a measurable one where the CPMs have already tripled.
Shoppers arriving from AI platforms are reportedly 30 times more likely to purchase than average traffic. I'd treat that number with some healthy skepticism. What it probably reflects is that chatbot shoppers are already deep in the purchase funnel when they start a conversation. They've done their research. They're comparing two specific products. The AI catches the tail end of the decision rather than creating new demand. That's genuinely valuable, but calling it a "30x improvement" is like saying people who walk into a car dealership are more likely to buy a car than people who drive past it on the highway.
Meanwhile, Walmart is testing sponsored placements inside its Sparky chatbot in what eMarketer described as a "strategic shift toward high-margin ad revenue." And an estimated 73% of consumers are already using AI tools somewhere in their shopping journey, according to industry surveys from earlier this year. The traffic is moving into these channels whether advertisers have figured out how to measure it or not. That, to me, is the part that's actually urgent. Not the ad formats. Not the CPMs. Whether your products are even visible to an AI that's helping someone decide what to buy.
The $500 research project your competitors aren't running
If your brand sells on Amazon, Rufus sponsored prompts are the lowest-risk entry point right now. CPCs are running below $0.50 in most categories. Turn them on for your top 5 ASINs, run for 30 days, and compare the assisted conversion path against your regular sponsored product campaigns. Don't look at click volume; it will disappoint you. Look at conversion rate per click and cost per acquisition. At current pricing, the data alone is worth more than what you're paying for it.
If you're not on Amazon but you sell products online, the bigger play is structured data hygiene. AI shopping agents read product feeds, not landing pages. Clear specs, real-time pricing, and parseable customer reviews are what Rufus, ChatGPT, and Google's Shopping Graph actually index when they recommend products. Your product page copy matters less than your data feed accuracy right now, and fixing that costs nothing but time.
And if you're a brand marketer watching all this from the sidelines, the one thing I'd do this week is check whether your products show up when you ask ChatGPT, Perplexity, and Google AI Mode "what's the best [your category] for [your use case]." If you're not in those answers, you have a visibility problem that no amount of ad spend on any of these platforms will fix by itself.
The brands treating chatbot advertising as a scaled performance channel are going to be disappointed for at least another year. The brands treating it as a cheap research project on where commerce is heading are going to learn things their competitors won't have access to until the pricing catches up.
I'd probably start with the cheap version.
