Turning session recordings into content assets your editorial team will actually use

Somewhere on a shared drive there is a folder from last year's conference with 84 files in it. The AV supplier delivered them a fortnight after the show, correctly named, and one person has opened one of them since. The marketing team asks twice a year whether there is anything usable in there and the honest answer is that nobody knows.

Turning session recordings into content assets is mostly a problem of definition. A 60 minute panel is raw material in the same way a court transcript is raw material, and a team without a fixed extraction routine will keep treating the folder as a library to browse, which nobody has time to do.

What an extraction pass produces

Define the output set once and apply it to every recording that enters the pipeline. The set I would fix on has four items.

The transcript, corrected to the point of being searchable, with speakers attributed. The correction work has its own defects and its own measurement, and it belongs to the conference transcript editing workflow rather than to this pass.

Two or three pull quotes, each with speaker, employer, and a start and end time. A pull quote without a timecode is a claim about what somebody said, and verifying it later means watching the session again.

One data point, meaning a figure stated from the stage that can be traced to a published source. Most sessions produce at least one. Many produce a figure the speaker could not source when asked afterwards, which is itself worth knowing before the number ends up in your show's marketing.

A clip list: start and end times for passages that would stand alone, with a one line description. The pass produces candidates and stops there. Whether they get cut, in what aspect ratio, and to what length is video clipping from session recordings work with its own selection rules.

Four items, one form, 40 minutes per hour of recording once a machine transcript exists. The fixed set is what turns an open-ended browsing task into something a coordinator can be given on a Tuesday.

Timing matters more than it looks. Run the pass within three weeks of the show, while somebody on the team still remembers which sessions were good, and while the speakers still answer email about a talk they gave recently. At three months the same request reads as an odd approach from a stranger, and the person who sat in the room has moved on to next year's programme.

Why should every output carry the recording id?

Because otherwise reuse is unmeasurable, and unmeasurable work gets cut in the first budget round.

The pattern to borrow comes from audiovisual cataloguing. PBCore, the cataloguing standard for audiovisual content, separates an asset, which is the intellectual content, from its instantiations, which are the particular formats and copies that content exists in. One recorded session is the asset. The archive master, the platform stream, the transcript, the vertical clip and the pull quote are instantiations and derivatives of it, and every one of them should carry the same identifier back to the source.

The log itself is small. Eight columns carry it: asset id, recording id, edition, session title, asset type, start and end time where the asset is a passage, the person who produced it, and a status field recording whether it has been used and where. Eight columns in a spreadsheet is a defensible first version, and a team that cannot keep eight columns filled will not keep a purpose-built system filled either.

Once the identifier is on everything, a set of questions becomes answerable in a query instead of a meeting. How many published assets came from the 2025 edition. Which sessions produced more than five. Which produced none, and were those the sessions that were badly attended or the ones where the audio failed. What proportion of everything the desk published in the second half of the year originated in a recording.

That last number is the one that decides whether recording the sessions is worth what the AV supplier charges. It is also the number that stops the same panel being mined four times by four people who did not know the others had been there, which is a portfolio-level failure the content repurposing rules are meant to prevent.

Timecodes need a written form

Write time ranges in a form that a person and a system read the same way, and use it everywhere from the extraction sheet to the CMS field.

The W3C published Media Fragments URI 1.0 as a Recommendation in September 2012, and it defines exactly this. A temporal fragment is written as t equals a start and an end in seconds, so t=1830,1902 addresses the 72 seconds beginning at thirty minutes and thirty seconds. The specification also allows the clock form, so t=0:30:30,0:31:42 means the same thing, and it defines the interval as half open, with the begin time inside the range and the end time the first moment outside it.

Half open sounds like pedantry until two people cut the same clip from the same numbers and get different last frames. It is worth having a rule.

The practical consequence is that a pull quote in your asset log has a machine-resolvable address. Given the recording identifier and t=1830,1902, anyone can jump to the passage, the fact checker can confirm the wording, and the clip editor does not have to be told what to look for. A note in a spreadsheet saying the good bit is about half an hour in costs each subsequent person ten minutes of scrubbing.

Which sessions are worth extracting?

Not all of them, and the arithmetic settles it quickly. Take a conference with 84 recorded sessions and use illustrative figures for the rest.

Extraction at 40 minutes a session across the full programme is 56 hours, or seven working days of somebody's time. Say the pass yields nine logged artefacts per session, so 756 in total, and that 120 of them get used in something published within 90 days. That is 28 minutes of extraction time per used artefact, and a reuse rate of 16 per cent.

Now take the ten highest attended sessions. That is 6.7 hours of extraction, 90 artefacts, and because these are the sessions with the subjects the audience actually turned up for, say 45 of them get used. Cost per used artefact drops to about nine minutes, and the reuse rate reaches 50 per cent.

The full-programme pass produced more total output, 120 used artefacts against 45, so this is a real trade rather than an obvious win. The point is that the second batch of sessions is three times more expensive per usable output than the first, and you should know that before committing a coordinator to seven days. Extract the top ten, measure your own reuse rate at 90 days, and extend the set only while the cost per used artefact stays acceptable.

Attendance is a proxy, and a rough one. Session-level demand data from your app or your scanning at room entry is better, and the on-demand view counts from the platform are better still if the sessions were streamed.

Making the session and the article point at each other

The relationship between a session and the things made from it can be stated in your markup, which helps the machines reading your event pages understand that the article is coverage of a real thing that happened.

Schema.org, in version 30.0 published in March 2026, defines recordedIn as a property of Event holding "the CreativeWork that captured all or part of this Event", with recordedAt as the inverse property on the CreativeWork side. So the session, marked up as an Event, points at the recording, and the recording points back at the session. Publishers reach for the video markup first and often skip this pair, which loses the connection between the programme and the archive.

Keep the same relationships in your own asset log whether or not they reach the page. The markup and the log are the same graph written twice, and the log is the one your team queries.

Where this stops

An extraction pass cannot recover what the room did not capture. If the panel's best twelve minutes happened after a question from the floor and the floor microphone was off, the transcript has a hole in it, the clip is unusable, and no amount of process fixes it in post. This is an argument for a short conversation with the AV supplier about floor audio, held before the show rather than after.

The second limit is that quality of extraction depends on the format of the session. A single-speaker presentation with slides yields cleanly, because the argument has a structure and the data points are on screen where the transcript can be checked against them. A six-person panel with a lively moderator yields quotes and very little else, because the substance arrives in fragments that need reassembly, and that reassembly is a writing job rather than an extraction one.

Permission is a separate gate and sits upstream of everything here. Extract into an internal log freely if your recording agreement allows the recording at all, and settle publication rights per speaker before anything derived from the session leaves the building. The content studio side of this work is where those two workflows meet.

Pick your last edition's ten best-attended sessions, run the four-item pass on one of them, and time yourself. Whatever that hour tells you about yield is a better basis for the AV conversation than any estimate of what might be in the folder.