
For you
Role Checks
Module · Does the model research, or write for you
The Analyst's AI Check
AI can draft an analysis in seconds that used to take you a day, and that is exactly the danger. The speed hides whether you still verified the sources, showed your method, and told the reader what you do not know. This module checks the five habits that separate an analyst who uses AI as a tool from one who has quietly handed over the thinking: source verification, method transparency, real synthesis, data discipline, and honest uncertainty.
What the five levels look like
Every dimension in this assessment is scored 1 to 5. This is what the levels mean, dimension by dimension. The graded report diagnoses where your own answers land and what to do about it.
You verify every source
- 1Never check
- 2Check if it looks odd
- 3Spot-check some
- 4Check the load-bearing ones
- 5Verify every cited claim
At the low end: You are signing your name to facts you have not seen. Open every source the model gives you before it reaches a deliverable; a fabricated citation in your work is your mistake, not the model's. What good looks like: Verifying every cited claim is the discipline that lets you use AI at speed without inheriting its fabrications. Keep the standard even when a deadline argues against it.
Your method is visible
- 1No trail at all
- 2In my head only
- 3Rough notes
- 4Reproducible steps
- 5Documented, prompts included
At the low end: Work nobody can retrace cannot be checked or defended. Start keeping the prompts and steps that produced each conclusion; it is the difference between analysis and assertion. What good looks like: A method a colleague can reproduce, prompts and all, is what makes your work auditable. This is what separates an analyst from a content generator.
You synthesise, not paste
- 1Lightly edited AI text
- 2Reworded output
- 3AI draft plus my edits
- 4My structure, AI inputs
- 5My judgment throughout
At the low end: Lightly edited model output is not analysis; it is transcription with a byline. Ask what you know that the model does not, and make that the spine of the deliverable. What good looks like: When your judgment runs through the whole piece, AI is an accelerator rather than a substitute. Keep asking what you are adding that a prompt could not.
You guard the data
- 1Paste anything
- 2Rarely think about it
- 3Avoid the obvious
- 4Follow a clear rule
- 5Rule plus safe tools
At the low end: Pasting sensitive data into a public tool is a leak you cannot reverse. Learn what your organisation classifies as confidential and treat the AI box as a public place until told otherwise. What good looks like: Knowing the line and having sanctioned tools for sensitive work is exactly right. Revisit the rule as you take on new data types; the risky category is often the new one.
You state what you do not know
- 1Present all as fact
- 2Caveat if pushed
- 3Occasional hedges
- 4Flag the weak spots
- 5Calibrated throughout
At the low end: Presenting estimates as certainties makes your work more dangerous the more it is trusted. Mark what is measured, what is inferred, and what the model simply asserted. What good looks like: Calibrated uncertainty is the rarest and most valuable analyst habit. It is what lets decision-makers trust your strong claims, because they can see you flag the weak ones.