Discover more from Steven’s Substack
Ads don't have to suck
AI will remake ads as features, not distractions.
Good ads solve problems. If I Google "toilet broken plumber" I want ads for local plumbers.
Most ads suck. Popups promising to "LOSE WEIGHT WITH ONE WEIRD TRICK" prey on our attention because one fool in 10,000 clicks them by accident.
In between, we have targeted ads thanks to FAANG's efforts to track us across the internet. Keep annoying me with Walmart ads for the coffee maker I left in my cart, and I might relent and buy it.
Done well (Amazon's recommendation engine) we don't feel like we're being advertised to. In a perfect world, we'd only see ads we were guaranteed to find useful.
Playing Jenga with user experience
Digital ads are a tax on our attention (worth $600B in 2022). Every website that runs ads decides its tax rate in the form of ad coverage, i.e. “how much of the page can we cover with ads before our users get so annoyed they leave.”
Ad monetization is like playing Jenga with gold blocks. If publishers can extract more value without the experience collapsing, they will.
Google’s not good enough
Following ChatGPT, a new kind of content publisher appeared: generative AI services. Chatbots, search engines, AI travel agents – these are still largely the domain of young startups and indie hackers. Currently, AI developers care more about delivering the best possible user experience new AI tech can deliver – GPU costs be damned. User acquisition is everything; monetization is an afterthought.
Talking to AI founders, they're skeptical of ads:
"We want to protect our users' privacy."
"We don't want trashy banner ads annoying users."
"We don't want to undermine the credibility of our results with paid content."
Unfortunately, the GPUs powering the AI revolution are not cheap. OpenAI burns $3 million/day on ChatGPT. Hacker projects that get any kind of exposure have cost their creators $100s.
Paywalls are an option, but that limits growth and restricts AI to only those who can pay.1
Consumer AI builders need ads, but don’t want to stalk and annoy their users.
A chance to do better
Current ad solutions like Google Adsense rely on tracking users or scraping static content to determine which ads to show. This breaks when AI content is generated in real-time.
But as we’re finding with many problems in AI, this can be solved with AI.
AI can deliver ads that not only target users' intent in prompts (much like Google search ads), but also adapt themselves to be contextually relevant to the users' prompts.
The three laws of AI advertising
AI publishers care about user satisfaction over revenue, and their users are providing extremely nuanced intent signals via AI prompts. It’s an opportunity to make AI advertising better for everyone – if we follow some guidelines:
Good ads solve problems. Respect the publisher app’s UX.
Good ads target what users want, not who they are. Respect users’ privacy.
Given a choice between showing a low-relevance ad versus no ad, show no ad. Respect users’ attention.
Ads that don’t suck
These are some of the ad formats that OpenAds is developing:
Advertiser and ecommerce AIs collaborate to generate recommendations specific to the user's problem.
Dynamic creative content speaks specifically to the user’s context
Conversational ads in chatbots add to the conversation
Chatbots embedded inside ads. Talk to a brand before even visiting their site.
BYO OpenAI API key is a low-friction, pay-as-you-go workaround for the tech-savvy, but it still means having to pay per use.