Turning a panel into an article that reads better than the session

Most panel write-ups follow the running order. The moderator opened by asking about supply chains, so paragraph one is about supply chains. Then the second question, then the third, then a closing line about a lively discussion and thanks to the sponsor.

Turning a panel into an article that anybody finishes means abandoning that shape entirely. The session was organised around a moderator's question list and a 60 minute slot, and neither of those constraints applies to a page. What the reader wants is the part where the panellists did not agree, which in the transcript is buried at minute 41.

The transcript order is the wrong order

A panel is a live format with live obligations. The moderator has to bring in the speaker who has not talked, cover the questions promised in the brief, and land the whole thing on time. Those obligations produce a sequence, and the sequence carries no argument.

Read a transcript with a pen and the structure becomes obvious. Roughly the first eight minutes are introductions and a question that everyone answers in turn, agreeing. The middle third is where the differences show up, usually triggered by one speaker giving a specific number or naming a practice. The last ten minutes are audience questions, of which one is genuinely good and the rest are pitches.

An article built on that sequence inherits its shape. An article built on the argument does not.

Find the disagreement

Divergence is the thing to hunt for, and it has recognisable signatures in a transcript.

The clearest is the qualified agreement: a speaker starts with yes, and then spends 90 seconds explaining a condition under which the previous answer fails. The second is the interruption, which in a well-moderated panel is rare and therefore informative. The third is the specific counter-example, where one panellist answers a general claim with a case from their own show or their own product line.

There is a structural reason those moments are longer than the surrounding talk. Yuan, Liberman and Cieri, in their 2006 Interspeech work on speaking rate in conversation, found that speaking rate as a function of turn length "rises rapidly for turns from one to seven words; remains level (when final words are included) or falls gradually (if final words are excluded) for turns of medium length; and rises slowly for longer turns", and also that speakers use slower rates and longer turns when talking with strangers. Panellists are usually strangers to each other in exactly that sense. The passages where somebody takes a long turn and pushes are the passages where the substance is, and a word count per turn, which any script can produce from a transcript, points at them without anyone having to listen again.

Two divergences carry an article. One is thin and reads like a report of a moment. Three starts to feel like a list. If your desk already runs an extraction pass over each recording, as turning session recordings into content assets sets out, the candidate passages arrive with timecodes attached and this stage takes twenty minutes.

Sort the candidates before committing, because two of the three kinds of disagreement are worth less than they look. A disagreement about definitions, where two people use the word attendee to mean different things, is worth a sentence of explanation and no more. A disagreement about predictions, where one panellist thinks the market grows and another does not, is unresolvable and reads as two opinions side by side. The one to build on is a disagreement about practice, where two organisers describe doing the same job in incompatible ways and both have a show that works. That has evidence attached on each side, and a reader running the same job can act on it.

What is the right length for a panel write up?

Around 900 words, and the arithmetic is worth doing because it explains why the instinct to write 2,000 is wrong.

A 60 minute session contains perhaps 45 minutes of actual speech once introductions, applause, AV changes and the pauses between speakers are removed. At a working assumption of 150 to 200 words a minute of sustained talking, that is between 6,750 and 9,000 words of transcript. Getting to 900 is a compression of roughly eight or ten to one.

That ratio sounds brutal until you count what is in the transcript. The moderator restating each question, perhaps 600 words. Introductions, 700. The four passages where a panellist agrees with the last answer and adds nothing, 1,200. Two audience pitches disguised as questions, 500. None of that survives into a written piece, and it is a third of the file before you have made a single editorial judgement.

The reading side supports the same target. Jakob Nielsen, writing for Nielsen Norman Group in 2008 and drawing on an analysis of 45,237 page views by Weinreich and colleagues, concluded that "users have time to read at most 28% of the words during an average visit; 20% is more likely", and that on pages of 111 words or less users read about half the content. Take the optimistic end of that: a 900 word article gets about 250 words genuinely read. That is your opening plus one section. Put the disagreement there, in the first 200 words, and let the supporting quotes come after it.

Add the context nobody had time for

Compression alone produces a shorter summary. The thing that makes the article better than the session is the material the panel could not include because it was live.

Three kinds of context are usually available and usually missing. The published source behind a figure a panellist quoted from memory, which you can find and cite properly. The counter-argument from a speaker who was not in the room, which you can get in a ten minute call. The history, meaning the two previous editions where the same question was asked and answered differently, which your own archive holds and nobody in the room had time to look up.

The first of those repays the effort most often. A panellist says the category grew by about a fifth last year, everyone nods, and the number goes into your write-up as a spoken figure. Twenty minutes with the published source usually produces one of three outcomes: the figure is right and now has a citation, the figure is right for a different period or a different geography and needs qualifying, or no published source says it. All three are more useful to a reader than the unattributed version, and the third one is a finding.

There is a search argument for doing this beyond the editorial one. Google's ranking systems guide, updated in December 2025, says the company has systems "to help ensure we are showing original content prominently in search results, including original reporting, ahead of those who merely cite it". A write-up that reports what was said on your stage is original reporting. A write-up that also resolves a disputed figure against a published source is original reporting with something in it that no other account of the session will have.

The archive point is worth a system rather than a memory. If your desk keeps its quotes in a register keyed by speaker and edition, as speaker quotes as reusable assets argues it should, finding what the same person said in 2024 takes a filter instead of an afternoon.

Should you send the draft to the speakers?

Yes, and the scope goes in the email rather than being left to negotiation afterwards.

The wording that works is short. Here is the piece, publishing Thursday. Please correct any factual error, your job title, and any company or product name we have misheard. We are not able to change quoted opinions, which were given on a public stage. That one sentence turns a courtesy into a defined request, and it heads off the version of this that consumes a week, where a communications team returns a rewritten quote that says nothing.

Two practical notes. Send the piece rather than the quote in isolation, because a quote read out of context reads worse than it did in the room and generates objections that the surrounding paragraph would have answered. And send it to everyone quoted at the same time, not in sequence, because a panellist reading a colleague's sharpened position tends to sharpen their own.

The attribution details you are asking them to check should already be right, because they came out of the conference transcript editing workflow rather than out of the programme PDF. Using the speaker check as your first attribution pass wastes it.

Where this stops

Some panels contain no disagreement at all. Four people who broadly agree, a moderator who kept it comfortable, and 60 minutes of consensus. The method described here has nothing to work with, and the honest responses are to write a shorter piece built on the single best example anyone gave, or to write nothing and use the material as background for a commissioned piece later.

What you must not do is manufacture the divergence. Two speakers emphasising different parts of the same position can be edited into apparent conflict by cutting the qualifiers, and the result reads well, gets shared, and is a fabrication. It also produces the phone call where a speaker says they were misrepresented, and they will be right.

There is a quieter version of the same failure. Quoting the most quotable line rather than the most representative one turns a careful speaker into a provocateur over a portfolio's worth of coverage, and it is invisible in any single article. That is a reason to keep the content studio view of what a given speaker has been quoted saying across your shows, and to look at it occasionally.

Take the transcript of your last panel, compute the word count of every turn, and read the ten longest. If the disagreement is not in there, it probably was not in the room, and you have saved yourself a day of writing to find that out.