The Descriptive Test Audience
Run 25 to 100 readers like a Hollywood test screening and ban them from suggesting fixes.
- Difficulty
- Moderate
- Time to result
- ~months to results
- Steps
- 7
- Confidence
- 92%
Sanderson treats books the way Hollywood treats films: with test audiences. He wants to know what his audience will say before release — often it changes what he writes, sometimes it does not, but he always wants to understand how it will perform. Most writers use a couple of early readers; he considers that insufficient to give a pulse and wants twenty to thirty, if not forty to fifty, drawn from a standing pool of around a hundred beta readers assigned per book. The critical instruction is descriptive, not prescriptive: don't tell me what's wrong or what to fix, tell me where you're bored, confused, excited, or turning pages so fast you don't want to stop and comment. The mechanism began in the late 1990s with physical manuscripts and coloured gel pens — each reader picked a colour, wrote their name in it, and responded to each other in the margins. That conversational structure is the actual innovation, because it lets agreement cluster: one reader says they started dragging here, another says me too, a third disagrees and explains, and the pattern surfaces. It moved to Google Sheets with a tab per chapter, then to a purpose-built tool pitched by one of his own beta readers, where readers add a check mark to agree and only write out disagreements. Finally, editors Peter and Karen condense the top 10-20% into inline annotations, so Sanderson reads end-of-part summaries and marked passages rather than the raw pile.
Origin
Sanderson started the practice in the late 1990s, before he had an agent or an editor, while running a local science fiction magazine — printing physical copies of his manuscripts and handing them round with packs of coloured gel pens so each friend's comments were distinguishable and could be replied to.
Core principles
- 01A handful of early readers gives you opinions; twenty to fifty gives you the pulse of an audience.
- 02You want descriptive response, not prescriptive repair — where readers were bored, confused, engaged, cheering.
- 03Readers should respond to each other, because agreement clusters are what reveal where the real problems are.
- 04The goal is to know how the work will perform, not necessarily to change it.
- 05Raw feedback must be condensed by people who know your taste before it reaches you.
- 06The condensers learn what you act on and what you ignore, and get better at filtering over years.
How to run it
- 1
Build a standing pool, sample per project
Maintain a pool larger than any single project needs — Sanderson keeps about a hundred — and assign a subset per book rather than always using the same faces.
Pro tip For the surprise Kickstarter he split the hundred into four groups of twenty-five and gave each a different book.
Watch out The same readers on every project drift toward predicting your intentions instead of reacting to the work.
- 2
Hit the size threshold
Use twenty to fifty readers on a given work. Sanderson's position is that a couple of early readers does not give an actual test audience or the pulse of one, even though fifty is still a tiny fraction of the real readership.
Pro tip Size is what converts individual taste into a distribution you can read.
- 3
Instruct descriptively
Ask only for experience: where were you bored, where confused, where standing up and cheering, where engaged and where not. Explicitly forbid fixes and diagnoses.
Pro tip His best-signal question is where readers were turning pages so fast they did not want to stop and comment.
Watch out Readers who propose solutions bury the actual symptom under their proposed cure.
- 4
Make the feedback conversational
Give readers a shared surface where they can see and reply to each other, chapter by chapter, so each chapter accumulates a discussion like a book club reading in real time.
Pro tip Sanderson's original version was coloured gel pens on a passed-around manuscript; the mechanism matters more than the medium.
Watch out Private, siloed feedback hides the agreement clusters that make a problem obvious.
- 5
Let agreement aggregate cheaply
Allow readers to simply tick agreement with an existing comment, and require writing only for disagreement. Sanderson's current tool works this way and also captures line numbers.
Pro tip Cheap agreement plus expensive disagreement produces a naturally weighted signal.
- 6
Have trusted editors condense before you read
Pay or assign people who know your taste to distil to roughly 10-20% and place it inline in the manuscript as tracked-change interstitials, so you see 'they said this at this point.'
Pro tip Sanderson's editors have worked with him since college; star the comments you found useful so their filter improves over time.
Watch out Reading the raw pile yourself is how writers end up either overwhelmed or selectively deaf.
- 7
Read the summaries, not the pile
Restrict your own reading to end-of-part and end-of-book general responses, plus the marked passages. That is enough to see the shape without drowning.
Pro tip Handle chapter-level speed bumps only where the annotation flags them.
Watch out Chasing every individual comment reintroduces exactly the noise the system exists to remove.
In the wild
In the late 1990s, before an agent or an editor, Sanderson printed physical copies and handed them out with a pack of gel pens. Each reader picked a colour, wrote their name in it so comments were attributable, read through, and was explicitly encouraged to respond to what other people had already written. Conversations formed in the margins about specific scenes.
→ The colour-coded margin conversation became the template for the spreadsheet and later the custom tool, and it is still the same mechanism twenty-five years on.
Sanderson cross-checks the panel against public Reddit threads on the chapters he serialises before launch, and finds they mirror the beta readers closely. On the most recent book only one signal appeared publicly that the panel had not caught: readers reacting against modernised language — 'okay' versus 'all right' — as he moves The Stormlight Archive toward a more modern register. He reads it as an appetite for sincerity rather than a craft error.
→ A near-perfect overlap between the private panel and the public audience, with one genuine blind spot found by cross-checking against an independent channel.
Common mistakes
Asking readers to fix the book
Prescriptive feedback replaces the symptom with the reader's amateur diagnosis. Sanderson wants the raw experience — bored here, confused there — because that is the data only a reader can supply.
Using two or three readers and calling it a test
A handful of readers gives individual taste, not a pulse. Sanderson considers twenty to fifty the minimum to see distribution, and even that a tiny sample of the real audience.
Reading the raw feedback yourself
Without a condensing layer run by people who know your taste, volume becomes paralysis or arbitrary selective attention. The 10-20% distillation into inline annotations is what makes the scale usable.
Is it for you?
Best for
Creators of long-form work with an existing audience and at least one trusted person who can filter feedback on their behalf.
Not ideal for
Early-stage creators with no audience to recruit from, or work so short that a large panel adds noise rather than signal.
From the transcript
“I look at books a little bit like Hollywood looks at movies with test audiences”
“I need like 20 to 30 if not 40 to 50 people reading it”
“I don't want people to fix the book I want people to give their descriptive responses to the book”
“don't tell me what's wrong don't tell me what to fix tell me what where you're bored and tell me where you're confused”
From the episode
#794: Brandon Sanderson on Building a Fiction Empire, Creating $40M+ Kickstarter Campaigns, Unbreakable Habits, The Art of World-Building, and The Science of Magic Systems