# What can an ad matcher learn from four fields?

Design a small context payload and test which details actually improve the match.

Published: 2026-11-05T03:01:04Z

![What can an ad matcher learn from four fields?](https://jolts.xyz/rails/active_storage/blobs/redirect/eyJfcmFpbHMiOnsiZGF0YSI6MywicHVyIjoiYmxvYl9pZCJ9fQ==--561fd31aa5caa5445f8bd35fd4e31ee97d756a03/context-without-the-transcript.webp)

Consider two messages a shopping assistant might receive.

> I need a waterproof daypack under $100 for my sister’s commute.

> My sister’s $100 waterproof daypack leaks. How can I patch it?

A parser can extract the same category, feature, price and use from both. In the first message the price is a ceiling. In the second it is a fact about an item the user already owns. Turning both into ad requests would throw away the most useful piece of context.

For the shopping turn, a publisher could send this small payload after establishing that the user wants a recommendation. The currency is assumed here for illustration.

```json
{
  "category": "daypack",
  "use": "commuting",
  "required_features": ["waterproof"],
  "max_price": {"amount": 100, "currency": "USD"}
}
```

Hand this payload and three product descriptions to someone who has never seen the chat. A waterproof $75 commuter pack passes. A waterproof $140 pack exceeds the ceiling. A $65 canvas pack lacks the required feature. The shopper’s relationship to the eventual owner contributes nothing to those decisions.

## The request you can leave out

For the repair turn, let the assistant answer the repair question. Translating an old purchase price into a new budget would introduce a commercial intent the reader never expressed.

One revealing integration test uses a distinctive fake name in a previous turn. Record the outbound request, including diagnostics, and look for that name. Run the repair case as well and count outbound placement requests. Passing the first check says the payload is small. Passing the second says the app can decide when to keep the conversation entirely to itself.


## Related terms
- [Contextual advertising](https://jolts.xyz/resources/glossary/contextual-advertising)

## Next step

[Explore advertising](https://jolts.xyz/advertisers) or [monetize your AI app](https://jolts.xyz/publishers).