Skip to main content

Try the solution on a small sample

Copy a prompt, use a fictional input, and check the result before trying your own work. Free downloads, no signup.

Download the coding, spreadsheet and draft-queue practice fixtures (JSON). Reference arithmetic, SQL and workflow checks were executed locally; these are fictional inputs, not AI benchmark outputs.

Fix a formula with an expected answer

A five-row fictional expense table. Supplies should total 20; Travel should total 25.

Copy this prompt

I am using Google Sheets. Given the attached expense CSV in A1:B6, calculate the total for Supplies. Show the formula, identify its ranges and test it against the rows. Then test a category with no matches. Do not assume the first row is data.

Reference result

=SUMIF(A2:A6,"Supplies",B2:B6) → 20. Travel → 25. A missing category → 0.

  • Keep the criterion and sum ranges aligned.
  • Confirm decimal values were imported as numbers.
  • Test a category with no matching rows.

The fixture totals are checked by the repository verification script. This is a deterministic formula example, not a recorded AI benchmark.

Extract decisions without inventing owners

Fictional notes with one explicit owner and two unresolved tasks.

Copy this prompt

Read the attached fictional meeting notes. Return decisions, commitments and open questions separately. For each action give the exact evidence quote, owner and due date. Use "not assigned" or "not stated" for missing details. Do not infer an owner from who spoke most.

Reference result

Landing-page draft: Asha, Friday. Pricing review: not assigned, not stated. Contact-form check: not assigned, not stated. Existing logo is approved for the draft; a rebrand is not approved.

  • Every commitment points to words in the notes.
  • Unassigned work remains unassigned.
  • An idea or prerequisite does not become an approved decision.

This is an illustrative reference answer from the fictional notes. Compare your model response with the source; wording may differ.

Build a source ledger before sharing AI research

Use a public product announcement or documentation page. Separate the claim from your interpretation.

Copy this prompt

Using only the sources I provide, create a claim ledger: claim, source URL, exact relevant passage, applicability limits and verification status. Mark unsupported claims unverified. Separate product facts, predictions and your inference. Do not invent citations or replace missing evidence with confidence.

Reference result

A claim-by-claim ledger with unresolved facts still marked unverified. A link alone is insufficient; its passage must support the claim.

  • Open each source and locate the supporting passage.
  • Check dates, plan limits and whether the evidence covers the exact claim.
  • Remove unsupported performance percentages and ranking guarantees.

A reusable editorial template, not a guarantee of factual accuracy. Human verification remains necessary.

A useful reason to come back

One small AI skill.
A few minutes each day.

Try a prompt, work with a small example, and check the result. Save your practice progress on this device.

Try today’s free practice

Start with a free resource

Download a social media prompt starter or source-checking template. Affordable editable kits are there when you need more.

We use optional cookies for analytics and referral tracking. Learn more about our privacy policy