Advertising is coming for GenAI consumer chatbots. In part 1, we looked at the parallels between web search and information retrieval with chatbots. Chatbots already have a rich collection of user data that can be leverages for ad targeting. Let’s take a look at how they can build an advertising platform to monetize usage.
The most straightforward approach is to mirror what web search already does. Allow advertisers to bid on search queries and then charge them based on impressions and clicks. Because chat users are entering conversational input as opposed to keyword queries, this might look a little bit different. The platform could share the underlying queries it runs with advertisers, or it could do something a bit more advanced, like selling ads on intent matching. Or possibly both, and charge advertisers more to match an intent vs a specific search query. The underlying LLM is well-suited to determine intent and pass that to ad matching. Once ads are retrieved, they can be displayed in the user interface just like search ads are today. The platform could also choose to show the ads first, while the final response is being generated.
You have probably noticed that some types of GenAI requests take longer to complete. Think image and video generation, reasoning models, or complex coding tasks. These longer-running tasks also cost the most money for the chatbot company. Instead of watching the pixels come together for your generated image, the platform can show you an ad while you wait. This would look a lot like how YouTube forces you to watch ads before showing you the underlying content. YouTube doesn’t have to show you ads, the content is already ready. You could argue that this approach is more justified in an unpaid GPT environment where the content isn’t pre-created, and there are real costs for longer inference tasks.
There are boatloads of money to be made through these traditional advertising methods, but they are a bit boring. There is a lot more we can do with GenAI. GPT chatbots have been shown to be incredibly persuasive. What it is advertising if not persuasion? With a basic advertising framework in place, platforms can start enabling AI generation for ad copy. Traditional ads are static. GenAI enables advertisers to build dynamic and personalized ads. Every ad you see could be unique to you. Instead of writing traditional ad copy, advertisers can write prompts that leverage user attributes and conversation history. The platform can charge advertisers more for prompt-generated copy vs traditional ads.
Prompt-generated ads could be incredibly personalized, especially in image/video generation environments. Think about an ad showing your Sora avatar happily using some new product. There are practical constraints on the time needed to generate ads like these, but that could be solved in the future.
Why stop here? Advertising could be integrated into the response itself. Unlike search, where users see individual results, the summarization nature of GPT chatbots allows the platform to obfuscate where the data is coming from. The advertising platform could enable advertisers to boost their search results in a given query to increase the likelihood that their page would be used in the summary. More blatantly, the platform could enable a direct pay-to-include feature, allowing advertisers to add their content into responses. This approach could make it very difficult for users to determine what is an ad and what isn’t.
Last, and probably the most extreme example, model creators could sell sponsorship for models. GPT 7, brought to you by Walmart! Training models is very expensive, and this would offset or potentially cover those costs. Post-training could build in preferences to create favorable responses for model sponsors. Building ads into the model itself would mean that they not only show up in chatbots, but even at the API layer for users building off of the consumer model.
In the next part, we will look at how mobile and voice enable more monetization opportunities.