Advertising Finally Has to Behave Like a Guest

Advertising is one of the oldest tricks in commerce, and also one of the most misunderstood crafts in product design. Every generation of technology thinks it invented the problem of "how do we make money without annoying people." It didn't. It just inherited it, with a chance to solve it better than the last one did.

I want to walk through where that problem has taken us so far, not as trivia, but because I think it's the only way to design ads for conversational AI responsibly. Skip that history and you'll either reinvent something that already failed, or reject something that already works.

A very short history of getting someone's attention

The first ads weren't digital. Egyptian merchants used papyrus to post sale notices, and walls in ancient Rome and Pompeii carried political slogans and gladiator announcements. The interaction model was simple: you walked past something, it told you what was available, and you decided whether to act. No targeting, but real relevance, because the ad and the product lived in the same physical space.

The printing press scaled that idea. Newspapers, magazines, and billboards let a brand reach thousands of people at once, but the relationship stayed one directional - you saw it, or you didn't. Leaflets pushed that to its cheapest extreme, easy to hand out, easier to ignore.

Then two things broke the model completely, in sequence. First the internet, which let an ad respond to a click and prove, for the first time, whether it actually worked. Then mobile, which put a screen in everyone's pocket and made big, loud placements impossible to sustain. Ads had to shrink, and to fit inside experiences people actually cared about such as a feed, a game, a map. That's the real throughline across three thousand years: every time the medium changed, advertising was forced to change with it, or it stopped working.

Who set the rules, and what they built

You can't talk about modern advertising without Google, Meta, and Amazon, they didn't just build ad products, they built the infrastructure that decided what "relevant" means. Google made intent the currency - you search, you're shown something related. Meta made behavior the currency - you like, share, and scroll, and a profile gets built from it. Amazon made purchase the currency - actual buying signals, not guesses. Together they trained an entire generation of users to expect ads that already know what they might want. That expectation is the baseline every new platform gets measured against, including conversational AI.

The format toolbox that grew out of that expectation is long by now: static banners, interactive units, playable ads you can try before downloading, skippable and non skippable video, rewarded video where the user opts into a trade, interstitials that interrupt a flow outright. Each one resolves the tension between attention and interruption differently, a playable lets the product prove itself first, rewarded video asks before it takes, an interstitial just takes. Knowing why a format exists matters more than knowing what it looks like.

The fine line between useful and intrusive

An ad can be useful. It can also be a bother. Sometimes it's both, to different users, in the exact same placement.

The difference usually comes down to timing, relevance, and control. An ad that interrupts you mid-task, has nothing to do with what you're doing, and can't be dismissed, will always feel intrusive, no matter how well designed it is. An ad that shows up at a natural pause, is relevant to something you already care about, and can be ignored or engaged with on your own terms, has a real shot at being welcome. Nobody in the industry has fully solved this, it's a constant negotiation.

Chat raises the stakes on that negotiation more than any other ad supported product, because it's a different category of intimate. A game is already a performance space. Users expect breaks and interruptions as part of the loop. A chat is closer to a private conversation, sometimes about work, sometimes personal, sometimes something the user wouldn't want anyone looking over their shoulder to see. That intimacy means the trust being spent isn't just "don't annoy me," it's closer to "don't make me feel exposed," and it's exactly the bar every ad experience in this space has to clear.

Understanding the rules before breaking any of them

Here's something I want to be clear about before going further. This isn't a call to throw out everything we know and build advertising from scratch. It's closer to the opposite. The goal is to deeply understand the formats users already recognize and the behaviors they already trust, and then ask where AI chat gives us room to do something genuinely new within that shared vocabulary.

Users already know what a banner is, what a video ad is, what a rewarded ad gives them. That familiarity is valuable, it lowers the cognitive cost of an ad appearing at all. The innovation isn't in inventing a new grammar, it's in finding new sentences within the grammar people already speak.

What this actually looks like right now

This isn't theoretical anymore, it's already happening, and it's worth looking at closely because the early attempts tell us a lot about what works.

Perplexity actually tried this, then walked away from it. The sponsored follow-up questions sat beside the answer, clearly labeled, with copy generated by Perplexity's own system, not the brand. Still, by February 2026 the company pulled the format entirely, saying even a labeled ad made users second guess the answer's objectivity, evidence of how thin the trust margin is inside a live conversation.

Amazon took the opposite bet. What started as Rufus, now folded into Alexa for Shopping, weaves sponsored product recommendations into a shopping conversation, framed as answers pulled from a brand's own listing content rather than written by the advertiser. It borrows the assistant's own credibility, which is exactly why the sponsored label has to stay unmistakable. Unlike Perplexity, this placement didn't get pulled, it moved into paid, general availability in 2026.

Both of these point at the same underlying idea. The ad doesn't have to break the conversation to exist inside it. It can be a parallel thread the user can pull or ignore, rather than a wall they have to click through.

