Skip to content

AI in TTRPGs

Using AI for TTRPG Writing

TL;DR: AI is not creative; it's an echo chamber of compiled data, sifted through the lens of a user's prompt and powered by an engine that uses programmatic pattern matching. That said, it has become a dramatic force multiplier for solo creatives.

Our group, RPGOracle, is home to some of the most neurodivergent people I've ever encountered in life, all connected through meeting and pretending to be heroes in other worlds. It's found-family, deep friendships and a literal lifetime of commitment I will never regret. For many, the group predates marriages, kids and several other adult milestones.

The Facts

AI is present in every modern convenience we use on a daily basis. It started with spellcheck, evolved to grammar check and now it's sorting our email, building shopping lists and constantly gathering more data to further personalize the experience.

Google and Microsoft have a strangle hold on search engine ranking systems. If a site is indexed it's because these two companies provide the means. These companies are also harvesting their consumer facing "free" email and file storage services to train AI "with permission".

Our phones and every search solution are now powered or backed by AI engines learning from every query "with permission".

Even security conscious companies like Proton and Duck Duck Go have an AI component now. It's a measure of convenience vs privacy and convenience is winning. If you participate in the modern world, you're using AI daily.

RPG Writing

AI is a force-multiplier for a solo creative that has no resources, patience or guidance to refine a rough idea into a reasonably functional RPG resource. Should the idea be discarded simply because the author can't market it to a major brand or comprehend the processes required to develop it into a comprehensible final product?

  • Is a product resulting from AI involvement honestly any worse than the thousands of 3rd party supplements that flooded the d20 marketplaces from around 1995-2005?
  • How about those word-count constrained Dragon and Dungeon magazine articles?
  • Is it any worse than the official tripe that flows out of a publisher's own official pipeline because it's 'time for a new edition'?

If the argument is that all results from AI assisted creation is "slop", then you've obviously never read any alpha release or play-test products. Humans write just as much slop, most of the human written TTRPG slop is found in binders full of GM notes gathering dust because the writer doesn't have the time or resources to fill in all the blanks that refine and make it useful to other people. The "slop" that starts under a major publisher cycles through several internal processes to which most solo creatives have no affordable, reliable or recognizable access.

AI excels at grinding data into passably useful formats that can guide an inexperienced writer toward creating something closer to marketable. It can also provide a convenient sounding board to explore ideas between regular game sessions. It's like a room full of mediocre personalities that can offer research, editing, formatting and even creative prompts to funnel a single individual's concept closer to a final result.

Is it perfect? No. Is it missing soul? Possibly. Is the resulting content completely invalidated because of the obvious ticks and tells that AI is involved? No.

A Proposed Use-Case

AI is a powerful tool. Capable of collating and summarizing massive quantities of data from a variety of sources. This is particularly useful when you have an archive of 20-30 years worth of digital text focused on a specific topic like ttrpg homebrew.

The "creative" work has at least been started by humans. Years of ideas, plot hooks, theories, notes and campaign details across too many systems to easily count is ripe territory for rediscovering forgotten gems, patterns and ideas. This task is made easier with the right tool.

Working in this way is often referred to as RAG (Retrieval-Augmented Generation). This harnesses an existing LLM to access, retrieve and incorporate new data (your archive) as part of their response. It improves the accuracy and responses generated by the AI.

  • Google's NotebookLM is a RAG engine used by students that need to study focused topics.
  • Claude uses Projects to contain a library of files that focus discussions using RAG.

Cloud vs Local

Using a commercial cloud-hosted AI solution (CoPilot, Gemini, Claude, OpenAI, etc.) runs the risk of your data being converted into training fodder for the core data set. This means you're relinquishing 'ownership' of what you're feeding the engine and whatever results come from the interaction. That ultimately means if someone else is asking RPG creation related questions, pieces of your conversations could be replicated as part of a different project. This also means YOUR responses could be contaminated by someone else's content.

If you have the resources and the technical capacity, you could develop your own locally hosted AI Agent using something like OpenClaw or Hermes. This requires some significant hardware investment and time to evaluate and tweak the various technologies involved for efficiency. In theory, it keeps your data from flowing into the wider AI ecosystem. These agents still leverage basic web searches to be useful in research tasks but your data isn't released into the wild as training fodder.

Is it still possible to experience contamination/plagiarism using the tool this way, possibly. It depends on the model being used, what it was 'trained' on and any specializations that might have been involved.

