Perplexity launches Comet: AI browser begins to push search to the task execution layer

The success criterion for traditional search products is "whether relevant results are found quickly enough." Comet goes one step further: It treats search results as intermediate states, with the goal of moving the user to an actionable next step. This means that the focus of product design is no longer just recall and sorting, but also information integration and action connection.

From "information retrieval" to "task advancement"

When the user enters a question, Comet not only returns candidate pages, but also tries to help the user complete subsequent actions: summarizing, comparing, concatenating contexts, and reducing tab switching. This means that products need to be optimized not just for recall and sorting, but also for the “last mile before action.”

Under this logic, search accuracy is still important, but it is no longer the only core indicator.

New evaluation index: task advancement rate

Products like Comet are more suitable for evaluation using three types of indicators:

• Whether the time it takes users from asking questions to completing key actions is shortened.
• Whether the number of page switches and repeated queries required for the same task is reduced.
• Whether the user is willing to continuously advance complex tasks within the same session.

If these three items are true, the value of the AI ​​browser is not just "better answers", but "lower execution friction."

Risks are also magnified: acting faster means making mistakes faster

The stronger the task advancement ability, the easier it is for users to skip the verification link. In the past, users would double-check between multiple tabs; when the system puts information and actions on the same track, wrong decisions will be amplified faster.

Therefore, Comet's long-term competitiveness lies not only in the quality of production, but also in "when to remind users to stop and verify."

in conclusion

The value of Comet's experiment is that it raises a harder question: Should search products ultimately serve "know" or "do". If the industry answer shifts to the latter, the search market will shift from competition in traffic to competition in task completion capabilities.

This article is based on Perplexity’s official portal and public product information, and does not make inferences about undisclosed functional routes.

Reference: Perplexity Official entrance https://www.perplexity.ai/