PerplexityapiWeb Grounding in the Decisions API
For developers using Perplexity's Decisions API: one setting now makes the model search the web before it answers, so it can handle facts that changed after training.
Details
- Adding "mode": "web_grounding" to a request runs one Perplexity Fast Search query per question and adds the results to the model's evidence
- The response returns a probability along with the pages the model read, so your code can act on the answer and check its sources
- You can also pass a URL instead of text to point the model at a specific page
- On a public 100-question test built on current web facts, accuracy rose from 51% to 86%, with median responses of 370 ms
- In Perplexity's own test of 328 true-or-false claims, median response time rose from 139 ms to 366 ms with grounding on
- Each completed search costs $0.001 on top of input tokens, search evidence counts as input tokens, and a request can include up to 100 questions
Best for
Yes-or-no and scoring checks that depend on current facts, such as whether an account raised funding in the last 90 days.
When to use it
Turn it on if the answer could have changed since the model was trained and nothing you send it includes that fact.
Who benefits
Developers building lead qualification, monitoring, or agent workflows, and the revenue and marketing teams those tools serve.
Join the newsletter to get ahead with AI.
I share strategies, new modules, free guides, and tools to get you better at using AI in your work.
Leading a team? See team programs →