AI in tabletop RPGs
Leave the imagination to your table
The part of tabletop roleplaying I care about happens between people: a player’s unexpected idea, a difficult choice, a character who becomes more interesting than anyone planned. That is where I want our attention to be. I use AI to help with the work around the game, so there is more room for the people playing it.
That distinction shapes how I build Recap Raven. Remembering a promise from six sessions ago is useful help. Deciding what a character should do, what their story means or how an artist should portray them belongs with people. The hobby needs its writers, artists, designers and players, and using AI should leave room for their work to be valued and supported.
Recording starts with a conversation
Before the first recording, talk with everyone at your table about what Recap Raven will process, who will see the recap and whether anything will be shared publicly. Get everyone’s explicit permission, and revisit that conversation when a new player joins or your sharing plans change. Nobody should discover afterwards that their game was recorded.

Our Discord bot posts a visible recording notice before it starts capturing audio. The notice explains that recording creates a transcript and session recap. Where a campaign uses a public Discord forum for delivery, it also explains that the recap and transcript will be visible there. The recording controls let the GM pause or stop capture.
A bot announcement supports that conversation; it cannot obtain permission on a GM’s behalf. Our Terms of Service require the person recording to obtain permission from every participant. We encourage GMs to make recording consent part of their table’s usual check-in.
Be clear about what AI produced
A generated recap is an interpretation of the session. It can mishear a name, miss context or draw the wrong conclusion. We explain those limitations and provide ways to review the result. Campaign answers include source references where supported, so you can revisit the session behind an answer rather than relying on how confidently it is phrased.
See our approach to accuracy for how we evaluate quality and make campaign memory easier to check.
Your table keeps the final say
We deliberately keep preparation suggestions high-level: reminders of what your players hinted they wanted to do next and threads you may want to revisit. How you build the encounter and respond to their choices stays with you. That distinction matters to us. Helping you remember what needs preparation is useful support; telling you how to run it oversteps the GM’s role.
Human creativity and artists
We value the writers, illustrators and other creators who make this hobby what it is. Our aim is to free up time for playing and creating. An artist brings their own judgement and interpretation to your world; generated imagery does not replace that relationship.
We use AI-generated illustrations on this website, in demonstrations and in Campaign Showcase. Supporting artists does not mean we avoid generated imagery altogether. The boundary for us is creative control: an illustration should not establish what a character looks like or add facts to your campaign. Those choices belong to your table. Generated artwork serves as a visual accompaniment, not a record of what happened or a substitute for an artist’s interpretation of your world.
How campaign artwork is created and handledUse the information to start a conversation
Our session analytics can help you notice who has room to contribute or where the pace changes. A quiet player may be thoroughly enjoying the game. Talk with your group before deciding what a chart means for them.
TTRPG Table Insights extends that curiosity to the wider hobby through anonymous aggregate statistics. We want to help GMs compare experiences and find ideas for their own games. There is no percentage that tells you whether everyone had fun.
Take responsibility when something goes wrong
We use testing, monitoring and support reports to find problems and improve the product. That includes checking access boundaries and player-safe outputs, alongside whether a feature works. Our engineering approach explains those checks. If you spot an incorrect result or have a concern about how your content is handled, please join our Discord community and help shape the next version of Recap Raven.
Many of our features have come directly from listening to our users. We’d love you to join those conversations, share feedback and help fellow GMs in the server.
