[keeda]

On a related note, I think that's a good monetization vector for chatbots. Long back I asked ChatGPT to recommend a few USB hard drives with very specific requirements -- including a small size and weight ("can be duct-taped to a tablet form-factor device without being obtrusive and constantly falling off"), low cost, and speed fast enough to boot an OS from. After a pretty technical conversation, it came up with 3 very specific products, one which ended up being the start point of the one I actually purchased on Amazon.

I came away thinking if those were presented as affiliated links, that conversation could have been monetized in a mutually beneficial way.

I keep thinking of that interaction whenever the discussion of ads on chatbots crops up. In an ideal world, model providers could better capture the value they provide. Unfortunately, I suspect such conversations are too rare, and often harder to monetize. (Like that time free tier ChatGPT helped me recover $500 in compensation for an airline delay. Great value for me, but no upside for OpenAI.)

Unfortunately given the amounts and timelines of returns that investors expect from OpenAI, and their scale, and with ChatGPT constrained to its “Free Consumer-facing Internet App” form factor, they are doomed to have to trade in the reserve currency of the web: ads.

[otherme123]

>I came away thinking if those were presented as affiliated links, that conversation could have been monetized in a mutually beneficial way.

I also ask LLMs for product recomendations. But the moment I suspect they are hidding the best items (not paying for the ad) to push the second best (not even talking about pushing shit as good products because they pay more) is the moment the LLM loses its value as recomender.

[toomuchtodo]

How do you know what recommendations you can trust from the LLM output?

[qup]

You don't, but if they recommend second best options with any frequency, you'll find out eventually.

[ProfessorLayton]

You don't. Use various other LLMs to check each other's work.

[dzhiurgis]

So the only way this can work fairly if Walmart, etc implements MCP/API?

[wat10000]

The trouble is that monetization and usefulness tend to be in conflict. It starts out as, show affiliate links if there are any. Then it turns into, prefer targets that we have affiliate agreements for. Then, don't show products unless we have affiliate agreements. Then, prioritize ones that give us more money. And on and on.

If they want to capture some of the value they provide, they should do it the standard fashion where they directly capture value from me by having me pay.

[keeda]

Yes unfortunately, the forces of enshittification will be at play with any model provider that has had billions invested in it.

> If they want to capture some of the value they provide, they should do it the standard fashion where they directly capture value from me by having me pay.

That would be ideal, and that's what their subscription products are for, but even their free versions provide so much value that is hard to capture. That's what I meant by chatbots being constrained by their "Free Internet app" form-factor... Consumers on the Internet largely just are not willing to pay, nobody managed to get micro-payments to work, and so the reserve currency of the web is ads.

[zymhan]

> I came away thinking if those were presented as affiliated links, that conversation could have been monetized in a mutually beneficial way.

Have you considered a world where the ideal outcome _doesn't_ involve making the most money?

[keeda]

It doesn't have to be the most really, but it's fair that they get some monetary returns commensurate with the value provided. All human society is based on an exchange of value, after all.

But realistically speaking, billions of investor money means that nothing short of the "most money" will be good enough.

[satvikpendem]

Gemini does this, it has Google Shopping links.