YouTube Key Findings Generator

There is a real difference between key points and key findings, and it is not pedantry. A key point is whatever a video emphasised. A finding is a claim someone is asserting is true about the world, usually with evidence and usually with conditions attached.

Those conditions are the whole thing. "Reaction times improved" is not a finding. "Reaction times improved by around twelve percent in the trained group, though the sample was small and self-selected" is — and the clause after "though" is the part that determines whether you can use it.

What separates a finding from a point

The fifth row is the one that causes real problems. Speakers cite other people's work constantly, and a summary that records "the speaker found X" when the speaker said "Chen and colleagues found X" has invented a research result.

Where findings hide in a talk

Research talks are structured, which helps, and the useful material is not evenly distributed through them.

The results section has the numbers, and they are the part most likely to be on a slide rather than spoken. Expect gaps.

The limitations section is where the honest speaker tells you what their work does not show. It is short, it is near the end, and it is the highest value-per-second passage in the entire talk.

The question session is frequently where the real caveats emerge, because someone in the audience asks the awkward question. It is also the worst- captioned part of any recording, since audience microphones are poor.

A summary organised by findings surfaces all three. One organised by topic tends to weight the introduction heavily, because that is where the speaker talks most fluently.

The slide problem is worse here

Everything quantitative in a research talk is on a slide, and slides are not spoken.

"As you can see, the effect is significant in both conditions" is what the caption track contains. The effect size, the confidence interval, the sample size and the p-value are on screen and absent from the transcript entirely.

So a findings summary of a research talk is structurally incomplete in the exact dimension you need it for. It will tell you what was claimed and not how strongly.

The workable approach treats the summary as a pointer rather than a source: use it to find which talks contain relevant findings, then get the numbers from the slides, the preprint or the published paper. Most conference talks correspond to a paper, and the paper is the citable artefact.

Why you cannot cite a summary

This is worth being unambiguous about, because the temptation is real when you have forty summaries and a deadline.

A summary is a paraphrase of an automatic transcript of a talk about a study. That is three transformations away from the evidence, each of which can drop a qualifier. Citing it would be citing a game of telephone.

Publication guidance is consistent on the underlying principle — the ICMJE recommendations are built throughout on the expectation that authors have engaged with the work they cite and are accountable for its accuracy. Citing something you have only encountered second-hand fails that expectation regardless of how the second-hand version was produced.

The correct chain is: summary finds the talk, talk names the paper, paper gets read, paper gets cited. The summary saved you the forty minutes of finding it, which was always the actual saving.

Recording a finding so it stays usable

The discipline that makes this worth doing is in how you write each finding down, and it takes one extra line per entry.

Every recorded finding should carry four things: the claim, who is making it, where it came from, and how confident they were. Drop any one and the entry becomes unusable six months later — a claim with no owner, or a number with no idea whether it was preliminary.

The fourth is the one people omit, and it is the one that decays fastest. "Effects were observed, though the speaker described the work as early-stage" is a different note from "effects were observed", and by the time you come back you will not remember which the talk actually supported.

This is the same working practice that summarizing research video depends on generally: the summary accelerates finding things, and the note-taking discipline is what stops the acceleration turning into error.

The accumulating archive

Where this pays off is across many talks rather than within one.

A field's conference output for a year is several hundred talks. Read as findings summaries, that is a map of what the field currently claims — including which claims keep recurring, which have quietly stopped being made, and which two groups are disagreeing about.

That view is not available any other way. Nobody watches three hundred talks, and the published literature lags conference presentations by a year or more, so the talks are where the field is now rather than where it was.

What it will not do

Give you numbers from slides. The most important limitation for this use case. Get them from the paper.

Distinguish reliably between the speaker's work and work they cite. Check before attributing anything.

Assess whether a finding is sound. A methodologically weak study summarizes exactly as cleanly as a strong one.

Handle poorly-captioned question sessions, which is unfortunate given that is where the caveats often live.

Tell you a finding has since been retracted or failed to replicate. A talk is a snapshot of what one group believed on one day, and a summary of it carries no more currency than the recording does. A finding from a 2022 talk needs checking against what happened to that work afterwards, and nothing in the summary will prompt you to do it.

Frequently asked questions