How to turn family interview transcripts into a book

By The Yourtale team · Published 29 September 2026 · 16 min read

To turn family interview transcripts (or any oral history transcripts) into a book, index every story on a timeline first, because the timeline is your chapter skeleton. Choose a structure that fits the material: chronological for one life, by generation for several narrators. Then edit each transcript into prose that uses only the speaker's own words, add short italic lead-ins for context they assumed, and gather retold memories under one theme.

A transcript is not a draft. It is a record of two people talking, and talk is built differently from writing. The story arrives out of order, the best detail turns up three questions later, and a third or more of the lines on the page are the listener saying "mm-hm". The job between the transcript and the book is mostly reordering and removing, very little adding, and never inventing.

Most of the pages that rank for this tell you to "organize chronologically or by theme" and "edit for readability", then move on to printing. They skip the part where people actually get stuck: how to build the timeline, how to decide between a life and a family, and how much you are allowed to change someone's words before they stop being theirs.

We built a transcript import for our own books, and building it forced us to answer those questions precisely. This guide is those answers, for doing it by hand, plus a plain account of where our import helps and where it does not.

If you only have recordings so far, with no transcript yet, start with how to turn old voice recordings into a written memoir, which covers digitizing and transcription. This piece picks up where that one leaves off: text in hand, book not yet.

Key takeaways

• Read every transcript once, all the way through, before you change anything. You cannot see the shape of the book from inside one session.
• Build a timeline with one row per story: session, time in the recording, approximate year, age, place, people, theme. The rows are the raw material of your chapters.
• Pick the structure from the material, not from an ideal. One narrator with dateable stories reads best in chronological order. Two or three narrators from different generations read best in parts, one per generation.
• Every sentence in the speaker's voice must trace back to a line in the transcript. Context they assumed goes in brackets or in a short italic lead-in in a clearly different voice. The Ohio History Connection's guide puts the limit at "a word or sentence here and there", not paragraphs.
• Cut the interviewer's "mm-hm" and rejoin the stories it split. In the first long interview we ran through our import, 52% of the interviewer's lines were nothing else.
• When a story was told twice, pick the fuller telling as the base. When two versions disagree, keep both and say so.
• Whatever does the cutting, a person or software, check what was left out. On its first pass over one interview, our own software placed only half of what the subject said.

Before you edit a single line

Gather everything into one place and read it through once without editing. The first read is for the shape of the whole: which periods of life are rich, which are missing, which stories come back in several sessions, and where the speaker is at their best. You cannot see any of that from inside one transcript.

Before that read, three pieces of housekeeping save hours later.

1. Copy the originals and never edit them. Work on a copy of every transcript. The original is your source of truth for the rest of the project, and you will go back to it.
2. Give each transcript a short code and put the date in the file name. S1 for the first session, S2 for the second, and 2024-03-10_S1_mum_kitchen.txt as the file name. Every note you make from now on refers to a session and a time: S2 14:30.
3. Keep the audio next to the text. If the recordings exist, keep them in the same folder. The edit is checked against the audio, not against your memory of the conversation, and a transcript without its recording hides every mistake the transcriber made.

While you read, mark only three things in the margin: stories (anything with a "one time" or "the day that" in it), names you need to check, and passages that are unclear. Save the editing for later.

How do you build a timeline from oral history transcripts?

With a spreadsheet and one row per story. The timeline is not decoration for the book. It is the skeleton the chapters hang on, and it is the tool that shows you, before you write anything, what the book can and cannot be.

Use these columns:

Most stories do not come with a date, so date them the way a historian would, from anchors:

• Ages. "I was fifteen" plus a birth year is a date.
• Life events. Which child was born yet, which house they lived in, whether they were married. "We were still in the flat" puts a story before the move.
• Public events. A war, a flood, a new king or president, the year television came, a famous winter.
• Documents. Certificates, school records, census entries, dated photographs, letters.

When a story cannot be dated exactly, place it between the two anchors either side of it and write "c." in front of the year. The last column is there so that anyone checking the book later can see which dates are known and which are inferred. Memories are not infallible, and the timeline is where a wrong date gets caught before it is printed.

Once every story has a row, sort by year and look at the result. You will see a few decades thick with stories and whole stretches with one line each. That unevenness is the most important fact about your book. The rows for a thin decade will not become a chapter no matter how you edit them, and the rows for a thick one may be three.

Should the book be chronological or organized by generation?