We ran into a version of this ourselves. As designers, we're always hunting for opportunities, that's the job. One we found at Velocity was the waiting time during generation, the seconds where an LLM is producing an image or a longer response and the screen mostly sits idle. That idle moment is unused real estate, and it turns out to be exactly the right moment for a slightly bigger ad unit, because it fills a gap that was already there instead of competing for attention. That's the thinking behind our folding ads, the unit expands while the model generates, then collapses back down once the response is ready. The same window works for a gentle playable moment too. What makes this interesting isn't the format, it's that it solves the user's problem first, and the ad revenue is a side effect rather than the starting point. An ad justified by a business need always feels like a tax. One justified by an experience need can actually feel like a gift.

Another one we built leans on a different lever entirely - curiosity instead of idle time. If someone asks about shoes, instead of showing a static banner for a shoe brand, we surface a short poll of a brand's question with three answer options. The click itself is the engagement. You're not being asked to buy anything yet, you're being asked a quick opinion. Answer it, and we reveal how everyone else answered, which is its own small payoff, then the CTA shows up after that moment of curiosity has already been earned. It's a small sequence, question, participation, reveal, offer, but it changes what the user is doing from the very first interaction. They're not dismissing an ad, they're taking part in something, and the ad only asks for the sale once that's already happened.

None of this works without the craft holding it together underneath, and that craft is just typography, color, and hierarchy applied under a stricter rule than usual. The "sponsored" label only earns trust if it's legible at a glance and never trying to blend in. That's a type weight and contrast decision as much as an ethical one. The folding ad feels calm instead of jarring because its expand and collapse use the same easing curve as everything else in the chat, so it reads as part of the interface, not an animation bolted on top of it. One rule overrides the others here: the ad is never allowed to look more important than the conversation it's sitting inside.

Designing the voice, not just the unit

There's a second shift happening that's just as significant as where ads appear, and it's about what they say. For most of advertising history, the creative was fixed. A brand designed a banner or a piece of copy and that asset ran as is - the same for everyone who saw it. In AI chat, the creative itself can be generated in real time, shaped by what the user is actually asking about in that exact moment, which means someone has to decide its tone of voice, its register, the specific words it reaches for, and that someone is increasingly a system, not a copywriter. Get it wrong and even a well placed ad feels off, too salesy for a technical question, slightly out of step with how the AI has been talking to the user up to that point. Get it right, and it reads as a continuation of the conversation instead of a foreign voice interrupting it.

That makes the job less about designing a unit and more about designing the voice guidelines and guardrails a generative system uses to write dozens of variations of that unit, consistently, without ever sounding manipulative or out of place. That's closer to building a style guide for an AI than laying out a static banner, and I think it's one of the most underrated design challenges this whole shift is creating.

Designing within real constraints

None of this happens in a vacuum. Every idea has to survive contact with actual development limitations. Android, iOS, and Web all render differently, latency budgets are tight in a live chat, and anything that breaks the model's response flow breaks trust immediately.

Good ad design in this space isn't about the most creative concept, it's about the most creative concept that survives production. I'd go further than that and say that the constraints aren't just something an idea has to pass, they're often where the idea comes from in the first place. Neither the folding ad nor the poll unit started as a concept looking for a home, they started as a limitation, a dead stretch of loading time, a static banner nobody wanted to look at, and the idea was the answer to that specific limit. An unlimited canvas rarely produces anything sharp, it just produces more of the same, bigger. I trust an idea more once I know what constraint it grew out of.

The designer's role: protecting the user

This is where the friction between product and design usually shows up, and I think it's a healthy friction, not a problem to eliminate. Product is optimizing for revenue and growth, and that's their job. Design's job, at least the way I see it, is to be the person in the room asking whether a decision will erode the exact trust that makes the product valuable in the first place.

That shows up in small decisions too, like how small an "Ad" badge could go before it stopped doing its job. A smaller badge looks cleaner, but it also gets easier to miss, and a badge nobody notices isn't really labeling anything anymore. Design held the line there, because the cost of a few extra pixels is nothing next to the cost of an ad that quietly stops announcing itself.

That doesn't mean design blocks monetization. It means design pushes for monetization that survives contact with real users over time, not just in a test cohort. Someone in the room needs to be thinking past the next sprint, and that's usually us.

Where This Leaves Us

Advertising has always adapted to its medium, from carved stone to banner ads to sponsored prompts inside an AI assistant. AI chat is the next medium, and it comes with its own intimacy and its own constraints. The history isn't a limit on what we can build, it's the map of what's already been tried, what worked, and why.

The work now is to take that map seriously, respect the trust chat interfaces are built on, and design something that feels like it belongs in the conversation, not something bolted onto it.

We're building these formats at Velocity, for teams designing ad experiences inside conversational AI. Get in touch.