- We tested AI-assisted translations on five wikis to explore whether localized content could help reach more fans globally.
- Results were mixed, and we learned a great deal about timing, context, communication, and where AI can (and can’t) help.
- One pilot (Clair Obscur: Expedition 33) showed strong results when timing and relevance aligned.
- One community chose to opt out, and we honored that request.
- We’re ending this pilot in its current form, applying what we learned - especially around hreflang and discovery - to better support existing international communities.
- We’ll continue using AI where it meaningfully improves the fan experience, including active testing in search.
Hi everyone,
When we launched our LLM-assisted translation pilot earlier this year, our goal was simple: to explore new ways to bring fandom content to more fans around the world. The idea, helping fans read and contribute in their own languages, still feels right - and we want to find new ways to continue exploring. However, as the pilot progressed, competing priorities around platform stability and urgent updates required more focus than anticipated, and our efforts and communication around this pilot suffered as a result.
We heard the confusion, the concerns, and the frustration, and we’ve taken that feedback seriously. This post is about what we learned, what worked, what didn’t, and how we’re using those learnings moving forward.
What We Learned[]
This pilot reinforced something we don’t want to underestimate: how we communicate matters just as much as what we build.
- Context and intent need to be clear early, especially when experimenting with new technology.
- Human review and editorial judgment are still essential. Transparency builds trust, and when it’s missing, people understandably fill in the gaps with concern.
- AI can be a powerful tool when applied thoughtfully, but using it well requires clear understanding of its strengths, limitations, and the context it’s being asked to work within.
We are sincerely grateful for all feedback received, from those who sent a 'thank you for trying' to those who provided tough, constructive criticism. While supportive messages validate that experimentation is how we learn, the fuel for improving the platform, your concerns were equally vital in shaping our path forward.
The Five Pilot Wikis[]
Below is a summary of each pilot wiki, why it was chosen, and what we learned.
Shōgun Wiki (Fandom-created)[]
Shōgun was the first wiki we worked on and was created by Fandom staff, allowing us to test the full workflow without impacting an existing editor community.
What we learned Because Shōgun is grounded in real historical context, the AI was able to start the translation with more accuracy using established historical references and terminology providing clearer signals for the model. This helped to streamline the human review for clarity, consistency, and handling in-universe interpretation layered on top of real history.
Barney Wiki[]
The Barney wiki was community-seeded and did not have an existing German-language version.
What we learned The AI struggled with imagined concepts and playful language central to this fandom. The translated output required more correction than was reasonable for a quality-focused pilot, so we chose not to launch it. This was a great illustration that AI-assisted translation works best as a starting point, and that some fandoms require significantly more human review and editorial involvement to meet quality standards.
Silo Wiki (Fandom-seeded)[]
The Silo wiki was seeded by Fandom and had no German-language version at the time of the pilot.
What we learned While closer to Shōgun in structure, Silo lacked the same historical grounding and would have benefited from stronger international glossaries. More importantly, it demonstrated that timing is just as important.
Silo Season 2 premiered on November 15, 2024, while the German-language wiki launched on August 16, 2025, well after peak audience interest. Engagement was limited at launch as a result. We’ll continue to monitor this wiki as the show approaches its next season.
Clair Obscur: Expedition 33 Wiki (Fandom-seeded)[]
We aligned translation timing closely with the game’s U.S. release. Interest later surged again during the recent Game Awards, where the game received multiple accolades.
What we learned During this period, the English-language wiki saw approximately a 60% increase in pageviews, while the German-language version saw a 58% increase. The wiki also brought roughly ~6K new users to Fandom.
This pilot validated our use of hreflang tags, demonstrating that translated content can rank effectively in specific territories without negatively impacting English-language performance.
Clair Obscur: Expedition 33 showed us the clearest example of the localization project’s potential, as fans of the game in both languages were able to enjoy their experience on Fandom.
Community-Created Wiki [Name intentionally withheld][]
One community-created wiki began translation after approval but opted out before launch following broader community feedback.
What we learned Editor and community buy-in is invaluable to building new fandoms, and in reaching new fans. We honored the opt-out immediately. The translated content was not published, and QA work stopped at the point of the request.
Key Takeaways[]
- Human review matters: AI-assisted translation can build a helpful foundation, but every effort requires editorial judgment and guidance, to ensure the best fandom representation.
- Community matters: Building by, for, and with the fandom community is crucial for every fandom, in every language, by every methodology.
- Timing matters: Localization is more effective when aligned with major fandom moments.
- Context matters: Historical or well-defined source material translates more reliably than highly imaginative content or imagined worlds (which is quite a bit of the Fandom universe).
- Discovery matters: hreflang validation showed clear potential to better surface existing international communities.
What Happens Next[]
With our pilot concluded, we’re looking to best apply our learnings.
For example, leveraging insights about hreflang tags to help existing international-language communities be more discoverable. If the timing is right and we believe the audience would genuinely benefit, we may also apply localization approaches to content created and maintained by Fandom itself, informed directly by these results.
More broadly, we’ll continue to test and use AI where it meaningfully improves the fan experience. One area we’re actively exploring is how to leverage AI to help fans find relevant content within the site, through better, more intuitive search results. As we know that you will be curious to know more about this, we will update you via a technical update in the coming year.
What’s important to emphasize is this: Editors remain the heart of Fandom. That will not change. Thank you to everyone who shared feedback, raised concerns, and encouraged us to keep trying. Your input helps us learn, iterate, and build a Fandom that works better for fans, creators, editors, and admins everywhere.
Click here to follow the Fandom staff blog.
Click here to sign up for the From the Desk of Community email newsletter.
Join our Official Discord server for registered editors!