Chronological if the transcripts are one person's life. By generation if you interviewed two or three people from different generations. Thematic if the material is rich on a few subjects and thin on dates. The timeline tells you which one you have.

Two variations are worth knowing about.

Braided chapters. When two narrators describe the same event, such as a mother and a daughter both remembering the move to the city, you can put their accounts in the same chapter, one after the other, each marked with the speaker's name. It is the most vivid structure a family book can have, because the reader sees one day from two ages. It only works where the transcripts genuinely overlap. Do not force it.

A section at the back for what fits nowhere. Every set of transcripts has good pieces with no home: a paragraph about a dog, a line about a boss, a song. Keep them together in a short final section rather than cutting them. They are still in the speaker's voice.

Whatever you choose, let the chapters follow the material. A book of eight chapters where the transcripts are rich beats eighteen chapters where ten are padding.

How do you edit a transcript into readable prose without inventing anything?

Remove and reorder, add almost nothing, and mark what you do add. Every sentence in the speaker's voice should trace back to a line in the transcript. The Oral History Association asks everyone who uses interviews to avoid "manipulations of the narrator's words" and to strive "to retain the integrity of the narrator's perspective" (OHA best practices). In practice that comes down to six moves, in this order.

1. Take the interviewer out. The questions and the "mm-hm"s were there to keep someone talking, and on the page they break the story into pieces. Baylor University's Institute for Oral History warns against exactly this: "If every few lines of the transcript is interviewer feedback, go back and carefully evaluate the merit of each instance" (Baylor style guide). When we built our import, we measured it on the first long interview we ran through it, a conversation of about 53 minutes between two people. More than half of the interviewer's lines, 52%, were nothing but that kind of feedback.

2. Rejoin the stories the listener interrupted. Once the "mm-hm"s are gone, a single story that was split into three or four fragments becomes one passage again. In that same interview we removed 567 bare feedback lines, merged the runs of lines by the same speaker that the feedback had separated, and stitched back 71 sentences cut off mid-clause. The transcript went from 1,439 separate turns to 448 coherent ones. Nothing was rewritten to get there. That is the scale of the change you get from removal alone.

3. Thin the filler, keep the voice. Baylor's rule is to keep "no more than two crutch words per occurrence per page", and to take "a middle course" on false starts, "leaving in enough to indicate individual speech patterns." Repetition that the speaker used for emphasis stays, always. "Four in the morning, you know, four" is not a stutter. It is how she told you it was early.

4. Add context in brackets, sparingly. If the reader needs a word the speaker did not say, such as who "he" is or what an abbreviation means, add it in square brackets so it is visibly yours. The Ohio History Connection's guide sets the limit: "a word or sentence here and there is appropriate while full paragraphs are too much" (Ohio History Connection, 2019). Anything longer belongs in a lead-in (see below).

5. Correct facts in the open, not silently. People misremember names and years. The same guide says: "Remember peoples' memories aren't infallible. If you find a mistake write the correction next to it in brackets." Silently changing the sentence makes the speaker say something they never said. A bracketed correction keeps both the memory and the fact.

6. Reorder within a story, never across stories. Moving a detail from a follow-up answer up to where it belongs in the same story is fair. Stitching sentences from two different conversations into one unbroken quote is not, because the reader can no longer tell that two tellings were combined.

Here is what that looks like on a short, invented but typical passage.

The transcript:

Q: So where did you work after school? A: Well, the bakery. Hansen's. On, um, on Storgata, no, on the corner, it was on the corner of Q: Mm-hm. A: of the square and Kirkegata. I was, what, fifteen. Fifteen. And you started at four. Four in the morning, you know, four. Q: Wow. A: And old Hansen, he, he would, he never said good morning. Not once, not in three years. He'd just point at the flour.

The edit:

Well, the bakery. Hansen's, on the corner of the square and Kirkegata. I was fifteen. And you started at four. Four in the morning, you know, four. Old Hansen never said good morning. Not once, not in three years. He'd just point at the flour.

The questions and the listener are gone. The false starts ("he, he would") and the self-correction about the street are trimmed to the version she settled on. "Four in the morning, you know, four" stays, because it is emphasis. Not one word was added. Notice what the edit did not do: it did not write "After school, I went to work at the bakery." That phrase came from the question, not from her, and putting it in her mouth is the smallest possible version of inventing a quote.

How do you turn a stretch of talk into a scene?

A scene is one place, one time, and one thing that happens. Transcripts are full of them, but they rarely arrive in order. The setting comes out in answer to a follow-up, the ending gets told first, and the best detail turns up ten minutes later when something else reminds the speaker.

