[bobbiechen]

I wrote about this idea here: https://digitalseams.com/blog/the-ai-lifestyle-subsidy-is-go...

Quick summary, I believe consumer AI experiences will feature ads because the profit opportunity is too large and company valuations depend on it. The hiring of Fidji Simo (ads at Facebook) at OpenAI + and just this week, Vijaye Raji/Statsig also point that way.

[rhubarbtree]

We’re in a golden period where AI results are ad-free.

One thing I’ve been doing is querying and storing results. For example, “what are the best books on X topic” for every topic I can possibly think that I may want to read about in the future.

I’ve found the results to be amazing if you give a sufficiently detailed prompt. I have enough reading to see me through to exit.

[Larrikin]

I've already started getting results infiltrated by SEO but for AI. Deep research did return seemingly a top book when I was looking into Ansible. I independently verified it with my own searching in a few different places.

But the other recommendations seemed like crap and when I followed the sources they seemed like AI generated garbage for AI that I couldn't find doing my normal searching.

[HSO]

By the time this „golden“ ad free time is over in 1-2 years you should be able to run your own custom model locally

[dataexec]

I hope so but we could also be staying in an eternal state of FOMO as the proprietary models keep getting marginally (or a lot) better.

[the_other]

How are they better if they spam the user with ads?

[mcny]

One possibility is that if they can manage to lower interest rates back to zero ish, we might see a hiring frenzy in machine learning/llm/genai, starving upstarts and free software of talent, slowing progress in custom local models? Or is this too pessimistic/ "out there" of a take that requires the stars to align just right?

[delfinom]

Interest rates at zero and there won't be any clients for these upstarts as people will be in bread lines from their worthless money.

[bloomca]

Some models will still trickle down; hell, better/cheaper hardware should enable to run hefty models available today, and they seem to be already okay-ish with such queries.

[stevage]

Would you consider sharing these somewhere?

[pickledoyster]

The training data is full ads. For books, you have publisher-influenced rankings, SEO slop and promotional social media posts. It's GIGO, and has been that way from the start.

[bobbiechen]

(me again) and Drew Breunig just posted about the tensions of actually getting those ads in: https://www.dbreunig.com/2025/09/02/considering-ad-models-fo...

[ankit219]

I like the thesis, (and probably not a fully informed opinion here), it's easier for openai like router to go the affiliate model than ads model. Router can determine how much a query is worth (eg: help plan a vacation is worth $50 vs say what is the capital of Australia as $0). This further informs how much compute to use, and this can lead to how much they get on affiliate fees if a link is clicked. Ads here are counter to the proposition in the sense that results are to be trusted. Then, despite everything, while contextual ads are good, personalized ads are still going to be more effective (gut feel). To get there, openai needs the kind of infra Facebook and Google has. They can get the same value (perhaps more) from affiliate links - especially in a world where they are kind of gateway to the discovery - and don't have to do as much work on the infra side. This also aligns incentives for all three - companies, consumers, and middlemen, in a way it only happened with Google before this.

[dbreunig]

Yeah, I agree. Almost mentioned in the post how I imagine an ad PM at OpenAI is jealous of an ad PM at Perplexity.

[BobaFloutist]

Affiliate links are literally ads

[rjh29]

ChatGPT is already returning lists of products (with photos and rating) saying they are impartial, I guess to collect affiliate fees. It's not a big jump to have sponsored products showing up first.

[unyttigfjelltol]

If folks corrupt the integrity of LLM responses, as it were, they’ll destroy the value proposition.

More likely the model will be payment for low friction enablement of transactions rather than overt steering. Pick Door #1: the LLM states the product is fit for purpose. Pick Door #2: the LLM will directly complete the transaction or close to it.

[imiric]

> If folks corrupt the integrity of LLM responses, as it were, they’ll destroy the value proposition.

Highly unlikely. Google's SERP is an ad-infested abomination that sometimes shows useful results, and yet people still use Google Search.

The same will happen with LLMs, except in far more subtle and insidious ways. Instead of showing you ads directly, they will be naturally interwoven in conversations, suggestions, and generated content. You won't be able to tell whether the content is genuine or promoted, as is common on the web today.

The ads will target you more accurately than ever before based on not just the data you've given them, but on the context of the conversation, your surroundings, and any other piece of real-time information they can use to secure a conversion, or to influence your thoughts on a particular matter. You will trust it more than any current ad channel since the AI will be personal, and the tone will be friendly.

As with the web, ad-free services will exist, but the only way to escape this entirely will be to use local and self-hosted models.

[LtWorf]

> Google's SERP is an ad-infested abomination that sometimes shows useful results, and yet people still use Google Search.

I've personally stopped doing that. I understand it's not a mainstream decision since most people don't even know what alternatives are there, but doesn't mean that in a couple of years google won't start to feel it.

[baq]

OTOH with the help of local models you should be able to post process any and all content on device to at least highlight, desensationalize if not outright remove quite a lot of those ads, at least until DRM folks lay their hands on it.

[hpdigidrifter]

>yet people still use Google Search

Google search is widely acknowledged as drastically drying up in the last year or so, that's despite more worldwide internet usage.

[tanseydavid]

IMHO a huge part of people choosing ChatGPT over Google for search is due to the absence of ads and other distractions.

Doing search with ChatGPT feels like doing research in a library setting whereas Google search feels like doing research in Times Square.

[lithocarpus]

Many people still use google search probably because the LLM response is at the top.

[seabombs]

Insightful article, thanks for writing/sharing!

[dfsegoat]

Very good catch. For all my interest this space, it's embarrassing that I did not remotely see this coming...