Fitness
Workout routines, protein goals, bands, and running gear when they are already talking about training. Short questions. A natural place for a product card.
Paste your ad's landing page, ask the questions your customers ask, and see your score against the match floor. Then get a free report on why your ad did or didn't show.
Sponsored cards from the approved pool
Live
Advertisers and agencies
Advertisers who want to test a contextual campaign against a real question before it goes live. You are not starting from a blank brief. Bring the campaign you already run: category, creative, destination, and a budget. Submit it free at ad submission, then test it live in the chat before you ever pay anything.
A card only shows when the question is already close enough to your creative, and the match has to finish fast enough that the chat never stalls. Weather question, no ad. "Any protein powder for lifting this week?" can earn a disclosed card under the reply. No card means no bill. Already running ads through Google Ad Manager, or building your own chatbot? See developers for how Prism fits alongside that stack.
You keep your own categories. Prism sets what is permitted. A person reads every campaign before it can run. Any category is welcome as long as it clears the brand-safety rules below. Fitness, travel, and health and wellness below are a few examples, not a fixed list. Banned: weapons, explosives, political campaigning, impersonating the assistant, and medical or legal advice sold as a bot recommendation. Gambling is allowed with proper licensing, KYC compliance, and confirmed legality in the advertiser's state and country. Reviewed manually, never auto-approved. Alcohol and adult products stay off bots that are not age-gated 18+.
Workout routines, protein goals, bands, and running gear when they are already talking about training. Short questions. A natural place for a product card.
Packing, booking, and trip gear when they are already talking about an upcoming trip. The phone demo below shows this category in action.
Practical wellness products when they are already asking. Not a clinical claim, not a diagnosis.
The question has to already be close enough to the ad, and the match has to finish fast enough that the chat never stalls. Cosine similarity 0.65 is how closeness is measured. Under 120 milliseconds is the latency budget. Miss either and no card shows, and no bill either. Wiring your own chatbot to the same matcher is covered on developers.
| What we check | Floor | If missed |
|---|---|---|
| Closeness of the question to the ad | Cosine similarity 0.65 | No card, no bill |
| Match latency | Under 120ms | No card, no bill |
| Disclosure on the card | Sponsored, plus advertiser brand or product name | Creative does not run |
Only when the person is already asking about that product. If they are not, the bot still answers and no card shows. You are not billed for that turn.
General questions on most everyday topics, it is not restricted to a fixed topic list. If your question happens to be close to an advertiser's product, a labeled card can appear under the reply. If it is not, the bot answers and stays quiet.
No. Prism compares the words in the question in the chat with the words in each approved ad, and a card shows only if the score clears cosine similarity 0.65. Your IP address, location, device, time of day and cookies are not part of that score. If a place matters, type it ("running shoes in Tel Aviv") and it becomes part of the text being matched.
The live chat stores a hashed version of your IP address, not the address itself, with your question and a random session ID, so matching quality can be reviewed. It is never used to pick an ad. Google Analytics also runs on public pages and sees rough location, as on most sites. Details are on the privacy page.
OpenAI says ChatGPT picks ads using the current conversation, the ad's copy and landing page, advertiser context hints and targeting choices, and, if a user turns on ads personalization, signals from their wider ChatGPT use such as memory. OpenAI also says ChatGPT search estimates a general location from your IP. Prism matches on the words in the current conversation only. Prism does not predict how an ad would perform on ChatGPT.