Reality Lab · 10 MIN · Use
Break a big ask into steps
One giant prompt asks for research, analysis, and a finished deck in a single go — and every part comes back mediocre. Split it into steps you can check.
A marketer pastes one prompt: “Research our top 3 competitors, analyse their pricing, figure out where we’re losing, and write me a 10-slide deck with recommendations.” The result is confident and useless — invented competitor facts, hand-wavy analysis, and slides built on both. Because it all happened at once, there was no point at which a wrong turn could be caught.
Your task: Rewrite the single ask as an ordered sequence of smaller steps, with a checkpoint where you verify before the model builds on it.
"Research our top 3 competitors, analyse their pricing, figure out where we’re losing deals, and write me a 10-slide deck with recommendations."
This box is local to your browser. Nothing you type is sent anywhere or saved after you leave.
Check your own work
There's no grade here — checking against these is the exercise. Compare each point to what you wrote.
- The work is split into distinct steps, each with one clear output.A single prompt asking for four things spreads the model’s effort thin and hides where it went wrong. Separate steps each get full attention and are individually checkable.
- The fact-gathering step is separated from the analysis and the deck.Competitor facts are the part most likely to be fabricated. Isolating them means you can verify before they harden into ‘analysis’ and then into ‘recommendations.’
- There’s an explicit human checkpoint: verify the facts before building the deck.Decomposition only pays off if you actually stop at the seam. A checkpoint is what turns a pipeline into something steerable instead of a faster way to be confidently wrong.
- Later steps are told to use only the verified facts from earlier steps.Otherwise the model happily re-invents at each stage. Pinning each step to the checked output keeps errors from re-entering downstream.
Reveal a worked approach
A stepwise version: “1) List our top 3 competitors and, for each, the pricing you can find — with the source URL beside every number. Mark anything you’re unsure of. STOP so I can verify. 2) Using only the pricing I confirm, compare it to ours and identify where we look expensive or underspecified. 3) From that comparison, draft 5 recommendation bullets. 4) Only then, turn the confirmed comparison and recommendations into a 10-slide outline.”
Notice the seam at step 1’s ‘STOP.’ That single checkpoint is the difference between catching a made-up price in ten seconds and discovering it after it’s been baked into a board deck. Everything downstream is told to build only on what survived the check.
The habit to carry: when a request bundles ‘find out X’ with ‘then decide Y,’ split them, and put your verification exactly where fabrication would otherwise become foundation.