All packs
Productivity & research · Any AI assistant

Reading AI answers

Judge what AI assistants and agents hand back: citations, numbers, formulas, charts, research reports, recaps and agent actions.

Start: Citations and claims

About Any AI assistant

Reading AI answers is a skills pack for checking what AI assistants and agents hand back before you repeat it or act on it. It works across tools like ChatGPT, Microsoft 365 Copilot and Gemini. You practice spotting uncited claims, wrong totals, percent mix-ups, broken spreadsheet formulas, misleading charts, thin research reports, wrong meeting recaps and agent runs that didn't finish.

Best for: Anyone who uses AI answers at work and passes them on to others.

Good for

  • Checking that each claim has a citation that really supports it
  • Catching totals that don't add up, wrong units and percent versus points
  • Reading formulas like VLOOKUP and SUMIFS before you trust the result
  • Comparing what an agent says it did with what its run log shows

Where people slip

  • Trusting a report with five sources that all trace back to one press release
  • Reading a maybe in a meeting recap as a decision
  • Missing a filter that limits an analysis to one quarter
  • Believing an agent's Done! message when a step is still waiting
How do I check if an AI answer is accurate?

Open the citations and confirm each one supports the claim next to it. Check the scope of the data, the units and the totals. For anything that matters, ask the tool to show its work, then check a few numbers yourself.

What is the difference between percent and percentage points?

Going from 4% to 7% is a rise of 3 percentage points. As a percent change, it's a 75% increase. AI answers sometimes mix the two, so check which one a claim means.

Why can VLOOKUP return the wrong value?

If you leave out the last argument, VLOOKUP uses approximate match. On unsorted data that can return the wrong row with no error. Set the last argument to FALSE for an exact match.

Can I trust an agent when it says a task is done?

Check the run log first. An agent can finish some steps and stop on others, like a failed sign-in or a send that's waiting for approval. Compare what it says it did with what actually changed.

Checked against Any AI assistant's own docs on 2026-11-06.

Milestone 1

Check claims, scope and runs

Trace claims to sources, read an analysis's scope, and review an agent's run log.

0 of 3 lessons complete
Citations and claims3 exercises
Numbers, recaps and scope3 exercises · Pro
Agent run logs3 exercises · Pro
Milestone 2

Check numbers, charts and formulas

Catch wrong totals, mixed-up percentages, misleading axes and risky spreadsheet formulas before they spread.

0 of 2 lessons complete
Numbers that don't add up5 exercises · Pro
Charts and spreadsheet formulas5 exercises · Pro
Milestone 3

Judge reports, recaps and agent actions

Steer and check deep research reports, fix meeting recaps, and verify what an agent really did.

0 of 2 lessons complete
Deep research reports5 exercises · Pro
Meeting recaps and agent actions5 exercises · Pro