The WordPress AI Guidelines are now part of the official handbook. If you contribute code to Core, or plan to, this one is worth ten minutes. Pasting model output into a patch and hoping nobody looks too closely is not going to fly anymore, because the core team has written down what it expects on disclosure, licensing and quality.
None of this bans your favorite LLM. The rules are about accountability: whoever opens the pull request owns what is in it. I have spent enough time debugging hallucinated hooks to understand where that came from. If you want the longer argument about quality versus shipping speed, recent community discussions cover that ground too.
What the WordPress AI guidelines say
There are five principles. Most experienced developers already work this way, but until now none of it was written down anywhere official:
- Individual Responsibility: Your contributions are yours. AI can assist, but it is not a contributor, and if the code breaks it is your name on it.
- Meaningful Disclosure: Say so when AI helped in a meaningful way, either in the PR description or in the Trac ticket comments.
- License Compatibility: Everything has to stay compatible with GPLv2 or later, which means reading the terms of whatever tool you use.
- Holistic Guidelines: This covers more than PHP and JS. Documentation, screenshots, images and educational material are all included.
- Quality Over Volume: Low signal “AI slop” is fair game for rejection. Reviewers have been told they can turn work away when it misses the project’s quality bar.
Licensing, and the risk of “vibe coding”
Licensing is where this gets messy in the WordPress world. GPL compatibility is not optional. If your tool claims ownership of its output, or limits how that output can be redistributed, it cannot go into WordPress Core. So provenance becomes something you track rather than assume.
I have watched plenty of developers slide into vibe coding, where the syntax looks right so they trust the architecture behind it. That is usually how you end up with technical debt and race conditions that take days to track down.
Before your next pull request
The guidelines are meant to change as the tooling does, so read the handbook entry and send feedback. The team especially wants to hear from maintainers who are already reviewing AI-assisted contributions in volume. In practice the rest of it comes down to disclosing what the model did and reading your own diff before a reviewer has to.
