Planet AI lets you search ten years of satellite imagery by chat – for free

What it's about

US company Planet (full name Planet Labs) runs one of the largest fleets of small Earth observation satellites and has been photographing almost the entire land surface of the planet nearly every day for years. On November 5, it launched Planet AI. The app connects a large language model, the kind of AI behind ChatGPT, to Planet's satellite image archive covering more than ten years. The idea: anyone can search for, track and analyze changes on the Earth's surface using plain language.

The app runs in the browser at ai.planet.com and is currently free. Users who sign in with a Planet Insights account, the company's commercial data platform, are supposed to get additional features later. According to available reports, Planet hasn't said exactly what those will be or what they'll cost.

Map plus chat instead of a GIS degree

If you wanted to analyze satellite imagery professionally until now, you usually needed GIS skills. GIS stands for geographic information system: specialist software where you stack map layers, pick image sections, compare time periods and kick off analyses by hand. It's powerful, but hard for non-experts to use.

Planet AI goes with an interface made up of a map and a chat window instead. You ask something like "Find newly developed areas in this region" or "Track how this forest has changed". The AI breaks down the request and searches a global vector map for matching areas. A vector map stores roads, borders and buildings as lines and shapes rather than pixels. The AI also taps into monthly image mosaics, where many individual shots from one month are stitched into a single image that's as cloud-free as possible. The tool then shows you the matching satellite images so you can check them yourself.

The three core features

Planet describes three main capabilities:

1. Search the Earth: The AI recognizes visible features in the images, such as structures, cleared land or specific facilities, and can find targets across large areas. You don't need to know beforehand exactly where something is.
2. Investigate change: The tool goes back more than ten years through the imagery and figures out when something changed. Results come as side-by-side before-and-after comparisons or as a time-lapse, a short image sequence spanning months and years.
3. Deep analysis: For more complex questions, Planet AI deploys AI agents on its own. These are programs that break a task into steps and work through them one by one. You can upload your own data, such as images or CSV files (simple tables like the ones you'd export from Excel). You can also track trends over time and set up recurring reports.

Fighting AI hallucinations: every answer comes with image proof

Language models are prone to hallucinations: they sometimes invent facts that sound convincing but are wrong. In satellite analysis that would be risky, for example if an agency acts on supposed illegal logging. That's why Planet stresses that all of the AI's conclusions are based on satellite images you can view yourself. The AI doesn't just make a claim, it hands you the image too. That way people can check the result with their own eyes before they act.

That doesn't rule out errors. The AI can misread a feature or miss something. But the approach deliberately leaves the final check to a human.

What early testers are using it for

According to Planet, the first testers mainly used the tool in three areas: conservation, land management and environmental compliance. The company gives two concrete examples:

• Conservation teams monitor habitats and spot illegal land grabs or fire risks. Before sending rangers into an area, they first check deforestation alerts on screen.
• Environmental compliance teams filter industrial sites by criteria like proximity to waterways and build a list of which sites to inspect first. Planet says this takes a few minutes, where it used to take several days of manual GIS work.

These time figures come from the company. There are no independent measurements yet.

Part of the "Planetary Intelligence" strategy

Planet frames the app as part of its "Planetary Intelligence" strategy. The idea is to combine daily satellite imagery with AI to understand changes on Earth in near real time. That shifts the business model from simply selling images toward delivering ready-made answers. Customers would no longer buy raw data and analyze it themselves, they'd just ask questions directly.

According to IT Home, industry observers see this as a turning point. Earth observation is moving from a niche for remote sensing experts to a decision-making tool that works via natural language and is usable by government agencies and the public. If image coverage, refresh rates and traceability keep improving, agriculture, disaster relief, city administration, insurance and infrastructure could become further use cases.

Our take

What makes Planet AI interesting is mainly the low barrier to entry. High-revisit satellite imagery has so far been mostly for specialists and paying enterprise customers. Free chat access where you can ask "What's been built here since 2018?" and get images as proof opens the archive up to journalists, NGOs, local politicians, researchers and anyone who's curious.

Some questions remain open, though. It's unclear what image resolution the free version offers. Monthly mosaics are coarser and less frequent than the daily individual shots Planet is known for. Planet also hasn't published numbers on how reliably the AI detects objects or how often it gets things wrong. The limits of free use and the price of the announced extra features are unclear too. Finally, such an easy-to-use tool raises questions about privacy and potential misuse, for example if someone deliberately keeps tabs on specific properties. The available sources don't address this.

Still, the direction is clear: satellite data is becoming searchable in much the same way as the web. Putting evidence images front and center is the right response to the well-known hallucination problem of language models.