How to Build a Paper Edit From Interview Transcripts

Six hours of interviews came back from a shoot in Trieste and I had four weeks to turn them into a 24 minute film. The first time I faced that, years ago, I dropped all of it straight onto a timeline and started cutting. Three days later I had a 90 minute assembly, no idea where the good material was, and a scrub bar I had worn out looking for a line I half remembered somebody saying.

The paper edit is the fix, and it is older than digital editing. You build the structure of the film in a document first, using transcripts and timecode, and only then touch the timeline. It feels like a delay. It is not. On that Trieste film the paper edit took two and a half days and the assembly took six hours, which is roughly the trade I have measured on every long form project since.

Why the Timeline Is the Wrong Place to Find a Story

An interview is dense. A single 50 minute sit-down might hold eight usable minutes, scattered. On a timeline you can see waveforms and thumbnails, neither of which tells you what somebody said. So you scrub, you listen, you find a good line, you cut it in, and you lose the thread of the argument you were trying to build.

Text does not have that problem. You can read 10,000 words of transcript in half an hour, and more importantly you can reorder it, delete it, and see two paragraphs from different interviews next to each other. The comparison is the whole value. Deciding that the engineer's answer about the 1974 flood belongs before the mayor's and not after is a judgement you can make in ten seconds in a document and not at all by looking at clips.

There is a second benefit that only shows up later. A paper edit is a record of your reasoning. When a producer asks in week three why a particular section was dropped, the answer is in the document, with the timecode, and you are not re-watching rushes to defend a choice you already made well.

Getting Transcripts You Can Actually Work With

Automatic transcription has changed the economics of this completely. Ten years ago I paid roughly a pound a minute for a human transcriptionist and waited two days. Now a service runs an hour of clean two person audio in a few minutes for a fraction of that, and the accuracy on decent audio sits somewhere around 95 percent.

What matters more than accuracy is timecode. Insist on a transcript with timestamps at least every paragraph, and ideally per sentence. A perfect transcript with no timecode is close to useless for a paper edit, because every selection becomes a search. If your transcription tool exports SRT or VTT, you already have per-line timing.

Two practical notes from repeated pain. Feed the transcription the separate audio recorder files rather than the camera scratch track, since the error rate on a camera mic across a room is noticeably worse. And make sure your transcript timecode matches the timecode in your project. If your camera files start at 10:00:00:00 and your transcript starts at 00:00:00, write the offset at the top of the document before you forget it. I have lost an afternoon to exactly that.

The Five Pass Method

I work through the transcripts five times, and each pass has one job. Doing two jobs at once is how a paper edit turns into a 40 page document nobody can use.

Read everything once without marking anything. No highlighter, no notes. You are listening for what the film is actually about, which is usually not what you planned on the shoot. This takes an hour per two hours of transcript and people always want to skip it.
Highlight only what is usable. Second read, mark any passage that is clear, well said and self contained. Be generous here. You are at maybe 25 percent of the total if the interviews went well.
Pull selects into a spreadsheet. Four columns: speaker, in and out timecode, the quote, and a short tag such as "flood 74" or "funding row". One row per selection. A 24 minute film usually lands between 90 and 140 rows at this stage.
Sort by tag, not by interview. This is the pass that builds the film. Group every row about the same subject together regardless of who said it or which day it came from. Contradictions jump out immediately, and contradictions are usually your best scenes.
Order the groups and cut the document to length. Roughly 150 words of spoken English runs a minute, so a 24 minute film needs about 2,400 to 2,800 words of interview once you allow for sequences without speech. Cut the document down to near that figure before you open your editor.

Keep the spreadsheet and the ordered document as two separate files. The spreadsheet is your library and stays complete; the document is the film and gets brutal. Conflating them means deleting a line from the film also deletes it from your records, and you will want it back.

The Advice I Disagree With

Text based editing in Premiere, Resolve's transcription panel and tools like Descript let you select words in a transcript and have the editor cut the timeline for you. They are genuinely useful and I use them. The claim I keep seeing, that they make the paper edit obsolete, is wrong, and I think it misreads what the paper edit is for.

Those tools automate the mechanical half: finding the clip and placing the cut. They do nothing for the structural half, which is deciding what the film argues and in what order. When you work inside the editor's transcript panel you are still looking at one clip at a time in the order it was recorded. The thing a paper edit gives you, every usable line from six hours of material visible on one screen, sorted by theme, is precisely what the panel cannot show you.

There is a quieter cost too. Cutting directly from a transcript panel produces exactly the words you selected and nothing either side, so the assembly arrives with no handles. Then you spend an hour extending every clip to find breath and room tone. My compromise on recent projects: paper edit in a spreadsheet, then use the text panel to locate and place the selections quickly, trimming two seconds wide on each end deliberately.

Where I will concede ground is short form. For a six minute corporate piece from one 40 minute interview, a formal paper edit is overhead you do not need. Read the transcript, highlight, cut. The method earns its keep somewhere past about two hours of source material or three interviewees.

Taking It to the Timeline

Before importing anything, your project should already be set up properly. The way you name bins and handle media now determines whether the assembly is pleasant or miserable, which is the subject of organising a video edit before the first cut.

Then build the radio edit: every selection in order, audio only, no pictures, no B-roll, no music. Listen to it start to finish. This is the most honest version of your film you will ever hear, and it is where you discover that the section you loved on paper runs 90 seconds too long or that two interviewees say the same thing twice.

Only once the radio edit holds up do you lay in picture, and that is when split edits start doing their work. Offsetting audio from picture is what stops an interview assembly feeling like a slideshow of talking heads; see J cuts and L cuts for interview edits for the frame counts I use.

One habit worth keeping: as you cut, mark in the spreadsheet which rows made it into the timeline. On a long project with a change of direction in week three, that column is how you find the 40 selections you set aside, without reading six hours of transcript again.

FAQ

How long should a paper edit take?

Budget roughly half a day per hour of interview material for a first paper edit, including reading time. Six hours of rushes comes to two to three days. That number drops with practice, though not as much as you would hope, because the reading is the work and reading is not something you get much faster at.

Do I need a paper edit for a scripted piece?

No. If the words are written in advance and performed, you already have your structure and a paper edit duplicates it. The method exists for material where you do not know what you have until you read it, which means interviews, vox pops, observational footage and anything shot without a script.

Spreadsheet or word processor?

A spreadsheet for the selects library, because sorting by tag is the step that builds the film and a word processor cannot do it. A document for the ordered edit, because you need to read it as prose and judge whether it flows. I have tried doing both in one tool and always drift back to two.

What about the shots, not just the speech?

Keep a second sheet for the visual material, logged the same way: timecode, a one line description, and a tag that matches your interview tags. When you reach the picture stage you then search for "flood 74" and find both what people said and what you have to show. Logging rushes nobody will transcribe is tedious and it is the difference between remembering you have a shot and not.