The problem
AI writing tools produce generic output because they don't know you. Not your tone. Not your audience. Not the piece you published last week. Not the style guide you spent months refining. Built-in memory may keep some preferences, but it is not the same as checking your source notes before a draft.
How writers use Hjarni
Store your style guide, sample pieces, and research notes in Hjarni. Set AI instructions that define your voice: "Short sentences. No jargon. Address the reader directly." Once your assistant checks Hjarni before it writes, Claude or ChatGPT works from your actual writing context.
A typical writer setup
- Style guide folder: tone, voice, words to avoid, formatting rules
- Published work folder: samples that show what good output looks like for you
- Research folder: notes, interviews, source material for current projects
- Ideas folder: pitches, outlines, half-formed thoughts
- AI instructions: "Match the tone of my published pieces. Never use bullet points in drafts."
I'm using Hjarni to host something like .ini files for my Claude projects. I write a lot, and Hjarni holds a style guide of how to write like me, which Claude can fine-tune when it arrives at new insights about my style.
A concrete workflow
You're writing about remote team culture. Five interview transcripts are in Hjarni. You ask Claude: "Draft an opening paragraph based on the strongest quote from my interviews." Claude reads your notes. Picks the quote. Drafts in your voice, because your style guide and past work are right there.
Save the final version. Next time, your AI knows even better how you write.
Prompts like "Match my newsletter voice" or "Draft this section from my reporting notes" get much stronger when your assistant can revisit your style guide and source material every time.
Why not just paste your style guide into the chat?
You can. Every single time. Until you hit the context limit. Or forget the research notes. Save it in Hjarni and ask your assistant to check your style guide and source notes before drafting.
The difference between generic AI output and useful AI output is context. Your context.
Long stories: keep your canon in one place
Novels, series, fan fiction, tabletop campaigns. The longer the story, the more your AI loses track. Chapter twelve contradicts chapter three. A minor character changes their name. Worst of all, the AI confidently invents a detail that was never part of your world.
Keep the canon in Hjarni: characters, places, the rules of your world, and what has already happened. Then tell your assistant how to use it: check those notes before drafting, and save the new version whenever you agree on a change. The next chat can pick up the story as it stands, not what the last chat half remembers.
That instruction matters. ChatGPT and Claude only look in Hjarni when they decide to, so a line in ChatGPT's or Claude's own instructions is what makes checking canon a habit. Here's the one line to add, in ChatGPT or Claude.
It works across assistants too. Plot with ChatGPT on your phone, revise with Claude at your desk. Both read and update the same notes.
For tabletop games, start with your latest recap, current party details, and unresolved quests. Ask your assistant to propose updates after play, then check the saved state before the next session. The D&D campaign notes guide includes copyable prompts and a worked example of an item changing hands while the session history stays intact.
A typical story setup
- Characters folder: one note per character, with appearance, voice, relationships, and how they change
- World folder: places, history, organizations, and the rules that can't be broken
- Timeline note: what happened, in order
- Chapters or sessions folder: a short summary of what each one established
- AI instructions: "Check Characters and Timeline before drafting. Point out anything that contradicts them. Ask before changing canon, and save the change once we agree."
A good first test: ask your assistant about an obscure detail buried in your notes, or whether a new chapter fits what came before. You'll see quickly whether it answers from your canon or from guesswork.
I use Hjarni to keep track of the characters, events and background details in a story I'm writing. I've given both ChatGPT and Claude access to it so they can check and update the same notes. I'd tried several ways of handling this before. Too much time went on uploading files and explaining things we'd already worked out. Hjarni has cut down on that and made it easier to carry on when I start a new chat or switch between assistants. I tested it before upgrading, then signed up straight away. I'd recommend it to anyone using AI for an ongoing writing project, especially if they use more than one assistant.
Jim Johnston uses this loop for a long writing project: ChatGPT checks the established story before exploring an idea, then saves their decisions back to Hjarni. Read Jim's full workflow.
What writers keep in Hjarni
- Voice and style guides: tone, rhythm, formatting rules, and phrases to avoid
- Reporting material: interviews, background research, quotes, and source notes
- Story canon: characters, places, timelines, and the rules of your world
- Published examples: finished pieces that show your assistant what good looks like
- Project-specific rules: one set of instructions for newsletters, another for essays or client work
- Idea capture: pitches, scraps, and outlines before they turn into drafts
- Portable drafts and notes: Markdown-first writing you can always export