Miku's Brand Agent Inside a Publisher Chatbot Converted 50% of Engagers
Dappier launched Sponsored Conversations on April 27, 2026: a programmatic ad slot that places a brand's own AI agent inside a publisher's chatbot. Lilly Broadcasting is the first publisher running it, and Miku, the baby monitor brand, told AdExchanger that roughly half of users who engage with its agent end up clicking the call-to-action button.
That CTR is the real story. A 50% click rate on a sponsored placement would be 30 to 60 times what most native advertisers see, and even allowing for a tiny test base, it raises a question the IAB hasn't drafted spec for yet: when a brand-agent and a publisher-agent exchange messages inside one chat session, who actually owns the conversation log?
What a Sponsored Conversation actually is
The format works in two layers. Lilly Broadcasting drops contextual prompts at the bottom of articles. A reader clicks one, and that opens a conversation with Lilly's own agent, trained on the publisher's site content. While that conversation is running, sponsored prompts may surface. Click a sponsored prompt and a separate conversation opens with the brand's agent.
The slot is auctioned. According to Dappier, the brand who wins the programmatic auction for the article's banner unit also wins the right to seed that prompt. So the same display impression you used to buy now buys a chat opener too, which is a much bigger commitment from the user than a banner click.
The brand agent is trained on whatever the advertiser hands over. In Miku's case, that was a website URL, an FAQ page, and a list of talking points (the no-wearable design, the age range, that kind of thing). Miku founder Eric White told AdExchanger: "In a short video or banner ad, we can't convey what it means to be a contact-free monitor."
That's a fair point. Long-funnel categories with high explanation cost are where this format makes obvious sense. Smart ring, hearing aid, robot mower, mattress, custom orthotics. Anything where the buyer was going to Google a comparison anyway.
The two numbers worth circling
Lilly Broadcasting's chatbot went live in March. Roughly 5% of readers engage with it. That's the floor of the funnel.
Miku says about half of users who start a conversation with the brand agent end up clicking the CTA. That's the conversion step.
Stack them: 5% times 50% gets you 2.5% of Lilly's pageviews ending in a sponsored brand-agent click. For a programmatic comparison, average display banner CTR sits in the 0.05% to 0.1% range, depending on which industry benchmark you trust. Even if the agent click is loaded with browsing intent and not buying intent, that's a 25 to 50x lift on the same impression.
Two caveats that I think matter more than the headline number. The Lilly engagement figure is from an early sample where the chatbot was still a novelty. From what I've seen with conversational ad formats (Reddit's reminder ads, the early Perplexity placements, ChatGPT's first sponsored prompts), the eye-popping early numbers regress hard once the format stops being new. I'd expect that 5% to settle to 1-2% over a quarter.
And Miku's been testing for a few weeks against what is almost certainly a small-base sample. A 50% click rate on, say, 200 conversations is a different statistical creature than a 50% click rate on 20,000. The press release didn't share the denominator, which is the part I'd want to see before pricing this against my Meta CPM.
Even the conservative version, though, is several multiples of programmatic display. That's enough to take seriously.
Why this only exists because Walmart and Amazon already won
Walmart launched Sparky in June 2025, then opened it up to sponsored ads in late 2025. Amazon's Rufus has been live longer. Both let advertisers buy placement inside the retailer's own AI assistant, and both built that capability without asking a single publisher for permission.
Dappier's pitch to publishers is roughly: if you don't build this, the only place agents will deliver shoppers is inside Walmart and Amazon. The framing they have been pushing for a while is that agents collapse the ad funnel and publishers either get a seat or get cut out entirely. Lilly Broadcasting's traffic has been falling double digits, which fits the broader pattern of AI crawlers eating publisher visits. The chatbot is a reactive defense: keep readers on-site by giving them the ChatGPT-style answer experience without them having to leave.
Sponsored Conversations is the monetization layer on top. Without it, the chatbot is a cost center. With it, the publisher takes a cut every time a brand wants to talk to a reader who already raised their hand by opening a chat.
I'd argue this is the first ad format that actually mirrors what AI agents are going to do anyway. If we're heading into a world where my buying agent talks to a brand's selling agent, then publishers are an obvious neutral ground for that conversation, the way newspapers used to be the neutral ground for classifieds. There's a real strategic case here, not just a feature release.
The data question nobody answered
Here is what's missing from the announcement. When a brand agent runs inside a publisher chatbot, three parties touch the same conversation: the user, the publisher, and the advertiser. Dappier's chief business officer Mark Balabanian told AdExchanger that brand agents "only pull from a publisher's own site content unless the site owner opts to augment with third-party content." That covers retrieval. It doesn't cover the inverse: what data flows back to the brand from the chat?
This matters because the chat log is qualitatively different from a click. A click tells a brand "user clicked." A chat tells a brand "user asked these specific questions, in this specific order, and pushed back on this specific objection." That's CRM-grade signal. Whoever holds that log has a real asset.
The press release doesn't say. It doesn't say whether the publisher gets the log, the brand gets the log, both, neither, or whether some Dappier middleware retains it for cross-campaign benchmarking. Nor does it say what happens under GDPR or CPRA when a user thought they were chatting with the news site but were actually chatting with an advertiser-trained model.
If you're a brand evaluating this slot, the contract clause to read first isn't the price. It's data return rights. If the publisher keeps the entire conversation log, you're paying for a click. If you keep the log, you're paying for first-party intent data and the click is incidental. Those are very different products at very different prices.
Pitching this slot to your CMO this week
Two scenarios where this format is worth a real test budget right now.
One: your category has high explanation cost. If your average buyer Googles competitor comparisons before purchase, a 5-minute chat with a trained agent is cheaper than a 5-minute YouTube ad and converts better, in theory. Test budget: $15-25k against a single category-relevant publisher running Dappier, with a 30-day window and a clean read on chat-log access in the contract.
Two: you already buy display inventory on a Dappier publisher. The slot is incremental, not net-new, so you can fund a test out of an existing line item. The downside risk is small. The upside is a CTR ceiling you can't get from any banner.
What I would skip: any test where the contract doesn't grant you the conversation data. If you can't read the log, the format collapses back to a glorified display unit at a premium price.
The real action this week, before any media buy, is the audit. List your top three competitors. Find one that's likely to test this format first. If they get there ahead of you and lock in the explainer slot for your category on a publisher your buyers actually read, the make-good is going to be expensive and slow.
One thing worth tracking through Q3: whether OpenAI or Perplexity build a competing native ad unit inside their own chats. If they do, Dappier's window shortens fast. From what I've seen, the publisher rails-of-trust play tends to lose to the platform-of-distribution play, but the platform version usually takes 18 months to ship. That's roughly the testing window publishers have to prove this format converts before the bigger AI shops absorb the demand themselves.
I don't think the brands who win this format will be the ones with the biggest budgets. It will probably just be the ones who read their data return clause carefully, and the ones who picked a category where a 5-minute chat actually answers a real question.
Notice Me Senpai Editorial
