Citing creators properly when your research is built on their work

The short answer: reference rather than reproduce — paraphrase the insight in your own words, attach creator, title, link and timestamp to the specific claim, and never ship someone else’s transcript or video as your own material. Claim-level attribution is also what makes the research usable a year later, so the careful version and the useful version are the same version.

Research built on public content sits on a foundation someone else paid for. A creator spent a decade learning the job and a weekend filming it; your conclusions exist because they did that. Treating attribution as paperwork misses that the same discipline solves a practical problem — a research artefact whose claims cannot be traced back to a source stops being evidence within about a month.

Two obligations, only one of them legal

The first obligation is about reproduction: do not republish material that is not yours. That means no re-uploaded clips, no wholesale transcripts presented as your content, no long verbatim passages doing the work your own analysis should be doing.

The second is about honesty in the artefact: every non-obvious claim should say where it came from, so a reader can check it rather than trusting you. That obligation has no legal force at all and matters more day to day, because it is the difference between research and assertion.

Referencing and reproducing are different acts

The clean mental line is whether your reader could get the same value without going to the source. If your write-up says “three of fourteen practitioners described month-end reconciliation as the step they stay late for” and links the three, you have referenced: the analysis is yours and the sources back it. If your write-up is four paragraphs of someone’s transcript with a link underneath, you have reproduced, and the link does not change that.

The product-surface row is the one that catches builders. A research tool that keeps notes and links is doing something different from one that hands every user a full transcript of a third party’s video, and the distinction is worth designing for early rather than retrofitting after launch.

Attach citations to claims, not to documents

A bibliography at the end of a research document is nearly useless. It tells a reader that fourteen videos were watched, and nothing about which one produced the conclusion they are currently doubting. Claim-level citation — source and timestamp next to the assertion — turns a doubt into a two-minute check.

The payoff compounds. Six months later, when a conclusion is questioned in a planning meeting, a cited claim can be re-verified while the meeting is still happening; an uncited one gets settled by whoever is most confident. That is the same argument as insisting on timestamps and citations from a summariser and the reason it matters when work is handed on, covered in sharing research with your team.

Cite what disagrees with you

Attribution has an under-used honest form: recording the sources that contradicted the conclusion as well as the ones that supported it. Three supporting videos and one dissenting one is a different finding from four supporting ones, and only the first of those is what actually happened in most corpora.

Keeping the dissent visible is what stops a research artefact from hardening into folklore, and it is the practical defence against the confirmation dynamics described in the failure modes of AI video research and in separating hype from signal.

Ideas travel; expression does not

Builders sometimes worry that learning from a creator’s content makes a resulting product derivative. It generally does not. Facts about how a job works, and the observation that a step is painful, are not anyone’s property, and every competitive market is built on people noticing the same public problems.

What is not fine is copying a specific expression: a creator’s own product, their branding, their distinctive materials, or their content republished under your name. The line is intuitive in practice — learning from a walkthrough that invoicing is a nightmare is research, cloning the invoicing template they sell is not — and it is the same boundary that governs a competitor read in a competitor teardown built from public video.

Attribution is also relationship management

The creators whose content anchors your research are often the best conversations available to you. They understand the job, they have an audience with the problem, and they are usually pleased to hear their work informed something real — provided the reference is specific and accurate rather than flattering and vague.

Getting the attribution right before you reach out is what makes that message land. It also matters at launch, since the practitioners you studied and the channels you eventually launch into are frequently the same people, which is the connection drawn in reading distribution signals out of creator content and part of the case for working in the open in building in public with a research trail.

A minimum standard worth adopting

Four elements, on every claim that matters: creator or channel, video title, link, and the timestamp. Paraphrase by default, quote only when the exact wording is the point, and keep the quoted fragment short enough that it illustrates rather than substitutes.

Adopt that once and the rest follows: research becomes checkable, published work becomes defensible, and the artefacts stay useful long after the session that produced them. It also makes the context files your team builds on genuinely maintainable, which is the discipline in keeping a CLAUDE.md current as research evolves.

What this costs

Attribution costs nothing when it is captured at the moment a note is made and is nearly impossible to reconstruct afterwards, which is the whole argument for tooling that keeps the source attached. As of September 2026, Hobby is $19 a month for 25 videos and 2 projects, Pro is $59 for 80 videos and 8 projects, and Studio is $199 for 250 videos, 20 projects and 3 seats, each with a 7-day free trial — see the pricing page.

Closing thought

The test is simple and slightly uncomfortable: if the creator you learned the most from read your research document, would they recognise themselves in it and be pleased to be named. If the honest answer is no, the problem is rarely the citation format.