Why StoryTrellis works the way it does

The method

Novelists have used spreadsheets to plan long fiction for decades. Writing guides regularly recommend Excel for outlining, and plenty of working authors keep their series notes in one.

The spreadsheet fits because a novel has two natural axes. Scenes run in one direction. The things you need to keep track of run in the other: characters, locations, objects, open questions, whatever matters to your book. Most of the continuity questions that come up while writing are lookups in that table. What does this character know by chapter 9? When was this object last mentioned? Which scenes touch this subplot? A grid answers those questions directly, in a way a folder of documents can't.

StoryTrellis starts from that observation. We didn't set out to replace the way writers already plan; we set out to build the best version of it.

The silo problem

The weakness of planning in Excel, or in any collection of separate documents, is that the pieces don't stay connected. The outline is one file. The character notes are another. Research sits in a bookmarks folder. Each document is accurate on the day you write it and drifts out of date as the manuscript changes, and nothing tells you which parts have drifted. Keeping them reconciled by hand is real work, and it's the kind of work that gets skipped once the drafting is going well.

A flat spreadsheet also does something less obvious: it makes your thinking more rigid than it should be. Every idea has to fit in a cell, at the intersection of one scene and one element. An idea about how two cells relate to each other, say a betrayal in chapter 3 that pays off a promise from chapter 1, has no cell to live in, so it ends up in a margin note or another document, or it gets lost. This is why StoryTrellis has its linking mechanism: cells, rows, and columns can be connected to each other directly, each link has a type you define, and each connection carries its own annotations. Relationships that cut across the grid's structure get recorded in the grid instead of around it.

The drift problem is handled the same way, by removing the copies. In StoryTrellis the grid is the only place information lives; the other views are derived from it. Every column produces a wiki page, with a structured overview and each appearance in story order, and every row produces chapter notes. These are views of the same data rather than copies, so they can't fall out of date. Research documents attach to the specific scenes they inform, and edits made in a derived view write back to the grid. Changing something once changes it everywhere it appears.

Why there is no AI

StoryTrellis has no AI features. It will not generate text, autocomplete a sentence, or suggest anything while you write. The reasoning is specific to fiction, and it rests on published research rather than a general objection to the technology.

Doshi and Hauser (2024), a randomized controlled trial published in Science Advances, found that giving writers AI-generated story ideas raised the quality of individual stories while making the full set of stories measurably more similar to one another. For most commercial writing, that trade is acceptable. For fiction it runs the wrong way, because a novelist's main long-term asset is a voice readers can't get anywhere else.

Jakesch and colleagues (2023) ran a controlled experiment in which people wrote with an AI assistant that held an opinion about the topic. Participants' own reported opinions afterward had moved toward the assistant's position, and most were not aware it had happened. The practical reading for writers: suggestions don't just save typing. They influence what you decide to say.

There is also early brain-imaging work pointing in a similar direction. Kosmyna and colleagues (2025) recorded EEG data from essay writers and found the weakest neural engagement, and the poorest recall of their own sentences, in the group writing with AI assistance. That study is a preprint with 54 participants and hasn't been peer reviewed yet, so we treat it as worth watching rather than settled.

The research also documents genuine benefits, and it would be misleading to leave them out. Noy and Zhang (2023) measured large, causal gains in speed and quality when professionals used ChatGPT for workplace writing tasks. We don't dispute any of that. Our position is narrower: those gains come from tasks where a standard voice is acceptable, and drafting fiction is not one of them. Writers who use AI elsewhere in their lives will get no argument from us; we've simply left it out of this tool.

In practice

Most writers who reach for AI mid-draft aren't asking a machine to write their book. They're trying to manage the amount of context a novel accumulates: who knows what, which threads are open, what happened forty scenes ago. Keeping that information organized and current is what the grid and the derived views are for, and none of it requires a model.

The drafting editor itself stays plain: rich text, a focus mode, and snapshots you can restore if a revision goes wrong, while the planning views hold the details of the story within reach. The user guide covers the mechanics in depth.