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.
A common view in tech circles is that Wikipedia has quietly become infrastructure for machines. AI models train on it, search engines and assistants quote it, and the people who write it are an afterthought. On that view, the volunteer community is a charming detail.
Wikipedia’s own fundraising team has been running an experiment this autumn that points the other way.
What the banner tests show
Every year the Wikimedia Foundation runs a large banner campaign on English Wikipedia. This year’s full campaign starts on December 2 in Australia, Canada, Ireland, New Zealand, the UK and the US, according to Wikimedia’s Diff blog. Before it launches, the team tests messages and discusses results with volunteers on a public planning page, Wikipedia:Fundraising/2024 banners.
That page is unusually candid. In July, a Foundation staffer wrote that AI-related messaging had “worked really well” in last year’s campaign. The page lists the “current best” banner from last December’s campaign, which says that “in the age of AI, access to verifiable facts is crucial” and describes Wikipedia as powering “everything from your personal searches to emerging AI technologies.”
But the slogans did not win. The team said it had tested more than 18 headlines, including AI-themed ones such as “Wikipedia is still human,” and none resonated as well as the long-running “Wikipedia is not for sale.” In a September update, newer headlines about trust, such as “Amid a crisis of trust, Wikipedia stays reliable,” also failed to beat the incumbent.
What did perform was description. The same update reported a strong result for copy explaining that “countless volunteers work tirelessly to create and verify the pages you rely on.” And the headline showing promise in a November 20 update, “The internet we were promised,” comes with body text about volunteers creating and verifying pages, “supported by tools that undo vandalism within minutes.”
One more result is worth noting. A minimal banner with just a donate button, often suggested by editors, produced a 95% drop in donation rates compared with the standard banners.
Process beats claims
The caveat is obvious: these are donation tests, not trust surveys. A banner that raises more money is not proof of what readers believe. Still, the pattern is consistent. Asserting that Wikipedia is human, or trustworthy, does little. Explaining how it is made, who writes it, who checks it and what happens to vandalism, does more.
That matches how reputation tends to work in general. Claims are cheap. Visible process is persuasive.
Why reputation teams should care
The reason Wikipedia sentences end up in Knowledge Panels, AI answers and journalists’ background research is that the text is assumed to come from a process: sourced, contested, reviewed, reverted when wrong. As I wrote earlier this year, the open question is who audits the sentences AI will repeat. The banner tests suggest that the answer, the human editing community, is also central to why the public gives Wikipedia credit in the first place.
So the myth has it backwards. The more AI systems repeat Wikipedia, the more value hangs on the human process behind it, not less.
For companies and executives, there is a practical implication. If the authority of a Wikipedia article comes from its process, then the only durable way to change the article is through that process: reliable independent sources, disclosed conflicts of interest, requests on the talk page. Shortcuts around the process do not just break rules. They attack the very thing that makes the article worth caring about.
Wikipedia’s fundraisers have spent the autumn learning that “still human” works better as a demonstration than as a slogan. That is a useful lesson for anyone trying to be believed online.