Display advertising is broken. Not because the people building it were incompetent, but because the systems were optimized for the wrong metric from the start. We built an industry around clicks instead of outcomes. That mistake cascaded into everything else: tracking infrastructure that grew more invasive every year, format constraints that never adapted to the medium, and an adversarial relationship between users and advertisers that only deepened over time.
I've spent years inside display ad systems. I know exactly where they went wrong because I helped build some of those mistakes. The good news: AI conversations offer a fundamentally different opportunity.
Optimization for Clicks, Not Outcomes
Display advertising was designed to work on the web. Advertisers wanted their brand on websites people visited. The natural metric was whether users clicked the ad. So entire networks—Google AdSense, ad exchanges, bidding algorithms—optimized for clicks per thousand impressions.
That's where things went sideways. Clicks became the target, not actual business outcomes. This created perverse incentives. Designs that looked like page content generated more clicks. Sensationalist headlines worked better than accurate ones. Fake urgency outperformed honest value propositions. Publishers could earn more money by making ads more deceptive. Advertisers could spend more money reaching people who had no actual purchase intent.
The system rewarded clickbait. It rewarded dark patterns. It rewarded making people feel bad enough to click. And since scale matters in programmatic advertising, these tactics spread across the entire ecosystem. Every publisher trying to maximize earnings fell into the same trap. Every advertiser trying to hit click targets used the same tricks.
The result: users learned to ignore ads entirely. Banner blindness became a documented phenomenon. Ad blockers became mainstream. The click-through rates that once drove the economics of display advertising collapsed. We optimized ourselves into irrelevance.
The Tracking Arms Race
To serve more relevant ads and improve click rates, the industry needed data. Where did users go? What did they like? What were they considering? Cookies seemed like the answer.
Third-party cookies let ad networks follow users across websites. Your browsing history on a news site could be combined with your shopping history on an e-commerce platform, your health searches, your financial research. A complete profile. Then advertisers could target that profile with laser precision—or at least, they thought they could.
But users didn't agree to surveillance. They just accepted it as the cost of free content. So the industry had to hide it. Cookie disclosures became legally necessary but deliberately designed to be impossible to parse. Cookie banners appeared on every site with dark patterns guiding users toward acceptance. The tracking continued in the background whether users understood it or not.
As cookies became restricted, the industry invented fingerprinting. Device fingerprints from browser characteristics, IP addresses, and other signals could identify users without explicit cookies. Then came first-party data partnerships, where retailers sold shopping data to ad networks. Then came walled gardens where platforms like Facebook built their own tracking infrastructure because they owned both the content and the ad network.
Privacy regulations like GDPR and privacy changes from Apple forced the reckoning. Third-party cookies were dying. The ecosystem panicked. None of it mattered because tracking itself was the problem, not the specific mechanism. You can't fix surveillance by using different surveillance tools.
Format Mismatch: A Problem That Never Evolved
Display ads came from print magazines. An advertiser bought a physical space—say, a half-page in the lower right corner. That space was designed for specific dimensions. 300x250. 728x90. 160x600. These standards made sense in print. They made some sense on desktop websites where you could reserve physical space.
But the web isn't print. And phones aren't websites. The format that worked on a magazine page doesn't work on a mobile phone screen. A banner ad that takes up 25% of a phone screen isn't an ad—it's an interruption. It's not a suggestion—it's a barrier to the content someone actually came for.
Publishers knew this. So they worked around it. They made ads autoplay. They made ads sticky so they followed you as you scrolled. They put ads between content. They made ads float over content. They invented interstitials that appeared for seconds before you could dismiss them. Every workaround was an escalation of intrusiveness.
Meanwhile, the creative never evolved. It stayed stuck in the dimensions of a print magazine. A 300x250 "medium rectangle" on a phone screen looks small, jarring, and utterly disconnected from the content around it. After a decade on phones, display ads still looked like they belonged somewhere else.
What Changed with Conversational AI
Here's the fundamental difference: users in a conversation are literally telling you what they want. They're typing it out in natural language. The intent signal isn't hidden in cookies or inferred from behavior. It's explicit. It's there in the conversation.
When someone asks an AI assistant about hotels in Berlin, they've signaled intent. When they ask about travel insurance, they've signaled intent. When they ask whether a certain camera is good for underwater photography, they've signaled intent. The AI doesn't need to follow them across the web or build a profile or use inference. The intent is in the conversation.
This changes everything:
- No tracking required. You don't need cookies, fingerprints, or behavioral data. The conversation itself is the data.
- No privacy violation. You're using what the user explicitly told you right now, not aggregating a surveillance profile.
- Intent matching, not profile matching. You're matching an ad to intent expressed in language, not to a demographic category or inferred preference.
- Format fit by design. An ad in a conversation flows with the conversation. It's not squeezed into a print-era box. It can be a helpful suggestion—"Here are two hotels that match your criteria"—instead of an interruption.
Good Advertising Looks Like Help
This is the key insight: ads don't have to be intrusive or deceptive. They can be useful.
If someone asks an AI assistant "What's a good laptop for video editing?" and the system responds with a relevant suggestion, that's not spam. That's help. The user asked the question. The ad answered it. Whether it's labeled "sponsored" or not doesn't matter if it's actually relevant.
This only works if the ad is genuinely matched to intent. If you show hotel ads to someone asking about hotels, you're being helpful. If you show hotel ads to someone asking about flight prices, you're being intrusive. The system has to respect the context and only show ads where there's a real match.
When done right, the ad disappears into the conversation. It's not "an ad"—it's a suggestion. It's not an interruption—it's part of helping the user accomplish what they asked for.
The Builder's Advantage
If you're building an AI application, you can monetize it without degrading the user experience. Show ads only when they're relevant. Show no ad rather than showing a bad match. Your metrics shift from impressions and clicks to actual outcomes: did the user find this useful? Did they take action on it?
The incentive structure reverses. You're no longer competing to show more ads or more intrusive ads. You're competing to show ads that are so useful that users see them as part of the product. That's good for users, good for publishers, and good for advertisers.
The Opportunity
Display advertising built an entire industry around surveillance and clicks. Privacy regulations killed the surveillance. Click economics collapsed when ads became too intrusive. The medium evolved in ways that format standards couldn't support. The whole structure broke under its own contradictions.
Conversational advertising doesn't start with those constraints. It starts with intent. It starts with context. It starts with the opportunity to show ads that are actually useful. That's not a small difference. That's the difference between an industry built on deception and one built on utility.