Planet’s AI app can find places to investigate. A blank result is not proof.

Planet’s November 5 release lets people search more than ten years of satellite imagery through a map-based chat app, with access free for now. Users can describe what they want to find instead of starting with coordinates. The immediate value is identifying places worth investigating—not proving that a whole region has been exhaustively checked.

This follows an earlier beta, rather than the invention of the system this week. Planet’s June 25 account already described an open beta combining imagery, change detection and public web information. The November release emphasizes improvements for the discovery stage, based on feedback from summer testers.

In a November 6 discussion with GIS users, Planet’s product team described the app as an entry point for people outside its traditional specialist audience. The team said nonprofit, education and research users had become especially active during beta, and that experienced GIS users could treat it as an exploration shortcut. Those are the company’s observations, not independent adoption measurements.

The chat searches a visual index

According to the release explanation, a language model interprets the request and queries a vector map built from visual features in monthly mosaics. The app can then show imagery across time and use Deep Research agents, external files and public information for analysis. This describes the vendor’s workflow; it is not a published accuracy score.

The distinction between finding and measuring matters. A search for a type of site can produce useful candidates even if some are irrelevant or others are missed. A claim about how much land changed needs a defined area, comparable observations and an assessment of what the method failed to detect. Fluent text does not supply those conditions.

Editorial visualization

Planet’s May 13 technical Q&A gives an unusually direct explanation of the search limits. In that beta description, broad-area screening first used an internal search engine, then selected some hits for closer visual analysis. Neither step was exhaustive. The company explicitly framed this as a way to surface targets, not rule out their existence.

That document predates the latest release. It should not be treated as a current specification for every interface feature or data entitlement. But the November announcement still describes vector search over monthly mosaics, and provides no measured completeness result that would justify treating a blank search as proof of absence.

Check the image, then the conclusion

The same May Q&A explains that optical imagery can remain cloud-obscured even in monthly mosaics designed to reduce cloud cover. It also reported variability between repeated searches during beta. These are dated disclosures, not new November measurements; they identify questions to test when using the current app.

Editorial visualization

For a conservation researcher or local reporter, a useful first question might describe a visible feature within a named area and period. Save the returned locations, open the underlying imagery and record the dates before deciding what changed. If a result suggests wrongdoing, the image alone does not establish who acted, whether they had permission or why the change occurred.

The product team’s November 6 answers also distinguish access from a complete professional workflow. They described the present launch as focused on individuals, with better developer onboarding still ahead. The team said chat can respond in multiple languages while buttons and menus remain English-only. Indonesian-language questions therefore need their own quality check, rather than an assumption of equal performance.

UpThat has not tested the app. The sources here are Planet’s own documents and employee explanations, so they establish what the company offers and discloses, not independent reliability. They do not establish a universal search success rate or a guaranteed time saving.

The useful output is a shortlist that survives inspection. If the tool helps a researcher choose the next image to examine or the next place to check, it has advanced the investigation. The evidentiary claim still has to be earned from the observations behind that shortlist.