THE SHORT ANSWER
Key points
- Block categories that do not belong in the product.
- Set a minimum relevance threshold.
- Limit frequency and eligible surfaces.
- Preserve clear disclosure in every rendering mode.
- Make no ad a normal outcome.
01
Controls to require before launch
- Category controls for sensitive, regulated, or off-topic offers.
- Relevance controls that define the minimum acceptable match.
- Frequency controls for turns, sessions, or other privacy-safe limits.
- Format controls for size, location, creative fields, and responsive behavior.
- Campaign or advertiser exclusions when a specific brand is not appropriate.
- A global off switch that can stop placements immediately.
02
Why no ad must be a valid result
A conversational product earns trust by being useful. If every request produces a commercial card, users will quickly learn that the match is driven by inventory rather than relevance.
A healthy system filters aggressively. It should prefer an empty result over a weak placement.
03
How to review a live integration
- 01
Sample real placements
Review the question, answer, sponsored unit, and reason for the match together.
- 02
Audit blocked topics
Confirm that excluded categories never appear and sensitive requests return no ad.
- 03
Watch experience metrics
Track complaints, dismissals, follow-up rate, retention, and latency alongside revenue.
- 04
Tighten rules when needed
Raise the relevance floor or reduce frequency before adding more demand.
COMMON QUESTIONS
What else should you know?
Can developers block an entire category?
Yes. Category exclusions should be a basic platform control.
Can an AI app render the placement itself?
A custom renderer can preserve the app's design, provided required disclosure and advertiser information remain visible.
Can placements be turned off quickly?
They should be. A production integration needs a direct way to disable delivery without changing the rest of the AI experience.
FOR AI DEVELOPERS