How to turn a recorded call into a transcript and a summary you can file

Last updated 2026-09-15 · 8 min read

What is the short version?

A recorded call is evidence nobody will ever rewatch. What people use is a document — one they can search, quote in an email and attach to a ticket — and that is a different deliverable from a wall of text in a chat window.

1. Go to Vizard Agent with the recording.
2. Ask for the transcript and a summary, as files.
3. Say you want the actions listed, not just the topics.

What do you need before you start?

The recording and a view on what the summary is for. Vizard Agent writes to that — a summary for people who were on the call is a different document from one for somebody who was not, and the difference is mostly how much context it repeats.

• The recording. Screen capture or audio.
• Who the summary is for. Attendees, or someone else.
• Whether you need timecodes. Usually worth having.
• The format you file in. Word, or plain text.
• Whether names matter. Say who is who if you know.

What do you type into Vizard Agent?

Ask for files in the request itself. A transcript in a chat is unusable at length — you cannot attach it, search it properly or hand it to anybody — and asking afterwards means the work is done twice.

Prompt

Variants worth knowing:

• "As a Word file." Not pasted into the chat.
• "A separate summary document." Two files, two jobs.
• "Decisions and action items." What a summary is for.

What does Vizard Agent actually do?

Here is the sequence on a real recorded call. Note that the summary is written after the transcript has been read rather than from an impression of the recording, which is the difference between a summary and a guess.

1. Checks the recording's length and parameters.
2. Transcribes the whole call with timecodes.
3. Saves a readable text version of the transcript.
4. Splits it into sections of manageable size.
5. Reads each section in turn, start to finish.
6. Writes the transcript and the summary from what it read.
7. Builds the document files.
8. Checks the file structure is valid and opens cleanly.
9. Fixes and rebuilds any document that did not.

Step two matters for what comes later. Timecodes in the transcript mean a disputed line can be listened to rather than argued about, and they cost nothing to include at this stage.

Step four is a quiet piece of method. A long call read as one block gets read unevenly — the middle of it least of all — so Vizard Agent divides the transcript into sections of a size it can actually attend to and works through them in order.

Step five is what makes the summary trustworthy. Summarising a long call from a skim produces the topics that were mentioned most, which is not the same as the decisions that were made — the decision is often one sentence somewhere in the middle.

Step eight is the step that sounds like housekeeping and is not. A document that will not open is worse than no document, and a file built programmatically needs checking rather than assuming, which is why it gets rebuilt when the structure is wrong.

What does the result look like?

Two files: the full transcript with its timecodes, and a separate summary carrying the decisions and the actions. Both of them open properly, both are searchable, and both can be attached to whatever ticket or email needs them.

Vizard Agent will produce the next call in the same shape, so a run of meetings ends up as a consistent set of documents rather than a pile of different formats.

When does this not work well?

A transcript is only ever as good as the audio underneath it, and Vizard Agent will say when that is the limit. Crosstalk, poor microphones and unfamiliar vocabulary all degrade it, and a summary built on an unreliable transcript inherits every single error in it.

• Poor audio. Everything downstream suffers.
• Heavy crosstalk. Attribution becomes guesswork.
• Calls with no decisions. There is nothing to action.
• Highly technical vocabulary. Names and terms get mangled.
• Confidential recordings. Check you may transcribe them.

How do you fix a result that came back wrong?

Say whether the problem is in the transcript or in the summary. Because Vizard Agent builds the second out of the first, a single wrong word in the transcript is often the reason the summary reads oddly, and correcting it at the source repairs both documents at once.

• "That word is wrong." Corrected in both documents.
• "The summary missed a decision." Pointed at by timecode.
• "Too long." A shorter summary, same transcript.
• "I need it as plain text." Same content, different file.

How does Vizard Agent compare to doing it yourself?

By hand a call gets transcribed by a tool and the output gets pasted somewhere. Nobody reads it through, so the summary is written from memory of the call, which is exactly the thing the recording exists to replace.

Common questions

Can it do both in one go? Yes. Vizard Agent produces the transcript and the summary together.

Will it list who said what? If the speakers are distinguishable, yes. Tell Vizard Agent who is who if you know.

Does it include timecodes? In the transcript, yes. Vizard Agent leaves them out of the summary.

What format are the files? Word by default. Tell Vizard Agent if you file in something else.

Can it pull out the action items? Yes, and Vizard Agent treats a summary without them as barely worth having.

How long can the call be? Long. Vizard Agent reads it in sections rather than all at once.

Will it summarise from a skim? No. Vizard Agent reads the sections through before writing a word of it.

Can it do several calls? Yes. Vizard Agent keeps them in the same shape, which makes them comparable.

What if a name is wrong? Tell Vizard Agent the spelling and it fixes both documents.

Does it keep the recording? The files are the deliverable. Vizard Agent leaves the recording as yours.

Can I get a shorter version? Yes. Ask Vizard Agent for a one-page summary as well.

Does it work on a video call recording? Yes, and it will check the picture if you need context from it.

Can it flag disagreements? Ask and Vizard Agent will note where the call did not resolve something.

Why not just read the chat output? Because at length it cannot be searched, attached or filed.