How should I fetch threads/posts from OSM Community Forum for a research project?

Help and support
1

Hi all,

I’m Gopi, a PhD student at New York University. I’m starting a research study that focuses on how OSM contributors discuss and debate the use of AI for mapping, tagging, and validation, and what those discussions reveal about the values and practices the community wants to preserve. Here is a poster I presented about the study at the State of the Map US 2026 conference last week in Madison, WI, USA.

For this research, I’m considering fetching publicly visible threads and posts from this forum (titles, post text, dates, tags, and engagement counts like views/likes/reactions) using the Discourse API.

My plan is to make read-only requests, include a clearly identified User-Agent with my contact, and rate-limit to roughly one request every couple of seconds. Discovery will be limited to the “ai” tag plus a defined keyword search list including names of AI-assisted tools/products (like Rapid or Microsoft Building Footprints), process-related terms (like AI-assisted tagging), and governance/policies (like Automated Edits code of conduct). Any published results will quote sparingly with attribution and pseudonymize usernames.

Does this seem like a respectful way to use the forum’s shared infrastructure? Is there anything I should do differently, or anyone I should check with first?

(Side request - if you know threads I should make sure to read to better understand discussions about AI in OSM, I’d be grateful for those suggestions).

Thank you so much in advance!

3 Likes
2

Awesome. I made it into your slide! (In the third column on the right!) :)

Since those May 2025 “AI Assisted” buildings posts, I’ve gone on a complete streak too.

Over these past 2 months, in the United States, I became:

  • #2 overall with “Created buildings”
    • 49,927 as of today, and no signs of stopping! :)
  • #7 overall with “Map changes”

These “AI buildings” save so much monotonous work with simple and super boring buildings—like garages/sheds (and trailer parks!!!).

These are just a whole bunch of small squares and rectangles.

And it’s exactly as I explained in that 2025 thread:

  • ~50% of the buildings are just Verify + Accept
  • ~25% of the buildings are just Accept, with minor adjustments.

This completely frees up time to then focus on all the “fun”/“harder” buildings, or adding many more micro-level details like building:levels, roof:shape, or entrances instead!

Note On Mass “AI” Slop

But I definitely noticed some serious QA issues where users are just mass adding clearly busted “AI” buildings.

For example, just this week, I ran across solid blobs, taking up an entire city block—where the townhouses and different color roofs clearly didn’t match, like these 2 cases:

OSM.-.Split.Microsoft.AI.Buildings.-.Very.Poorly.Accepted.Buildings
OSM.-.Split.Microsoft.AI.Buildings.-.Very.Poorly.Accepted.Buildings1240×1914 719 KB

OSM.-.Split.Microsoft.AI.Buildings.-.Before.and.After.Blobs

(I try my best to quickly “karate chop” buildings using JOSM’s awesome Alt+X shortcut, but even that can only go so far…)

But if the AI shape doesn’t match? Just move on! There’s no need to 100% accept every single little “AI building” you see! I say tackle the most easy+correct ones first, and then more complex cases can get filled in later.

One of the absolute best ones was 2024’s:

And if you want hugely negative “AI” topic, see March 2025’s:

Similar to what I said above, there’s absolutely no need to import 100% slop.

Just verify it, accept “the best ones”, and leave those “worst ones” for the “future you” to come back with even better tools/algorithms! There’s plenty of the Earth left to map! :P

Take care of all those perfectly square/rectangular/circular swimming pools first, then you could start importing all those wibbly wobbly blob-shaped ones later! Or better, why import complete AI trash when you could just drop a leisure=swimming_pool NODE on it instead‽

  • High confidence? Draw the way.
  • Low confidence? Drop a node.
  • Can’t verify it? Don’t add it!