๐ฐ ๐๐ฒ๐ป๐๐ ๐๐ฑ๐๐ฒ๐ฟ๐๐ถ๐๐ถ๐ป๐ด ๐๐ผ๐ฒ๐ ๐๐ฒ๐ฒ๐ฝ๐ฒ๐ฟ ๐ง๐ต๐ฎ๐ป ๐ฌ๐ผ๐ ๐ง๐ต๐ถ๐ป๐ธโ๐๐ฟ๐ผ๐บ ๐จ๐ ๐๐ฎ๐ป๐ป๐ฒ๐ฟ๐ ๐๐ผ ๐ง๐ฟ๐ฎ๐ถ๐ป๐ถ๐ป๐ด ๐๐ฎ๐๐ฎ
written by Stefan Christoph
๐ฐ ๐๐ฒ๐ป๐๐ ๐๐ฑ๐๐ฒ๐ฟ๐๐ถ๐๐ถ๐ป๐ด ๐๐ผ๐ฒ๐ ๐๐ฒ๐ฒ๐ฝ๐ฒ๐ฟ ๐ง๐ต๐ฎ๐ป ๐ฌ๐ผ๐ ๐ง๐ต๐ถ๐ป๐ธ, ๐๐ฟ๐ผ๐บ ๐จ๐ ๐๐ฎ๐ป๐ป๐ฒ๐ฟ๐ ๐๐ผ ๐ง๐ฟ๐ฎ๐ถ๐ป๐ถ๐ป๐ด ๐๐ฎ๐๐ฎ
Just before heading out for my lunch run, I read the Reuters Institute for the Study of Journalism’s article “Advertising is coming for GenAI”[1],ย which is interesting, but from my perspective just scratching the surface. ๐ค
๐ง๐ต๐ฒ ๐ฆ๐๐ฟ๐ณ๐ฎ๐ฐ๐ฒ
The surface being, in a way, just the user interface which is used by human users to interface with some AI Agent. Here you can place ads, informed by the context of human-AI interactions, in multiple places:
1๏ธโฃ Outside of the conversation flow, like the good old banner advertisements on websites/TV/… 2๏ธโฃ Inside of the conversation flow, like inline ads
Likely other places and models are possible. I agree that users would want to have this clearly separated from the real content of the conversation.
๐๐ผ๐ถ๐ป๐ด ๐๐ฒ๐ฒ๐ฝ๐ฒ๐ฟ
But like I said, this is just the surface. Conceptually, think about product placements in video, just appearing to be content but actually sponsored ads. Or the good old search engines selling ad placements that sit above and alongside the organic results. ๐ฐ
A similar approach would be to interweave the advertisements directly into the agent’s answers. Possibly not easily recognizable as an advertisement, and, I’d guess, still impactful and comparatively easy for the provider of the Agents to manage.
๐ง๐ต๐ฒ ๐๐ผ๐๐๐ผ๐บ
But what if we drill deeper? What if Advertisers could pay AI Labs to include their advertisements into the training data of their models? The models could absorb a tilt from consuming these advertisements, how much depends on how often the material appears and on how the training data is sampled, filtered, and weighted. The intuition is like humans being mostly unconsciously impacted by the advertisements they have been exposed to, an analogy for the reader, not a claim that models learn the way we do. ๐ง
Science Fiction? Paid placement in training data is, as far as I know, not a business model in action yet. But an unpaid version is already happening: the models most agents use are trained largely on internet data, and marketing content rides in with it. And that is just the training path. At inference time, many agents also retrieve answers from knowledge bases sourced from the same internet, a second, separate door for the same marketing content.
And the internet, well, is full of advertisement and marketing data. Paid influence content. So watch out in your agent’s responses. What is the source? โ ๏ธ
๐ง๐ต๐ฒ ๐ค๐๐ฒ๐๐๐ถ๐ผ๐ป
What do you think about this? And, beyond an advertisement-free user interface, would you also pay to access agents/models built to minimize advertising influence? ๐ญ
From Technology perspective both are possible and in order to create good performing models, the training data is typically filtered and cleaned, which strips much of the obvious noise like banner ads, but cannot guarantee that embedded marketing is gone. But as we know it is hard to tell what of today’s internet content is just content.
I go now out having a coke, ARGH, No, for a run… ๐โโ๏ธ
๐ฆ๐ผ๐๐ฟ๐ฐ๐ฒ๐ [1] https://www.linkedin.com/pulse/advertising-coming-genai-reutersinstitute-btayf/
About the Author
Stefan Christoph is a Principal Solutions Architect at AWS, focused on agentic AI, media & entertainment, and helping builders move from demo to production. He writes about AI architecture, developer productivity, and the future of software.
This is a personal blog. Opinions expressed here are my own and do not represent the views or positions of my employer.