Margin Points - Arnold Engel

July 15, 2026 · [Essays 121, 122, 123]<br>OpenAI advertising, Hinge, Chargebacks

→ AI chatbots know exactly how to sell to you.

→ Hinge introduces dating profile endorsements from friends. Fun.

→ Credit card chargebacks are built on assumptions that are crumbling.

Made possible by<br>Spacebar Studios<br>The #1 Newsletter Agency For B2B Companies. Work with us →

Selling Pintos<br>Tomasz Tunguz1, GP at Theory Ventures, was on The Information talking about how AI chatbots will be particularly good at personalizing advertising for each individual based on chat history and individual context.

Google Ads are great at giving users options when the intent is clear in a search. Meta, with Facebook and Instagram, has remarkable targeting of interests and creative formatting. OpenAI is banking on being able to steer users through the consideration stages of a purchase nimbly without turning users off.

LLM ads are imagined as a single trusted salesperson voice. That world won't materialize, which will mean disappointing results for OpenAI.

Google Ads shows you words that you wrote2 to sell you whatever is in the blue link. Meta shows you pictures or videos that look like things you might like. OpenAI can show you words that ChatGPT writes just for you and in the format you are most receptive to.

Ads in ChatGPT can be counter-programmed—users can ask another chatbot why the product being pitched is or isn’t a good fit for them and get an immediate alternative perspective. Ads that Google and Meta serve can’t be counter-programmed. We don’t yet know how important the counter-programming distinction is. I think it will push ads in ChatGPT to focus on lower-value purchases rather than higher ticket items and services. The high-value decisions will warrant a second opinion from another LLM.

There is reason to think that steelmanning and strawmanning for purchase decisions is important, particularly for high-value purchases. People can be indecisive for large, infrequent purchases and often seek advice or perspectives from friends. Today, we see many people getting advice from chatbots. The highest chatbot usage for purchase consideration is for bigger purchases. Big purchases are where there are lots of dollars at play but also other voices to consider.

Let’s say I’m interested in buying a new car. ChatGPT is steering me towards a Pinto,3 letting me know how great the heating system is for cold winters. I am really partial to warm steering wheels, so I lean in. ChatGPT seems to be making some sense! And ChatGPT kindly told me with a little note that it does have an advertising relationship with Ford, so I know the Pinto is being shown in the best light. Since buying a car isn’t something I do regularly, I want to feel like I am being diligent, and so somewhere mid-conversation I decide to get another opinion. Google’s Gemini informs me of my impending folly and links me to some YouTube videos to watch of cars going up in flames.

For every signal, strand of logic, or emotional pull that an LLM can execute on a user to induce purchasing or steer behavior, there is a potential counter-signal. The LLM telling me that I should upgrade to leather seats because I will appreciate the easy cleaning with kids and because it has a strong positive impact on resale value is hitting high notes that I like to hear. It’s skipping over the parts of the script about impressing a date or comfort for long trips. Another LLM might very well counter-program the advertisement by highlighting that I’ll be happier waiting another year before buying as early leaks of the next model year features are a better fit for the cargo space I’m looking for. I feel like a very smart car buyer as a result. I don’t know how the car companies feel. It’s like demand was flipped topsy-turvy by an LLM.4

The world of the single trusted salesperson is an illusion. The world will always be filled with multiple voices competing for your cash, time, attention, and energy. Can OpenAI build an unbelievably good salesperson? Sure. In fact, I think I’ve already been sold on that. But if other AI labs build great salespeople as well—OpenAI’s conversion rate will drop.5

I think you'll like him<br>Hinge, the popular dating site owned by Match Group, rolled out a new feature enabling friends to leave testimonials6 on a friend’s profile page. According to Bloomberg:

The tool is meant to offer endorsements from well-meaning, preapproved wingmen. Daters can enlist as many as 10 close contacts to share written feedback, photos, voice notes or videos about them. Friends will be invited to answer prompts, such as recalling a moment they still laugh about or imagining what the person is presently doing.

Hinge is trying to shrink the gap between what people are expecting before the first date and the real person they encounter, which they view as a top priority to improve.7 The friend endorsement feature has a lot of second-order implications for both...