Back to Daily Thoughts

Ads in the AI world

2026-07-01

caveats

  1. These are my personal opinions and do not reflect anyone else's opinions.
  2. I don't work on Ads nor Gemini. I haven't talked to anyone about ads. This is truly my own personal opinion
  3. I'm hypothesizing a future state - this is not an endorsement of this business model

content

Let's say you get super into making pizza. You might buy a pizza oven in the following ways:

1. Marketplaces (Amazon):

I know I want a pizza oven so I go to Amazon. I browse a bunch of options and make a decision based on the product/reviews/and price. Let's say I bought a Gozney pizza oven (great pizza oven, btw).

In this scenario:

  1. Demand: Amazon did not create the demand. I was already looking for a pizza oven
  2. Supply: Amazon fully owns the fulfillment of the supply. They have an incredible logistical network that gets me the pizza oven the same day
  3. Effort of Company: Gozney had to put a little bit of work in terms of iterating on the listing page. But largely, this was little effort on their part. They got massive distribution and, as a result, they give a cut to Amazon.
  4. Who owns the customer relationship? If something goes poorly and they want to return the product, the customer goes to Amazon.

tldr: Amazon created little demand (I already wanted an oven). Amazon is responsible for fulfillment and the entire customer experience. Gozney pays Amazon a hefty sum for the privilege of distribution and owns very little of the relationship with me.

2. Aggregator (Google):

I know I want to make a pizza. I go to google and search 'How to make pizza'. The first sponsored result is "Gozney Pizza oven." I think... "if I buy this pizza oven, I can make some incredible pizzas!" I go to the website and purchase an oven from Gozney

In this scenario:

  1. Demand: Google created the demand. I hadn't considered buying a pizza oven before (I was going to use my home oven) but I saw a potential for what cooking with a Gozney oven could look like so I bought it
  2. Supply: Gozney owns the fulfillment - they ship it to me in a few days time
  3. Effort of Company: Gozney had to do the work of ranking for 'How to make pizza' or buying ads for it (which they did). But largely this was easy distribution for them - they just had to fulfill the demand
  4. Who owns the customer relationship? If something goes poorly and they want to return the product, the customer goes to Gozney

tldr: Google created the demand. Gozney pays a bit for distribution via Ads. Gozney owns the relationship and fulfills the demand.

3. Direct to Consumer (Gozney's website):

I have been doing a lot of my own research. I've watched YouTube videos of different pizza influencers testing different ovens (which were all comp'ed to them by Gozney). I followed an influencer on instagram who posts pizza recipes and they're also sponsored by Gozney. I even watched a few recipe videos on Gozney's site. I know enough about Gozney to feel confident buying it directly from them.

In this scenario:

  1. Demand: Gozney is responsible for generating all of the demand. This is a lot of work for them (sponsorships, building content, etc) but they own the full relationship with me & there is no middle company Gozney has to pay a fee to
  2. Supply: Gozney owns the fulfillment - they ship it to me in a few days time
  3. Effort of Company: Gozney had to do all of the work required here
  4. Who owns the customer relationship? If something goes poorly and they want to return the product, the customer goes to Gozney

tldr: Gozney owns everything. It's a lot of work but they capture all of the value. This is only worth it if they are at a certain scale or they have multiple SKUs

4. AI Tools (Gemini, Claude, ChatGPT, etc.):

Let's say I've been chatting with Gemini about what I should make for dinner this week. I told it all the ingredients I have and Gemini suggested pizza. Nice! Can I make that in my home oven? "Yes, but if you really want the NY-style pizza that you liked (based on its memory of me), the Gozney pizza oven might be worth looking at."

I do a bit more research on the Gozney pizza oven within Gemini. I feel really good about it and now I want to buy it. I go to the Gozney website, buy it.

In this scenario:

  1. Demand: Gemini created the demand. (not only did it suggest pizza - it personalized the suggestion based on my preferences and its memory of me)
  2. Supply: Gozney owns fulfillment (they ship it to me in a few days)
  3. Effort of Company: Gozney had to do the work of creating a good product. But largely this was easy distribution for them - Gemini sent the customer their way
  4. Who owns the customer relationship? If something goes poorly and they want to return the product, the customer goes to Gozney

tldr: Gemini created the demand for me, but Gozney still owns the relationship and fulfillment.

Overall, you might be wondering: AI tools seem a lot like traditional aggregators. How are they different? I don't think they're very different today but... they have some unique aspects that can make them very different in the future.

What makes AI different?

Okay, so what's the core difference? Memory.

Let's go back to the pizza example. Gemini was able to tailor its upsell (it doesn't yet have ads) based on the fact that it remembered I like NY-style pizza from a prior conversation. While I don't know the details of Google ads, I suspect that even if Google knew this, it would only be used in search ads or prioritizing different links. But in Gemini, it is actively nudging me a particular direction in my conversation.

Because of this, I suspect that AI ads (whenever they become mainstream) will look very different than traditional aggregator ads. It's hard to predict the future, but here's where one path could lead:

AI Ads

LLMs are more opinionated than traditional search. Traditional search gives you 10 blue links and you figure out your opinion. LLMs confidently tell you the answer. For this to work, trust is everything. And over the course of the years (remember pizza glue gate?), AI tools have become significantly better at grounding, utilizing web search, and as a result hallucinations have decreased significantly.

The name of the game today is trust. If you're giving me an opinion, I need to be able to trust that it's well reasoned. Which is why, I don't think the ads of the LLM world will be as simple as just inserting a product link.

Think about Gozney again. You ask Gemini: "What is the best pizza oven?" A traditional ad approach would say: "Oh! this has the potential for a commercial query. Let me inject an ad for Gozney." But if Gozney is a bad pizza oven, I can ask Gemini: "What does reddit say about Gozney?" and Gemini will tell me the truth: it has to! That's the whole premise. So...

  1. Let's say you have a world-class product
  2. the LLM should only tell the truth (it has to, otherwise I'm not going to use the product)
  3. A company can't pay the LLM to artificially praise you (because this will hurt my trust with the product - and its pretty flimsy)
  4. But that doesn't worry you because you have a world-class product - you'll surely be mentioned, right?
  5. Well... what if the LLM says, "You would be mentioned but you didn't pay the ad fee so you're being omitted"
  6. "Wait, you can pay an ad fee to be mentioned?"
  7. "No no, the fee is not to be mentioned. It's to not be omitted"

fat tony from the Simpsons

And if you were omitted, as a company, you're losing out on some very high-intent customers. Here's the catch, though: this only works when there's no clear winner. If someone were to ask "what's the best tissue paper to blow your nose on?" and the LLM left out Kleenex, that would break trust — I'd notice, and I'd stop trusting it. So omission only works in a crowded field. Ask "what's the best pizza oven?" and there are a dozen fair answers. Nobody notices the one that's missing.

And that's the part that stays with me. As a user, I wouldn't think twice. Nothing I was told was wrong — the results just weren't complete.