This article was AI-generated as part of an experimental historical-content project. The date reflects the period being analyzed rather than the date the article was originally written.
The Wikimedia Foundation published its AI strategy yesterday, and the opening is built for headlines. Someone asked when the Foundation would replace Wikipedia’s human-curated knowledge with AI. “The answer? We’re not.”
It is easy to read that as a conclusion for everyone who cares what Wikipedia says about them. Wikipedia is staying human, so the AI question is settled, and the old rules still apply.
I think that reading misses what the strategy actually describes.
What the Foundation committed to
The announcement, by Chris Albon and Leila Zia, lists four areas of investment. AI-assisted workflows for moderators and patrollers. Better discoverability of information so editors spend more time on judgment and consensus. Automated translation and adaptation of common topics. And guided mentorship for new volunteers. It also commits to prioritizing open-source or open-weight models.
The longer strategy document on Meta-Wiki explains the reasoning. It warns that people and governments can now “within minutes, generate thousands of Wikipedia-like articles” that may be hard-to-detect hoaxes, while “the verification of content has remained slow and costly.” So the Foundation chose to put AI into content integrity before content generation.
None of that reduces the stakes for a company with a Wikipedia article. In three ways, it raises them.
The bottleneck is verification
The strategy states plainly that new knowledge “can only be added to Wikipedia at a throughput that is defined by the capacity of existing editors to moderate that content.” That is the queue every company enters when it asks for a correction. Humans-first means the queue stays human. It does not get shorter because AI exists, and requests that arrive looking like machine-written advocacy are likely to get less patience.
Patrolling gets better tools
Wikipedia has used machine learning against vandalism for years. The strategy makes moderators and patrollers the first priority for new AI investment. If that works as intended, promotional edits, unsourced claims and undisclosed conflict-of-interest changes will be noticed faster and more consistently. For any company still tempted by the quiet edit, the odds are moving the wrong way.
Translation moves sentences between languages
Automating “the translation and adaptation of common topics” is aimed at helping smaller language editions, and the examples are core encyclopedia subjects, not companies. But the direction matters. As translation gets cheaper, an English framing of a topic is more likely to show up elsewhere, and multilingual reputation work has to watch for sentences that arrive rather than sentences that are written.
Why the human sentence matters more
The announcement contains one more line worth noticing: Wikipedia “is at the core of every AI training model.” That is the Foundation’s own description of its position. When ChatGPT began citing Wikipedia in 2023, I argued that the neutral sentence was becoming infrastructure. A humans-first strategy keeps the authorship of that infrastructure with volunteers. It does not remove it from the pipeline. If anything, a sentence written and defended by people becomes a more trusted seed for systems that cannot do the defending themselves.
Early last year the open question was who audits the sentences AI will repeat. The Foundation’s answer is clear: humans, with better tools.
For companies, the practical response is the familiar one, with less room for shortcuts. Bring reliable independent sources, disclose any affiliation, use the talk page, and accept that the process is slow by design. The strategy says nothing about entity risk getting smaller. It says the people managing it are getting better equipped.