Genres and story forms

Historical fiction: research, voice, and the one true detail

August 16, 2026 · The StoryProp team

Historical fiction asks you to be two people at once. One is a novelist, who needs a scene to move and a character to want something. The other is an amateur archivist, who has spent four hours learning how a household lit a fire before matches were cheap and now, understandably, wants credit for it. The archivist has the louder voice. Research produces evidence, and evidence feels like an obligation — every fact you dig up seems to demand a place on the page.

It doesn't. Readers aren't grading your inventory. They decide, sentence by sentence, whether to believe the world, and belief is cheaper to buy than accuracy: it runs on a few well-chosen specifics and a voice that doesn't break. The working ratio is one true detail per scene rather than ten, and the working method is period voice built by subtraction, a named archive for each kind of question, and a confidence rating attached to every fact you write down. Most of what you learn will never appear, and that isn't waste — it's why the sentences that do appear sound like someone knew. What follows is the period-specific part of the job. Organizing research in general is its own subject; this is what working in the past adds on top of it.

Period voice is flavor, not transcription

The first instinct is imitation. You read letters from the era, notice the inverted clauses and elaborate courtesies, and start writing sentences that sound like a costume. That is pastiche, and it dies fast. Real period prose is dense, slow, and full of conventions that carried meaning then and read as noise now; transcribed faithfully, it sets a pane of glass between reader and scene. What works is closer to seasoning. Keep modern syntax, because the ear needs a rhythm it knows, and let a few period choices do the flavoring. Hilary Mantel's Wolf Hall is the standing demonstration: Tudor England rendered in plain present-tense modern English with almost no archaism in the sentences at all, the period arriving instead through what Thomas Cromwell notices, counts, and is afraid of.

The most efficient levers are contraction and address. A character who says “I do not” rather than “I don't” sounds a century older at no cost. A servant who says sir and gets no honorific back has established a whole social geometry in four words. Compare “I'm not going to lie to you, Tom — it's a mess in there” with “I will not lie to you, Mr. Ashe. The room is in a state.” The second isn't archaic. It's modern English minus the contractions, plus an honorific, plus one faintly formal noun phrase. Full immersion is the exception, and it works when you are copying one documented person rather than a whole era: Peter Carey wrote True History of the Kelly Gang as Ned Kelly's own unpunctuated letter to his daughter, modelled on Kelly's surviving Jerilderie Letter. A voice that exists in the record can be held for three hundred pages. A voice assembled from impressions cannot.

The complementary discipline is subtraction: removing words that are too new. Vocabulary from psychotherapy, corporate life, and systems thinking is the usual offender — a character who feels stressed, who needs to process a death, who calls a family toxic. These aren't errors of fact but errors of interior architecture. They imply a way of understanding the self that hadn't been built yet, and attentive readers feel the wrongness without naming it. Keep a short list of words this book doesn't use and what it says instead. If grief is a weight here and not a process, that has to hold in chapter three and in chapter twenty-nine.

Spend one detail, not ten

A scene with one precisely observed period detail is more convincing than a scene with ten. Ten reads as a museum placard; one reads as memory. The difference is what the detail does. Catalogue details are inert — they sit in the description, hold up a sign saying research was done, and vanish by the next paragraph. A true detail does work: somebody uses it, wants it, resents it, or is defined by their relation to it. The test is whether removing it would change anything. If not, it's decoration.

Picture a kitchen in a house with no electric light. The catalogue version lists the range, the copper pans, the scrubbed deal table, the pump at the sink, the meat safe. Accurate, and thin. The other version notices only that the woman of the house keeps a cracked cup by the stove for spent matches, because matches cost money and a half-burnt one will light a candle if you're careful. One object, and it tells you the household is poor, that she meets scarcity with system rather than worry, and that she's done it long enough to have invented the system herself. It also gives her hands something to do while she says a difficult thing. Penelope Fitzgerald's The Blue Flower opens on a great house's annual wash — linen coming out of the upper windows in armfuls, a visitor standing bewildered in the middle of it — and the eighteenth century arrives complete, through laundry.

Detail selection, then, is a character question in a research question's coat. Ask not what was in the room but what this person would touch, avoid, count, or be ashamed of. A soldier notices boots. A woman running a household notices what things cost. A physician notices smells. The period arrives through a nervous system rather than a floor plan, which is also what keeps description from stalling the scene. Spend your best details at thresholds — a chapter's opening, an entrance, a change of place. Readers calibrate trust at transitions and coast in between.

Anachronism triage

You will get things wrong; the field is too large not to. What matters is knowing which errors break trust, because that tells you where to spend limited checking. Errors of mind do the most damage. If a character in 1780 holds a moral position that only became thinkable much later, and holds it without cost or contradiction, the book has quietly become a costume drama about the present. Readers who know the period will close it. That category deserves real work — not because the past needs defending, but because a dissenter who pays no price isn't dissenting.

Errors of technology and consequence come next, and they matter because plots run on them: the invention that doesn't exist yet, the journey that takes two days instead of two weeks, the letter that arrives impossibly fast. A message that can cross a country overnight dissolves the suspense of a story built on distance. Get the constraints right and half your structural problems solve themselves. Get them wrong and the book stops being set in the world it claims. Check these before you outline, not after.

