What is an AI assistant for GIS? A plain-English guide

Ask a general AI chatbot how to reproject a shapefile and you’ll get a decent answer. Ask it to actually reproject your shapefile, check the result, and add it to Monday’s client report, and it can’t. That gap is what an AI assistant for GIS is meant to close.

The work that eats a GIS week

Talk to analysts at consultancies, plantations, or forestry agencies and the same tasks come up:

• Checking whether new satellite scenes cover your areas of interest, and whether they’re usable through the clouds.
• Merging files from surveyors, fixing invalid geometries, and getting everything into the right coordinate system.
• Re-running the same analysis every month: NDVI, change detection, area per land cover class.
• Turning results into map layouts, tables, and dashboards for people who don’t use GIS.

None of these are hard. They’re repetitive, easy to get slightly wrong, and they push the actual analysis to the end of the day.

What makes a GIS assistant different

A useful GIS assistant needs four things a general chatbot doesn’t have.

1. Spatial understanding. It knows that EPSG:4326 and EPSG:32748 aren’t interchangeable, that a self-intersecting polygon will break an area calculation, and that a 10 m Sentinel-2 pixel can’t tell you about a single tree.

2. Access to your tools and data. It works inside QGIS or ArcGIS, reads your layers and attribute tables, and can query imagery catalogs. Without that, it can only give advice.

3. A habit of asking first. GIS data is often a client deliverable or a legal record. A good assistant shows a preview and waits for approval before it overwrites a file or sends a report.

4. Memory of how you work. Your default CRS, your office layout template, your client’s AOIs, the folder structure of each project. You shouldn’t have to explain those twice.

What it looks like day to day

A realistic morning with a GIS assistant:

1. At 7 AM it sends a short brief: two new Sentinel-2 scenes over your estates, one with heavy cloud; a map revision due Thursday; a surveyor’s shapefile with 14 invalid geometries.
2. You reply “fix the geometries and add area in hectares”. It runs the fix, shows the result as a temporary layer, and saves only when you approve.
3. You ask whether anything changed in block C-07 this month. It compares the two clear scenes, highlights 12 ha of possible clearing, and drafts a map.
4. You check the map, adjust the legend, and send it. The assistant logs what it ran, so next month’s report is one message away.

The analyst still makes every judgement call. The assistant handles the steps in between.

What it isn’t

• Not a replacement for QGIS or ArcGIS. It works on top of them.
• Not a replacement for the analyst. It can flag a change; deciding whether that change is a violation, a harvest, or a cloud shadow is still your job.
• Not a black box. You should be able to see the code or tool it ran, so the result is reproducible.

Questions to ask before you use one

• Does it work with the software and file formats you already use?
• Can you see and reuse what it ran (Python, SQL, or tool parameters)?
• Does it ask before overwriting data or sending anything?
• Where is your data processed and stored, and can it run on your own servers if your clients require that?

Where Yerin fits

Yerin is the AI assistant for GIS work we’re building, starting with these everyday tasks: monitoring your AOIs, cleaning data, and turning results into maps and reports. It’s in development now, and we’re shaping it with the people who will use it.