Giving good context

Supplying the information the model cannot know — and nothing it should not.

A model knows the world in general and your situation not at all. It has never seen your project, your customers, your codebase, or the email thread that explains why everything changed on Tuesday. When it answers plausibly but wrong, the missing ingredient is usually context you could have supplied.

Good context is selected, not shoveled. The skill runs in both directions: including what matters, and deliberately leaving out what does not — especially what should never leave the building.

1. Stage 1 · E1 FOUNDATIONAL

   Realize the model cannot see your desk

   If you ask about "the project" and get a confident, wrong answer, the model was not lying — it filled a gap you left. Anything specific to your work must be pasted, attached, or described. Assume the model knows nothing about your situation until you tell it.

   For documents, give the document. Naming it is not enough — a summary from a model that has actually read the text beats one recalled from training data every time, and asking for answers "with citations to the text" keeps it honest.

   Try it this week

   Ask the model a question about your current work twice: once cold, once with the relevant document or background pasted in. Study the difference — that gap is what context is worth.

2. Stage 2 · E2 PROFICIENT

   Select what matters, explain what it means

   Data without meaning is half-context. If you share a spreadsheet, say what each column means and what decision the analysis feeds. "Column C is churn reason, free text; I need to know whether pricing complaints are growing" turns a data dump into an answerable question.

   Resist both extremes: the two columns you guessed were relevant may not be, and "here is everything, figure it out" buries the signal. Choose the material a smart new colleague would need for this specific question — that is the standard.

   Try it this week

   Before your next analysis request, write three lines: what the data is, what each relevant field means, and what decision the answer feeds. Prepend them to the request.

3. Stage 3 · E3 DISTINGUISHED

   Manage context over time

   Long conversations degrade: constraints set at the start get crowded out, and the thread slows and drifts. The fix is deliberate context hygiene — periodically restate the rules that matter, or start a fresh conversation seeded with a tight summary of decisions, constraints, and only the artifacts still in play.

   For work you will return to repeatedly, prepare context once: a working digest of the 400-page manual, a standing project brief, a glossary of your team's terms. Reusable context turns every future request from an essay into a sentence.

   Try it this week

   Write a one-page standing brief for your main project: goal, status, constraints, key vocabulary. Start your next three AI conversations by pasting it, and note how much re-explaining disappears.

4. Stage 4 · E4 EXCEPTIONAL

   Engineer context as a system

   The exceptional operator designs what the model sees the way an editor designs a briefing pack: ordered, labeled, and scoped to the decision. They exclude with intent — irrelevant history that invites drift, and sensitive detail that has no business in the request at all.

   They also build verifiability into the ask: requesting themes with counts and verbatim examples rather than impressions, so the model's reading can be checked against the raw material. Context in, traceability out.

   Try it this week

   For a real analysis, write the full briefing pack: labeled inputs, field meanings, the decision, the output format with traceable examples — and a note of what you deliberately excluded and why.

Ready to see where you stand? The free check scores all six competencies in about fifteen minutes.