Why we research before we write a single sentence

Generating first and checking later produces confident fiction. The scan runs first so the draft is written against something real.

Thinking Plans · 17 August 2026 · 5 min read

The failure mode of AI-written plans is well known: fluent paragraphs, invented figures, no sources. It happens because the model is asked to write before it is given anything to write about.

Order of operations

Here the sequence is inverted. Before section one exists, the engine runs a live scan against your industry, country and horizon, and returns signals you can read — each with its source. You keep the ones that matter and discard the noise. Only then does writing begin, and the sections are written against the signals you kept.

  1. 1Scan the market for your specific venture context.
  2. 2Review signals; keep, drop or add your own.
  3. 3Generate sections grounded in the kept set.
  4. 4Cite. Sources stay attached to the document, including in the share view.

Assumptions are allowed — silence is not

No scan covers everything, and small local markets are thinly documented. Where a number has to be assumed, the document says so in plain language. A labelled assumption is defensible in a lender's meeting. An unlabelled one is a credibility risk.

Readers do not punish uncertainty. They punish uncertainty that pretends to be fact.

Put this to work on your own venture

Describe the business once and the engine researches, drafts and scores the document — then tells you what changed when you come back to it.

Start a plan

Keep reading