Language sits in an odd middle, and it is the one part of the job that can be checked mechanically. Three tools answer three different questions, and confusing them is where writers go wrong. The Oxford English Dictionary gives a first-citation date: the earliest use anyone has yet found in writing. That is a floor, not a start — speech runs ahead of print, and the OED's rolling revision keeps pushing individual dates earlier. Google Books Ngram Viewer plots how often a word appears across digitised books year by year, which is nearer the question you actually have: not whether the word existed, but whether it was ordinary. Read it with two cautions — it measures print rather than speech, and in text printed before about 1800 the long s is routinely scanned as an f, which makes the early curves unreliable. The Historical Thesaurus of English, compiled at the University of Glasgow, answers the most useful question of the three, because it groups words by meaning together with the years each one was current: not is this word too new, but what did they say instead.

Conspicuously wrong-era vocabulary is a real problem, but much of what readers call anachronism isn't — plenty of words feel modern and are old. The usable standard is perceived anachronism: if a word will pull an attentive reader out of the scene, replace it even when the OED says you could keep it. Errors of texture, meanwhile, are the ones to stop worrying about — the exact coach fare, the precise cut of a sleeve. Note the uncertainty, write the scene, verify later.

Research with receipts

Before any of that, know where to look, because the distance between a general history and a primary source is the distance between a summary and a voice. The Proceedings of the Old Bailey, free online, is the closest thing English has to a recording of how ordinary people spoke under pressure: witnesses, thieves, servants and shopkeepers, taken down in London's central criminal court from the late seventeenth century to the early twentieth. Digitised newspapers do the other half of the work — the Library of Congress's Chronicling America for the United States, the British Newspaper Archive for Britain. One week of a local paper hands you prices, weather, crimes, patent medicines and the rent on a house, which is a year of general reading compressed into an hour. Arrive with a question about a scene rather than a plan to learn the period, and you will walk straight past the wiki trap.

In period work, the thing to record is not only what a source said but how far it stood from the event. A ship's log differs from a memoir written at eighty about the same voyage, which differs from a history written a century later, which differs again from a film that dramatized the history. Each remove adds interpretation, and on anything interesting your sources will disagree. So attach three things to every fact: what it says, where it came from, and how sure you are.

That last field is the one people skip and the one that saves you. Six months on, staring at a note about how long a wound took to turn fatal, you won't remember whether it came from a period medical text or from a film you half-watched. Recorded confidence turns an unanswerable question into a ten-minute task. Draft with brackets as well: write a marginal note to check how long the crossing took, and keep moving. Verification and drafting are different states of mind, and doing both in one hour yields four hundred words and eleven open tabs.

Keep sources separate from story decisions. Research is what the world was; canon is what your book has committed to. The two diverge constantly and legitimately — you compress a siege, move a real building two streets, give a character a composite trade. Written down as decisions, those choices hold across four hundred pages; left in your head, they mutate, and by chapter twenty-two the building is back where history put it. That drift is the main failure mode of long historical drafts, and it's the same problem as keeping any long story consistent, with two sets of rules to hold instead of one. StoryProp is built to that split: the compression you settle on becomes a written record the agent works from, and anything you haven't confirmed stays a suggestion rather than hardening into fact behind your back.

Invented people among real events

Most historical novels place invented characters in a documented world, which raises a question of conduct. The workable line runs along proximity and harm. Invented characters can move freely: attend the battle, carry the letter, be in the room. Real people who lived should keep their documented actions, and what you're free to invent for them is interiority — what they felt while doing what they verifiably did. That is precisely the licence Marguerite Yourcenar takes in Memoirs of Hadrian: the emperor's public life stays as the record has it, and the novel supplies only the inside of that life, in his own voice. The pressure point is causation. If your fictional character causes a real event, you've edited the record rather than filled it in, and readers who know it will feel the hand on the scale.

Fictional people work better as witnesses, as the ones who bear the consequence, as small actors whose choices matter enormously to them and not at all to the record. That's also the more interesting position, since it's where most people live. The harder cases involve real people who suffered, and real people with living descendants. Inventing crimes or humiliations for a named historical figure is a form of accusation, and shelving it as fiction doesn't undo that. A composite — several documented figures merged into one invented character — is usually the honest solution.

When you compress, say so. A short author's note listing what you changed and why costs a page and buys a great deal of goodwill: it tells the reader you knew, which is the difference between a choice and a mistake. Declared departures can travel a very long way — Colson Whitehead's The Underground Railroad turns the metaphorical railroad into a literal one, and the book is stronger for announcing the invention than it would be for hiding it. All of it collapses into one discipline: the research exists to make the sentences confident, not to be visible. Hold that across four hundred pages and the past stops being a costume and becomes a place, which is the whole reward of the form.

A period novel asks you to keep two things straight at once for hundreds of pages: what the sources actually said, and what your book has decided to do about them. StoryProp saves research with its sources and dates, keeps that evidence separate from the canon and style rules you've confirmed, and reads both before it drafts a scene.

← All posts