Research does not fail at
gathering. It fails at deciding.

Write the question and the decision it feeds: Not the topic, the question. Not "research competitors" but "should we price above or below the two tools our buyers already use, and why". A question has an answer, so it can be finished. A topic cannot, which is why topic-shaped research runs until the deadline hits.

Agree what a good enough answer looks like: Decide up front how confident you need to be and what evidence would satisfy you. A pricing decision may need three comparable data points. A strategy bet may need ten interviews. Writing the bar down converts an endless task into a finite one.

Set a stopping rule before you start reading: The practical version is saturation: stop when new sources keep landing in themes you already have and stop introducing anything new. On a canvas that moment is visible, because every fresh card lands in an existing pile rather than starting a new one.

Keep a list of what you deliberately did not look at: Scope you excluded on purpose is not a gap, it is a decision, as long as it is written down. Parking it on the board stops you rediscovering the same rabbit hole three times and gives anyone reading your work an honest picture of its edges.

Put the claim on the card, not just the URL: A bookmark is a promise to read something later, and later never comes. Write the one sentence the source actually gave you at the moment you read it. If you cannot state the claim in a sentence, you have not finished reading it, and saving the link will not fix that.

Keep the quote, the number, and the date: Capture the exact line or figure you would cite, who said it, and when it was published. Research goes stale silently, and a market number from four years ago looks identical to one from last quarter once it has been retyped into your notes without its date.

Separate what the source says from what you conclude: This is the single most valuable habit in research. Keep the source's claim and your own interpretation visibly distinct on the card, because in a week you will not remember which was which, and inherited assumptions that were never actually evidenced are how confident wrong answers get made.

Note who is speaking and what they want: A vendor's benchmark, an analyst's forecast, and a user's complaint are three different kinds of evidence. Recording the source type beside the claim lets you weight it later, and stops five press releases from feeling like five independent confirmations.

Move cards until the themes name themselves: Push each finding towards the findings it resembles and stop when the groups feel stable, then name each one. A theme reached this way survives contact with your evidence, because it came out of the sources instead of out of a heading you wrote before reading them.

Run a deliberate contradiction pass: Once the clusters exist, go looking for conflict on purpose rather than waiting to notice it. Two sources disagreeing usually differ in something specific: the sample, the market, the year, the definition, or the incentive. Write that difference on the connection between them.

Mark confidence, not just conclusions: Some of what you found is well evidenced and some is one person's opinion repeated by three blogs. Labelling each claim as strong, mixed, or thin means the person reading your board, including you next month, can tell which parts are safe to build on.

Leave the open questions on the board: What you could not answer is part of the output. An explicit list of unknowns beside your conclusion is far more useful than a tidy summary that quietly pretends the gaps are not there, and it is the natural brief for the next round of work.

It clusters and structures what you put there: Because the AI has your whole canvas as context, you can ask it to group the cards you wrote into candidate themes, to propose a structure for the write-up, or to point at two cards whose claims look like they conflict. It works from your notes, so what you get back is only as good as what you captured.

It does not search the web or fetch sources for you: Storyflow does not run searches, crawl sites, or scrape pages. Finding the sources is your job, and honestly that is the part you should not automate, because reading is where the judgement comes from. Bring what you found and the canvas is where you make sense of it.

It does not automatically read or summarise your PDFs: Attaching a report to a card stores it beside the claim so the original is one click away, but it does not mean the document has been ingested and summarised. Pull out the claim, the number, and the quote yourself, then the AI can work with those.

Re-cluster without losing anything: Drag a card into a different theme and everything on it moves too, including the quote, the source, the confidence label, and the attached file. Trying a completely different structure for your conclusion costs a few drags instead of a rewrite.