Research Before Writing: How to Gather Your Material and Verify It

Published Updated Say
Rows of archive boxes on wooden shelves, representing gathering material and verifying its sources
Photo by Luke Caunt via Unsplash License

The Figure With No Origin

He said in the video that seventy percent of craftspeople had left the trade in ten years.

A striking figure. He'd found it in an article, and that article had found it in another article.

After publishing, someone wrote in the comments: where is this figure from?

So he went back looking for the origin.

The first article cites the second, the second cites a report, and the report doesn't contain that figure. It contains a different one, about a single region, from twelve years ago, measuring something else.

He wasn't lying. He'd believed what he read and never climbed to its source.

And the video is published under his name, and the comment is still there.

Why Researching While Writing Fails

Researching while writing makes you select the information that supports what you just wrote, rather than what describes reality. The result is a script that looks coherent and falls apart under review, because the order of work itself pushed you toward what was comfortable.

Separate the two stages: gather first without writing, then write from what you gathered.

Four Layers of Sources

Rank your sources by their closeness to the event:

  1. Direct testimony: what you saw yourself, or what someone who lived it told you. The strongest source and the hardest to gather.
  2. Primary documents: official statements, figures issued by the relevant body, documents and archives. Strong if read in full rather than excerpted.
  3. Specialist coverage: journalists or researchers reporting from the two layers above. Useful for context, and requires tracing back to the original.
  4. Recycled content: articles and clips citing someone who cited someone. The weakest, used to establish that a subject exists rather than as a citation.

The rule: whenever you build an important claim, climb a layer or two before fixing it in place. The figure at the top of this article came from layer four, and its user never climbed a single step.

Practical Verification of a Number

Before you put any number in your script, answer three questions.

Who issued it? A named body, not "studies suggest".

When? A figure with no date is meaningless, because conditions change.

And what exactly does it measure? Plenty of figures get cut from their context and end up saying something other than what they measured.

And if you can't reach a clear answer, you have two options: delete the figure, or state it with explicit attribution to its source and date and let the viewer judge. Presenting it as a bare fact is a risk published under your name.

Signs That Warrant Doubt

A figure circulating everywhere without its original source ever being named.

A claim that suits your narrative rather too well, which is the most dangerous case, because wanting it true disables the checking.

A translation from a foreign source without going back to the original text.

Information several years old on a fast-changing subject.

Organise Research So It Serves the Writing

Keep research notes inside the video project itself rather than in a separate file, so the material stays next to the script that will be built on it.

And record each piece of information's source and date at the same moment, because going back two weeks later to remember who said what costs more time than noting it immediately. The project structure is covered in: managing content production projects.

When to Stop Researching

Research can go on forever, and becomes a disguised way of postponing the writing.

The right moment to stop is when you can write the core idea in one sentence and defend it with three independent sources. Anything beyond that improves the margins without changing the substance.

AI at This Stage

An AI assistant is useful for summarising material you gathered, for suggesting angles you hadn't thought of, and for exposing contradictions between two sources.

Generating the facts themselves isn't a safe use: any figure coming out of a generative model needs verification against a named source exactly like any other piece of information, as explained in: AI tools inside TikTok: what to delegate and what to keep.