For years, Google has been diluting the core value of search in favor of profit. Paid rankings and sponsored content now dominate the results, forcing users to dig through layers of ads just to find real, useful information.
Then, ChatGPT came along and exposed the stark contrast: no scrolling past ads, no clickbait-filled pages masquerading as genuine content. Instead, users got clear answers, relevant follow-ups, and a truly useful search experience – like having “the world at your fingertips.”
But the ad-free experience is temporary. Soon, LLMs like Gemini, Perplexity, and ChatGPT will integrate ads. The difference? These ads promise to be smarter, more relevant, and far less intrusive. In this newsletter, we will look into the first LLM-based ads with insights from Andrej Kiska , founder of Kontext and my friend for over a decade.
Truly Contextual Ads Powered by AI
Our kids convinced us to get a rabbit. Filip, as we call him, sometimes has digestion issues, especially after eating fresh grass. Now, imagine I ask ChatGPT for advice, and it suggests feeding him pellets and hay for a few days.
Imagine now that truly contextual and personalized ads come in place. In real time, an ad platform processes the session, fully understanding my concern. It doesn’t just register “rabbit”—it recognizes the problem and context. Based on this, I get a relevant ad for an automatic pellet feeder that ensures Filip gets the right amount of food at the right time.
The result? A native, helpful ad that blends seamlessly into the conversation. No intrusive banners—just an offer that solves my actual problem (strange-looking rabbit poop all over the house).
From One-To-Many to One-To-One Communication
Mass advertising began with the printing press in the 17th century, evolving into radio, television, and digital ads that speak to the masses.
But now, GenAI apps generate over 1 billion user impressions daily, growing 20% month-over-month. We’re shifting back to one-on-one selling, like a medieval marketplace where sellers tailor their pitch to each customer.
Each ad can now be generated in real time, uniquely for each user, based on their conversation. That’s a fundamental shift.
LLMs fall into three main categories:
- General Chatbots (e.g., ChatGPT): Broadly accessible conversational tools, numbering in the dozens.
- AI Search Engines (e.g., Perplexity): A handful of focused tools delivering search-like results.
- Companion Apps (e.g., Character.AI): Ranging from shopping assistants to virtual fitness coaches, there are hundreds of niche implementations.
Right now, none of these have ad formats, but relevant ads could enhance shopping assistants, AI fitness coaches, trip planners, and virtual assistants.
How LLM Ads Work
Andrej Kiska—an early investor in Deepnote, Assetario, and part of the fund that backed UiPath (NYSE: PATH)—is now leading the way in AI-native advertising with Kontext which shows great results for partners like Masterclass, Amazon, and CD Projekt. “We refine each model individually for publishers, delivering 25-50% better ROI than Meta ads. CPMs are as low as $2-4,” explains Kiska.
The key? Perfect ad matching.
• The query and product must align perfectly.
• The wording of the ad must blend into the conversation naturally.
• The LLM behind the ad network creates the response dynamically, ensuring it fits the context and tone of the discussion.
Another advantage? Instant deployment. “The beauty of text-based ads is that you don’t need anything beyond product information,” says Kiska. “Defining a target audience helps but isn’t even necessary.”
The Future of AI-Native Visuals Tailored to You
While text ads are a strong starting point, the next frontier lies in multimodal ad formats.
With most pictures and videos becoming synthetic in the future, we can expect advertising to become much more interesting than it is today. For example, if you like videos about skis, then Atomic, Rossignol and Head can bid on which skis appear in the video.
Personalization Without Privacy Trade-Offs
The LLM does not know a user’s age or location unless explicitly shared. Gender is known in companion apps (but not in search/general chatbots), and publishers – who serve the impression – provide an IP address based on which the ad network determines the location of the user.
Yet, thanks to the context of the conversation, users receive valuable recommendations without compromising trust.
For Now, Enjoy the Ad-Free Experience—but Embrace What’s Coming
Ads are evolving. Instead of being an annoyance, they will become genuinely helpful recommendations.
We’ve gotten used to clear, straightforward answers from LLMs. There’s no reason commercial messages can’t follow the same principle.
Subscribe to our monthly AI Newsletter for more content like this!
