I'd make this shorter and more of a statement without opinion (e.g. is or isn't AI useful).

For example "If you use AI tools please state so in your Pull Request description."

Yes, I can make it shorter, but should this also become more of a "please do so", or keep the stricter tone? Or should we even be more explicit on the consequences, like in the paragraph above on fully-automatic submissions ("Maintainers reserve the right, at their sole discretion, to close such submissions")?

Not sure. I like being polite.

I also like to try and keep it short because no one reads anything, and the longer the text the less people read it :-/

Referring to an earlier version where you had a list of AI assistance that was used. stdlib uses a similar list (https://github.com/stdlib-js/stdlib/blob/develop/.github/PULL_REQUEST_TEMPLATE.md#ai-assistance) with checkboxes. Maybe a checkbox to state whether they used AI makes it more explicit/difficult to ignore. e.g., this is what a stdlib PR looks like:

I like the checkbox approach.
For reference, here is what I had initially (minus the correct formatting):

Please select one of the following:

• No AI assistance was used in the creation of this PR.
• I used AI assistance in the creation of this PR (specifically <ADD TOOLS/DETAILS HERE>),
  but I confirm that I checked and understood all changes and can explain them on request.
• This PR was created by an AI Agent.

The stdlib template goes on with a disclaimer section similar to what I added in the second option:

Disclosure

If you answered "yes" to using AI assistance, please provide a short disclosure indicating how you used AI assistance. This helps reviewers determine how much scrutiny to apply when reviewing your contribution. Example disclosures: "This PR was written primarily by Claude Code." or "I consulted ChatGPT to understand the codebase, but the proposed changes were fully authored manually by myself.".

{{TODO: add disclosure if applicable}}

I'm not sure what the best way here is to not add too much burden on contributors and maintainers side, but still to have some easy way of detecting non-compliant PRs.

Maybe we can start by copying just the upper part of the stdlib checklists and see how good Agents are at adapting to that?

Jumping in from the site with a comment 😅 : I think adding this to the PR Template is the wrong approach. I can really connect to the argument that we should not add additional burdens / bureaucracy to people who want to contribute nor on reviewers.

My hope on adding "disclose AI use" to our policy would be to easier tell people off that use AI in an irresponsible, harmful way, but I really don't care if someone has used a bit of AI or not at all and I don't want maintainers to be in the position to investigate what people claim to be doing compared to what they are doing.

In my opinion, informing people in contributing.rst that they have to disclose AI use on the PR, is enough. Hardly anybody will do that and that's fine. We don't want to discuss with ai-spammers whether they have pushed AI code they have not reviewed at all or whether they have only used AI as a helper.

If they fail to disclose their AI usage, we can then easily tell people who irresponsively open gen-ai PRs that they did two things wrong: 1. hardly or not supervised gen-ai PRs and 2. not telling us about it.

That puts us in a position where we don't need to hint people to the Automated Contributions Policy, because we expect them to be informed. This in turn makes it easier to deal with those cases. (I am speaking of the nastiest 3% of contributions and the rest of the other contributors would be untouched by that, no matter if they use AI as a helper or not.)

Yeah good thoughts.

Hardly anybody will do that and that's fine.

I think this is the situation that's not ideal, I have not seen anyone offer this info in a PR. What do you think of a simple checkbox yes/no for use of AI? It sort of forces a response as it makes it 'standard'. It would be nice to know before reviewing, and hopefully make people less reluctant to share - it is very commonly used for research/understanding (and even from a data collection perspective it may be interesting).

I see the value in having a clear signal from PR authors for reviewers. Sorry that I didn't reflect on that and only expressed concerns in my last message. I didn't mean it to sound dismissive, @lucyleeow and @AnneBeyer.

What do you think of a simple checkbox yes/no for use of AI?

I like it. Though posed like this, I am sure most people would have to check "yes". (I would certainly always do, since I constantly chat back and forth with llms to explain me the coding world.) What we mean is if they used AI as a coding assistance, I think.

What do you think of adding a simple checkbox "[ ] AI assistance used for coding?" without a "yes"/"no" option, only check if you have used AI for coding.

We could try it to collect some data on how people use it and if it is useful for reviewers and adjust later if it doesn't prove helpful.

I think, so far we have not asked people to disclose their AI usage (adding this to the guidelines is also part of this PR), so I'm not too surprised people don't do that yet.

I think we can go with a trial and error approach here. Adding the section heading to the template is a first step towards making it more obvious that this disclosure is expected. Adding a checklist could actually also make it less effort for both sides, because we might not get as detailed information as with a free text field (like which kind of tools people used), but people might be more likely to actually set a check mark than to fill in text. So we could go with something like this (and observe what AI Agents make of it for a while):

I used AI assistance for (please check all that apply)

• Code generation (e.g., when writing an implementation or fixing a bug)
• Test/benchmark generation
• Documentation (including examples)
• Research and understanding

What do you think? @lucyleeow @StefanieSenger

I'm fine with this, especially since it's important to others on the team.
I do still have concerns about adding an extra task for contributors while collecting information that may not be reliable but we can try it and adjust the PR template later if it turns out not to be helpful, or once we feel we’ve learned enough from it.

I'm +1 for this. I think a checkbox is easy enough to do, so I don't think it will be a burden in that sense.
But, I can understand that it would be a 'burden' because some people may feel like we would 'value' their contribution less if they say they used AI. As you said, we can always iterate.