To build a scene, collect every piece of the story from the same session and put the pieces in the order things happened: where it was, who was there, what happened, what happened next, how it ended. Use the timeline to find all the pieces. Then read it aloud against the audio. If it sounds like the same person telling the same story, only more clearly, the scene is done.

Three rules keep it honest:

• Only their details. If she never said what the weather was, the scene has no weather. A scene with invented sensory detail reads well and is fiction. If you want more detail and the speaker is alive, ask them. That is a better scene and a better afternoon.
• Dialogue only as they reported it. "He said, you'll be back" is quotable if she said it. A conversation reconstructed from "we argued about it" is not.
• Leave the seams where the tellings differ. If the same scene came up in two sessions with different details, use one telling and put the other after it with a short line such as In another conversation, she remembered it differently:. That line is a lead-in, which is the next tool.

What is a third-person lead-in, and when do you need one?

A lead-in is one to three sentences in italics, in a narrator's voice that is clearly not the speaker's, placed before a passage to give the reader what the speaker assumed they knew. It is the place for dates, places, who people are, and what happened between two stories. It is not the place for feelings, interpretation, or anything that did not come from the transcripts or a document.

For the bakery passage above, the lead-in might be:

Autumn 1953. Kari is fifteen, has left school, and starts work at Hansen's bakery on the square. She stays three years.

Every fact in it comes from the timeline: the age and the three years from the transcript, the year from her birth year, the season from a later session. It reads differently from her voice on purpose, so a reader always knows whose words they are reading.

Use lead-ins in four places:

1. At the start of a chapter, to set the year and the place.
2. Where the speaker assumed knowledge, such as a name the reader has not met yet or a place that no longer exists.
3. Between stories that are years apart, so the jump does not confuse.
4. Where the record has a gap. She never said, in any of the sessions, why the family left the farm. That line is honest, and it tells a grandchild what to ask about while there is still someone to ask.

Write them in the present tense if you want them to feel like stage directions, or in the past tense if you want them to feel like a history. Pick one and keep it for the whole book.

Memories scattered across several sessions

Group them by theme using the timeline, then decide for each story whether it was told once, told twice, or told differently. People return to their important stories in almost every session, and a book that includes every telling reads like a repetition.

Sort the timeline by the theme column and you get every piece of "the farm" or "the war" in one place, whichever session it came from. For each story:

• Told once: use it.
• Told more than once, consistently: use the fullest telling as the base. If a later telling adds one good detail, you may bring it in with a lead-in (Years later, she added:), not by weaving it silently into the first.
• Told differently: keep both versions and say so. The OHA's guidance lists "recognizing the subjectivity of the interview, including, when possible, verification of information presented as factual" among the duties of anyone using an interview (OHA best practices). Check what documents can settle, such as a date or a spelling, and leave the rest as two memories. Two accounts that disagree tell a reader more than one smoothed account.

If the transcripts are from more than one person, the same method works across narrators: the "wedding" rows from the mother and the daughter end up next to each other, and that is often where a braided chapter comes from.

Which route is right: doing it yourself, or a supported import?

Doing it yourself costs time and nothing else, and everything above can be done with a word processor and a spreadsheet. The edit is the slow part. Expect it to take many times longer than the conversations did, because every passage is read, cut, checked against the audio and read again. A supported import removes the mechanical part of that work. It does not remove the judgment, and it cannot remove the parts that need knowledge of the family.

Doing it by hand fits if you have one or two transcripts, you enjoy the work, the person has died and you want to spend time with their words, or you want full control of every line. The only tools are the ones in this guide.

What our import does

In a Yourtale workspace, you upload a transcript to the book's library: a plain-text transcript with speaker names, an export from a meeting tool, or a WebVTT or SRT caption file. Then:

1. It works out who is speaking, and assumes the person who spoke most is the subject.
2. It removes bare acknowledgements ("mm-hm", "yes", "okay") from either speaker and rejoins the stories they split, without changing a word of what it keeps.
3. It splits the transcript into titled sections for one chapter.
4. Before anything goes into the book, it shows you how much of what the subject said made it into a section, and lists every stretch it left out with its opening words. You choose "Add this to the book" or discard it.
5. An AI drafts the chapter from the sections, and you edit that draft yourself.

There is no human writer at Yourtale and no editorial team, and no employee of ours reads or edits your book. You are the editor. If the person is still living, our interviewer's next conversation with them starts from what the imported transcript already covers, rather than asking it all again. Importing a transcript does not use any interview minutes.

