[lubujackson]

It is a valid concern. We are firmly in the goldilocks phase of LLMs, like in the first couple of years of Google when it was truly amazing. Then SEO made Google defensive, then websites catered to Google and not users, then Google catered to Google and not websites and we end up with 30 page recipe sites.

LLMs are obviously different and will have different challenges, but their advantage is how deep into a user's request they go. Advertising comes down to a binary choice - use product X or not. If I want implementation instructions for a certain product on specific hardware an ad will be obviously out of place and irrelevant.

So "shopping comparison" asks might get broken, but those have been broken for a while.

[wrs]

There wouldn't be an "ad" anywhere, though. You'll just ask the LLM for alternative implementations in plan mode, and it will be selling you one of them during the conversation rather than giving you an unbiased comparison. If you become suspicious it will make sure the pros just slightly outweigh the cons, or mention how well the thing works with something else in your stack, or whatever else a skilled salesperson would do to guide your choice without you realizing.

It's already doing this by telling everyone to use React and Tailwind, it's just that nobody's getting paid for it to do that.

[xnx]

> Then SEO made Google defensive, then websites catered to Google and not users,

Google was created in response to simple proto-SEO techniques (e.g. keyword stuffing) that already ruined Alta Vista.

Google has been combating adversarial information retrieval since inception.

Google's background with that is one of the reasons to expect they will stay on top of the AI race. The recipe is: lots of good/novel data x careful weighting of trust x algorithm.