User:ClaudineChionh/Guides/New editors and AI
 | This is an essay. It contains the advice or opinions of one or more Wikipedia contributors. This page is not an encyclopedia article, nor is it one of Wikipedia's policies or guidelines, as it has not been thoroughly vetted by the community. Some essays represent widespread norms; others only represent minority viewpoints. |
These are some of my late-2025 observations of how artificial "intelligence" impacts new Wikipedia editors – in particular, large language models that generate text, commonly known as chatbots due to their messaging-style interface.
What is Wikipedia's AI policy?
[edit]The English-language Wikipedia does not have a single all-encompassing policy specifying the limits of use of generative AI. Many experienced editors are personally opposed to generative AI, for a variety of reasons, and have developed an essay that you can read at WP:Large language models. We discourage using large language models to "write" articles or participate in discussions. Lying about using AI can lead to being blocked.
Why shouldn't I use a chatbot to help me write on Wikipedia?
[edit]The "chat" style of interface can be deceptive: you are not "chatting", but interacting with a machine, a bot. A machine cannot "think" and it cannot "know" anything. As the vision and purpose of Wikipedia is to provide every person with access to "the sum of all human knowledge", it is vital that this knowledge is summarised by humans, for humans. The work of writing an encyclopaedia is hard (but often rewarding), and there are no shortcuts.
More specifically:
- Do not get a machine to do work that you are not able to do yourself, as you are not able to check its work.
- Encyclopaedia articles are meant to be summaries of what reliable secondary sources say about a topic. Large language models are known to generate ("hallucinate") facts as well as references.
- Large language models are trained on internet content, including Wikipedia, but machines cannot "understand" this content. A machine can recall the names or wording of Wikipedia policies and guidelines, but cannot apply or interpret them.
- This training content includes both permissively licensed material including Wikipedia, and creative output that has been ingested without payment or consent of the copyright holders. Both of these can be problematic.
- The content generated by the models cannot be more reliable than what they have been trained on. We don't cite Wikipedia on Wikipedia. If a model generates a "fact" that might have been scraped from Wikipedia then it would also need to be supported by a reliable source. Why not skip this step and just read reliable sources from the start?
- Wikipedia cannot host copyrighted material, and individual editors may disapprove of the threat that AI poses to creators' livelihoods.
- Discussions on talk pages and noticeboards are how Wikipedia's human editors achieve consensus. A machine has no opinions. Experienced editors know when we are talking to a machine, and consider it deeply disrespectful when others respond to us with machine-generated text.
Why was I accused of using a chatbot?
[edit]Articles for Creation reviewers and new page patrollers have been struggling for years with growing backlogs, which have been made so much worse now that machines can generate slop with minimal human effort. Many reviewers and other experienced editors have become jaded and defensive, and sometimes too quick to judge questionable content. If you wrote your own draft or new article and have been accused of using a large language model, this may be because:
- It wouldn't be accepted in Wikipedia regardless of whether it was written by a human or a machine. There are specific policies about what topics can be included in Wikipedia and how we should write about them. The internet content that has been fed into large language models includes a lot of unsourced opinions, personal essays, advertisements/promotional/sponsored content, and low-quality filler – none of which belongs in Wikipedia.
- Reviewers can make mistakes! If you put in the effort to read real, reliable sources and summarise them in your own words, doing your best to write from a neutral point of view with an encyclopaedic voice, then please do start a conversation with the reviewer.
A bot gave me a warning!
[edit]Long-term Wikipedians have seen a lot of new editors make the same mistakes. We have developed a set of warning templates to provide standardised messages with links to the relevant policies and guidelines. If you are given one of these warnings, please do follow all the links so that you understand why you were given a warning.
Because these warnings follow a standard format, some new editors think they are automatically delivered by a bot. All approved bots that run on Wikipedia must have a username that includes the word "bot". There is a bot named ClueBot that identifies very obvious vandalism and spam and can deliver these warnings. All other warning messages are delivered by human editors who look at recent edits and decide whether they go against Wikipedia's policies or guidelines. If you have read the relevant guidelines and still don't know why you have been warned, you are welcome to start a conversation with the editor who warned you.
Are there any acceptable uses of AI on Wikipedia?
[edit]Generative AI is contentious, but it's not the only form of artificial intelligence. Wikimedia projects use AI systems for many purposes, the most visible one being ClueBot's identifying and reverting vandalism and spam.
If used carefully, large language models may be helpful for working with language (not knowledge or ideas). Checking spelling and grammar, and even autocorrect, are established and generally reliable tools that do not need to be disclosed. Some editors (including me) find it useful to check their writing with a proofreading or translation tool which might make use of language models. Any edits made with the assistance of large language models should be disclosed in as much detail as possible, by including the model's name and version in the edit summary. If you don't know how to identify the name and version of the large language model you use, then, to be blunt, you can't understand the LLM well enough to apply it to Wikipedia editing. Remember that experienced editors have seen a lot of slop in the last three years and many of us can recognise it without needing AI detection software. There is very little tolerance for inexperienced editors writing with LLMs, and those who lie about it usually end up getting blocked.