AI interview editing: keep the qualification with the quote

A short interview clip can contain only real words and still change what the speaker meant. The problem often sits just outside the selected range: a question, a qualification, or the sentence that explains who “we” refers to.

This walkthrough uses one fictional interview to build a main story, a standalone clip and a focused revision. The useful artifact is a paper edit you can trace back to the recording—not just a transcript with deletions.

Published by Valmera. Written and published with an AI agent. All speakers, passages and timings in the worked example are fictional teaching data, not a customer recording or a measured product run.

Decide what you are making

“Edit this interview” can describe several different jobs:

A finished MP4 and an editable timeline are different deliverables. Settle that requirement before committing to an editor.

Vendor pages illustrate the distinction. Eddie AI describes importing multiple angles and separate audio, generating a rough edit and exporting to desktop editing applications. Endroll describes building cuts from real takes using a script or selected transcript lines. MirrorCut describes reviewing agent changes and exporting MP4 or captions, while marking Premiere XML and OpenTimelineIO as forthcoming. These are vendor descriptions checked November 7, 2026, not our comparative test results.

The expanded interview guide includes a fourth workflow and specific trial questions. Whatever editor you choose, test your required delivery on a representative passage before processing the whole interview.

Keep two records: what exists and what you propose

A source ledger identifies passages in the original recording. A paper edit places selected passages into a proposed story order.

Keep source time separate from output time. A passage beginning at 00:26 in the recording might begin at 00:24 in the main story and at 00:16 in a standalone clip. Replacing its source reference with either output time makes later corrections harder to trace.

For the fictional file workshop-interview.mp4, use this ledger:

These ranges are start-inclusive and end-exclusive. They describe straight cuts at normal speed. Real boundaries require listening and frame review; tidy invented numbers do not establish natural cut points.

In your own ledger, add verified speaker identity, exact words or a locator phrase, necessary context, and any uncertainty. The transcript helps find a moment. The recording is the reference for words, emphasis and overlapping speech.

Build the main story

The order Q → A → B → C → D gives the viewer a question, an observation, its limits, an explanation and a bounded next step.

The planned program lasts 56 seconds. That excludes any added titles, transitions, pauses or end card. Measure the actual rendered duration separately.

A starting brief could be:

Make an interview story for [audience] using [source recording]. Treat the attached ledger as a proposal and verify the words, speakers and boundaries against the recording. Preserve the pilot question, the observed benefit, the team size and the stated limitations. Remove repetition where it does not change meaning. Do not invent speech or use a reaction from another moment to imply a response. Render a preview and identify the source selections used. Report anything you could not verify.

Replace the placeholders with your real material. Do not give the assistant the fictional filename and expect it to describe your upload.

Give the standalone clip its own editorial decision

Selection A alone is a tempting 16-second clip. It retains the team size, but it drops the explicit limits about scale and reliability.

For a clip presenting the pilot's conclusion, A + B is a better proposed selection: 28 seconds. A occupies clip time 00:00–00:16; B occupies 00:16–00:28.

That still needs review. Does the opening identify the checklist? If not, include more source context or use an accurate editorial label that is visibly distinct from the guest's words. A title claiming the checklist works for every team would overstate this interview even if the spoken excerpt were perfectly transcribed.

Also check the surrounding presentation. The title, opening caption and thumbnail should not erase the qualification that the spoken clip retained.

Create a separate clip from [verified ranges]. Keep [qualification] with [claim]. Check that the opening identifies the subject without requiring the main interview. Preview this clip separately and flag unresolved context.

Approval of the main story does not approve the clip.

Follow one correction through both deliverables

Suppose a reviewer confirms that the first four seconds of C are redundant and asks to retain source 00:44–00:52 instead of 00:40–00:52.

The main story becomes 52 seconds. The standalone clip remains 28 seconds because it does not use C.

Captions inside C need remapping to the retained words. Captions belonging to D move four seconds earlier. Moving every caption earlier would damage the sections before C. See the source-to-output caption walkthrough for a more detailed timing example.

Keep the request narrow:

In main-story preview [version], shorten only the explanation corresponding to original source [range]. Retain [words/action] and preserve the other approved story decisions. Re-render the main story and update its captions. Identify the new version and outstanding checks. The separate clip uses different selections; do not replace it with the main-story export.

Check that the new opening of C sounds natural. A correct duration calculation cannot establish an intact syllable, a comfortable breath or an honest reaction shot.

Record the result, including what remains unchecked

For each deliverable, record the reviewed version, export filename, dimensions, program duration, total file duration and remaining issues. Watch the downloaded file through its ending.

When comparing editors, record hands-on review time separately from waiting time. Note actual cost or credits consumed instead of assuming a subscription price tells you the cost of this job. Keep first-pass problems in the record even after a revision fixes them.

If you have multiple cameras, verify a shared visible-and-audible event and another later event to check for drift. Matching timecodes alone do not establish synchronization. Check the joins after editing, including whether a reaction shot suggests a response that never happened.

The copyable interview worksheet collects the source ledger, paper edit, revision log and delivery review. It also includes a trial record for comparing tools with the same source and brief. “Not checked” is a useful result; it should never silently become “passed.”

You can bring your recording and brief to Valmera Studio or use a supported connected assistant. Editing requires an eligible paid plan; external assistants may have separate billing. Check the current plans for your route. The worksheet is public and requires no account.