I've written before about knowing when to stop researching. This is the other half of the problem: how to do it properly in the first place. Across the Legions of Honour novels, the Khemaret Cycle and now five volumes of Detective Files, the method has settled into something fairly consistent, and it owes more to twenty-nine years of building cases than to anything I learned about writing.
Start wide, then go to the source. A good general history, a documentary, or an encyclopaedia entry is the right place to begin, because it tells you what happened, who matters and what the arguments are. It's a terrible place to finish. Every general account is someone else's summary, and summaries lose things. Use them as a map, then follow their footnotes and bibliography back to what they were built from.
Primary sources beat secondary ones. A primary source is as close to the event as you can get: a trial transcript, an inquest report, a contemporary newspaper account, a letter, a diary, an official record. A secondary source is someone interpreting those. Most of the real texture of a book comes from the first kind. A newspaper report from the week of a crime gives you what people actually knew and feared at the time, before hindsight tidied it into a story. For British material, the National Archives, the Old Bailey proceedings and the British Newspaper Archive are the obvious places to start. For other countries, Trove covers Australian newspapers, Papers Past covers New Zealand, and Chronicling America covers the United States. All of them are searchable from your own desk.
Separate what's documented from what's repeated. This is the discipline that matters most in non-fiction, and it's the one I'm strictest about in the Detective Files books. Any famous case gathers rumour, folklore and sensational reporting around the documented facts, and the legend usually ends up in the earliest and most-quoted accounts. I keep three piles for every claim: what the primary record supports, what is credibly reported but unproven, and what is simply repeated. Disputed suspects and unproved victim counts get labelled as such on the page. A reader can forgive an unsolved case. They shouldn't be handed a guess dressed up as a fact.
Keep a source log from day one. This is straight from detective work: an exhibit with no record of where it came from is worthless in court, and a fact with no record of where you found it is nearly as useless in a manuscript. For every note, write down the source, the page or date, and how far you trust it. When two sources disagree, and they will, you need to be able to go back and see which one is closer to the event. A plain spreadsheet works. So does a research folder in a tool like The Bindery, which keeps notes next to the chapter they belong to.
Fiction and non-fiction research differ in what they're for. In a novel, research serves plausibility and texture. You need to know how a Roman survey party worked, or what a flood gauge on the Nile measured, so that the scene feels solid and the reader never trips over an error. You're allowed to invent, as long as the invention is consistent with what was possible. In non-fiction, research serves accuracy and attribution. Nothing is invented, including the conversations. You can describe what was said only where a record says it was said. That's a much stricter standard, and it changes how you write: more qualified statements, more "according to", more honesty about gaps.
Verify everything, whatever its source. That includes books, websites and AI tools. A confident paragraph from any of them can still be wrong, and a single error repeated across enough sources starts to look like consensus. If a fact matters to your plot or your argument, find it in two independent places, and prefer the one nearest the event. Be especially careful with dates, names and numbers, which are where copying errors accumulate.
If you can, go there. Walking a location teaches you what no document does: how far it really is from the road to the river, how little you can see from a given window, how dark a lane is at night. A good half of the details readers remember from a book come from standing somewhere and noticing it. If you can't visit, large-scale maps, satellite imagery and local-history society publications get you surprisingly close.
One practical note for non-fiction about real people: writing about real events and real individuals, especially anyone living or recently dead, carries legal and ethical responsibilities beyond accuracy. I'm a novelist and a retired detective, not a lawyer, so for anything sensitive take proper advice before publishing.
Done well, research is mostly invisible. The reader never sees the source log, the three piles of evidence or the afternoon spent in a newspaper archive. They just feel that the world holds together. That's the whole point.