Perplexity adds live web evidence to probability-based decisions
Perplexity has added optional web search to its Decisions API, which returns probabilities for predefined questions rather than conversational answers. The new mode searches before a decision and exposes the retrieved material to developers. Questions about changing public facts can draw on current pages, while responses still retain the service’s structured numerical format.
Artificial Intelligence··Midday
Search becomes an optional step before the decision
Perplexity added optional live web grounding to its Decisions application programming interface on 4 November. This service gives numerical probabilities for yes-or-no questions, predefined choices and ordered scores. The new mode lets a decision use current public information, including changes in a person’s executive role or a software release. It retains the structured answer format rather than turning the response into a conversational account.[1], [2]
Retrieved pages are visible in the response
When grounding is enabled, the service constructs searches from each question and the beginning of the supplied text. It combines results, removes duplicate addresses and adds page titles, links and text to the material considered by the model. Developers can inspect a separate search-results field. Questions that generate the same query share a search, while images supplied as input are not searched.[1]
A decision can return even when searches fail
The gateway uses the searches that succeed if others fail. If none succeeds, the model answers from the original input and returns an empty search-results list. Retrieved text counts toward the input budget. Developers using the Python or TypeScript libraries currently pass the mode in the request body because the library types do not yet expose that setting or the results field.[1]
