Perplexity is putting ads in AI answers — what it means

This week, the answer engine Perplexity began running ads inside its AI-generated answers. The format is “sponsored follow-up questions,” the rollout is a US test, and the launch partners include names like Indeed and Whole Foods. It’s small, it’s experimental, and it is one of the most important things to happen to advertising this year.

Not because the format is novel. Because of who is doing it, and what it confirms.

Perplexity ads, in one paragraph

On November 12, Perplexity announced it would start experimenting with ads on its platform, beginning that week with a controlled rollout in the United States. The ad unit is a sponsored follow-up question — a paid suggestion that appears alongside the “related questions” Perplexity already surfaces under an answer, plus some sponsored media positioned to the side. Crucially, the company says the answer to a sponsored question is still generated by Perplexity’s own model, not written or edited by the brand. The initial cohort of advertisers and agencies includes Indeed, Whole Foods Market, Universal McCann, and PMG. Perplexity is selling the format on a CPM basis.

That’s the whole launch. No display network, no banners, no retargeting pixels following you across the web. An ad that lives inside the conversation, as a question you might actually want to ask next.

Why this matters more than the format

For the past year, the open question hanging over AI products has been: how does any of this get paid for? Inference is expensive. Subscriptions alone don’t cover it — Perplexity said as much, framing ads as the way to fund a sustainable revenue-sharing program with publishers. Every team building an AI chat product has been circling the same question, and most have been waiting for someone credible to move first.

Perplexity just moved first. An established answer engine, with real traffic and real brand advertisers, put paid placements directly inside model-generated responses. That single decision does more to validate the category than a hundred think pieces about the future of advertising. As we argued earlier, the page is being compressed into a paragraph, and the ad has to live inside that paragraph or not at all. Perplexity just shipped the first mainstream proof.

The interesting part isn’t that ads showed up in AI answers. It was always going to happen. The interesting part is the shape Perplexity chose — and how close it lands to what a good ad in an answer should be.

The publisher revenue-share is the real story

Buried under the ad-format headlines is the part publishers should care about most: Perplexity is positioning advertising as the engine for sharing revenue with the publishers it cites.

This is the through-line from the Publishers’ Program the company launched earlier this year. The logic is straightforward. When a publisher’s content is referenced in the answer that surrounds a paid placement, that publisher gets a cut of the revenue. Advertising is what makes the pool big enough to share.

That’s a meaningfully different deal than the one the open web offered. In the old model, an answer engine that summarized your article was a threat — it satisfied the user’s intent without ever sending them to your page, and your ad inventory went unsold. In this model, being the cited source is itself monetizable. The economics of getting summarized flip from pure loss to partial recovery.

It is early, and the percentages and mechanics will be argued over for years. But the principle is the one that has to hold for AI answers and the open web to coexist: if the model uses your work to earn, you share in what it earns.

What good ads in answers should look like

Perplexity’s launch is a useful moment to be precise about the standard. An ad inside an AI answer is a fundamentally more intimate placement than a banner. The user asked a question and is reading a trusted response. Abuse that trust and the whole surface dies. A few principles that the early format gets mostly right, and that the category should hold itself to:

• Relevant in the moment. The ad has to match what the user is actually trying to do, not what they happen to be reading. A sponsored follow-up question only works if it’s a question the user might genuinely want answered next. The prompt is the targeting signal, and it’s a far better one than any cookie.
• Clearly labeled. “Sponsored” has to mean sponsored, unambiguously, every time. The moment users can’t tell the paid suggestion from the organic one, the answer engine stops being an answer engine.
• Ranked honestly. A paid placement can sit next to organic options, but it can’t quietly corrupt them. Perplexity’s claim that the model — not the advertiser — still writes the answer to a sponsored question is the right instinct. The advertiser buys the slot, not the truth.
• Useful, not interruptive. The best version of this is an ad the user is glad to see because it advances what they were already doing. The worst version is a mid-answer interruption that breaks the thread. Conversation is a hostile environment for anything that feels like a pop-up.

These aren’t lofty ideals. They’re survival conditions. The reason banner ads decayed into something users install software to block is that nobody enforced them. The teams building advertising into AI chat get to set the norms now, before the bad patterns calcify. (If the vocabulary here — sponsored follow-up, in-thread placement, honest ranking — feels new, we keep a running glossary of conversational advertising terms.)

Where this goes

We’re a small team building an adserver for AI chat applications — we started Elo this year because we think in-thread, conversational advertising is the model that will fund the next era of AI products, and it needs real infrastructure underneath it: bidding, creative matching, and labeling that publishers and users can trust.

Perplexity’s launch is the clearest signal yet that the category is real. One answer engine, one US test, four launch partners, a CPM, and a promise to share the proceeds with publishers. Modest on paper. But it draws the line in the right place: the ad belongs inside the answer, it has to earn its spot in the moment, and the people whose work makes the answer good should get paid.

That’s the bar. The rest is execution.