That review screen exists because of our own first run. The first time our software split that 53-minute interview into sections, the sections covered only half of what the subject had said, and a childhood memory was simply missing. We changed the instructions to keep material by default and to report coverage, and the same interview came back at 88%. The lesson holds whether a person or software does the cutting: look at what was left out before you accept the rest.

What it does not do

• It is not quick yet. The work runs in the background, and today an import can take up to two days to reach the review screen.
• It does not handle a transcript without speaker names. Add a name before each turn first, the way a meeting tool exports it.
• It does not transcribe audio yet. A recording you upload is stored, but you need a text transcript to import.
• It does not read Word documents. Save a .docx transcript as plain text first.
• It does not build your timeline, date your stories, or write lead-ins. The structure decisions in this guide are still yours.
• It makes one chapter per transcript. Arranging chapters across sessions and themes is still your work in the editor.
• You cannot move a section boundary on the review screen. If a left-out stretch matters, add it to the chapter by hand afterwards.
• It drafts prose from the transcript, so every chapter still needs the checks above: nothing added in the speaker's voice, facts corrected in the open.

The book is $199, once, with no subscription. The checkout is not open yet while we finish our first books.

If you are working from a relative who has died, there is no hurry and no reason to use us. The method in this guide is the whole of it.

FAQ

How do you turn family interview transcripts into a book?

Read all the transcripts once without editing. Build a timeline with one row per story, then choose a structure: chronological for one person, by generation for several. Edit each passage by removing the interviewer and filler and reordering within a story, adding only bracketed context. Add short italic lead-ins for dates and places, group retold stories by theme, then print.

Should a family history book be chronological or organized by generation?

Chronological if the transcripts are one person's life and most stories can be dated. By generation if you interviewed two or three people of different ages; give each generation a part and keep each part chronological inside. Choose thematic chapters when the material is deep on a few subjects and vague on years.

How much should you edit an oral history transcript for a book?

Enough to make it readable and no more. Remove the interviewer's questions and "mm-hm"s, trim crutch words and false starts, and rejoin interrupted stories, but keep repetition used for emphasis and the speaker's own word order. Baylor University's oral history guide keeps no more than two crutch words per occurrence per page.

Can you add words to a transcript to make it read better?

Only visibly. A word or short phrase the reader needs, such as who "he" refers to, goes in square brackets so it is clearly the editor's. Longer context, like a date, a place, or what happened between two stories, belongs in a short italic lead-in in a narrator's voice. Never put words in the speaker's mouth that are not in the transcript.

Can AI turn interview transcripts into a book?

It can do the mechanical part well: removing feedback words, rejoining interrupted stories, proposing sections, and drafting prose. It cannot know what it should not have left out, or which facts are wrong. Check coverage before accepting anything, compare each chapter against the transcript, and never let it fill a gap with plausible detail the speaker never gave.

Where to go next

If the transcripts came from recordings of someone who has died, how to turn old voice recordings into a written memoir covers the parts this guide assumes are done: digitizing, transcription and the gaps you can no longer ask about.

If what you care about most is that the book sounds like the person, how to preserve a parent's voice in a book explains why drafting from the transcript with the audio open matters more than any other habit.

If the family's material also includes paper, what to do with old letters and journals from a deceased relative covers what to keep, what to scan and how letters can fill the years the interviews skip.

And for the whole project in one place, from deciding what to capture to getting it to the next generation, see our complete guide to family story preservation.

Sources cited above

• Oral History Association, "Best Practices": avoiding manipulation of the narrator's words, retaining the integrity of the narrator's perspective, recognizing the subjectivity of the interview. Retrieved 29 September 2026.
• Baylor University Institute for Oral History, Style Guide: A Quick Reference for Editing Oral History Transcripts, revised May 2018: feedback words, crutch words, false starts, brackets for added material. Retrieved 29 September 2026.
• Ohio History Connection, Guide to Transcribing and Summarizing Oral Histories, 2019: limits on inserted material and bracketed corrections of misremembered facts. Retrieved 29 September 2026.
• Yourtale, internal measurement of our transcript import on one 53-minute, two-speaker family interview, August 2026: 1,439 raw turns to 448 after cleaning, 567 feedback lines removed, 71 interrupted sentences rejoined, 52% of interviewer lines were feedback only, subject coverage 50% on the first segmentation and 88% after the fix.