Regarding Artwork

TTRPG art is a significant factor in product identity and first impressions.

TTRPG materials have always had a thriving ecosystem of recognizable staff artists connected with the products everyone associates with the hobby-space. There is absolutely no doubt that products from the 80's and 90's are almost universally identified by the artwork that created the visual landscape for the product lines.

Can a solo writer understand the nuances of finding an artist for exclusive artwork and all the processes required to make the art fit the project? Possibly. Are most going to settle for stock art from the open market thus diluting their product identity? Likely. Can producing original art be a realistic and affordable goal? Yes, with attention, patience and contacts.

Communication & Compensation

Artists are literally creating something from nothing. This creativity deserves payment because it costs time, requires creativity and a talent for translating your text into a scene or character. Do the research, find artists that can match your theme or vibe and assure you have a budget that isn't an insult. Be honest, let them know up front if you have no concept of what's reasonable for the art you want or if the planned scope is way out of line.

Can AI generate some impressive art? Sure. Is it going to get identified, picked apart and criticized thus limiting your products appeal? Most likely. Is AI art derived from ingested intellectual properties? Being able to prompt "in the style of xxx" would indicate to me that, yes, it actually is. Otherwise how would it know to whom you're referencing and how their particular style looks? Even AI generated placeholder art is considered a black-eye to your potential product if you release preview copies. Just create blank boxes to mark the space for editing layout.

If you pay for human created art, don't feed it to AI as source material for creating 'similar' spot art you suddenly need. In doing so, you've just provided a fresh sample of art, that you paid for, to an engine that will likely be used to churn out similar compositions. If you start with an artist, work with them to create your look and feel and keep the communication lines open for those last minute fillers that fit your existing theme.

RPGOracle, AxiomRPG & AI

We are players and game masters. We've all been playing longer than we'd like to admit most days. We all have day jobs or other 'real life' things that impact our time for creativity and escape.

We have crossed systems, editions and genres and settled into a rhythm of casual, bi-weekly sessions with all but one active player being over 50. We've realized that all those years of notes and creation amount to zero if the creator dies before reaching some formalization process. We've lost members to the passage of time. The only legacy they leave are scattered notes, stories of epic campaign moments and past settings.

RPGOracle, through AxiomRPG, is our attempt to formalize something that can be carried forward by anyone in our group since none of us are getting younger. It's a passion project for a bunch of neurodivergent "nerds" to leave a legacy.

We use AI to supplement the roles of planning, outlining and detailing the less exciting aspects of our settings and effectively applying the basic mechanics of our system. Are we shattering some barrier with our innovative ideas or planning on becoming filthy rich through this process? Absolutely not.

If you enjoy what we've created, cool. If you hate that we used AI to fill the gaps, cool. In the end it's not about appealing to the masses, it's about creating something a group of old gamers can call their own and enjoy.

The SRD site is not the raw dump it may appear. Translating our scatter-brained ideas into the framework provided by AxiomRPG for Genre Catalogs still requires real life effort. Literal hours at a keyboard reading and re-reading text. As we complete our internal check list we have gaps. We use AI to fill those gaps with starter content so we can remain focused as we work through the editing and re-editing processes before we play test something.

There is a constant process of offline editing and expansion happening as we play and adjust. AI provides an engine for our old and easily distracted brains to outsource the details and guidelines that make our madness more coherent to others.

Our Workflow

  • Original compiled archive data (word, onenote, treepad, realm works, campaign suite, notepad and transcribed voice or hand-written notes)
  • Convert date into a "working" Obsidian Vault
  • Outline, applying the Axiom structural framework
  • First round of AI RAG queries based on the original data
  • Creation of a development priority check list
  • Step through the checklist and develop human-based content to focus concept, direction and style
  • Present to AI to evaluate missing content and propose gap filler
  • AI interaction to explore how the concepts actually form setting lore
  • Revise, refine and rewrite gap content to better meet the intent and style of the author
  • Playtest, review, revise and re-write for more specific details
  • Convert content into static HTML using MkDocs
  • Add converted content to the SRD
  • Continue to review, revise and re-write.
  • Update MkDocs site as revisions are moved to stable.

Notice, there is no "package and sell this product for zillions of dollars" anywhere in that workflow. This is not a money grab, it is a hobby.

If we ever get around to compiling PDF versions of these settings, we may offer them for purchase to make pizza money for our games.