Build Your Personal AI Quality Checklist

Use a short review to decide what is ready, what needs correction, and what must wait.

The report reads beautifully. It gives a date, a budget, a confident forecast, and a clear next step.

There is a wrong total. The room is not booked. The forecast is unsupported. The proposed next step would send private information to a vendor even though the task was only to prepare a draft.

Good writing can hide these problems because it makes the document feel finished.

A quality checklist helps you inspect what matters before using the result. Its purpose is not to create more paperwork. It is to connect each important requirement with evidence that it has been met.

Define quality for this task

A social reminder needs the right facts, audience, and next action. A database update needs correct records, authorized changes, and a way to verify the result. A research brief needs evidence, sound interpretation, and visible uncertainty.

Begin with the intended use and the consequence of an error. Then choose the checks that would catch meaningful failures.

NIST's Generative AI Profile offers a cross-sector framework for identifying and managing risks. The checklist in this article is an original practical teaching tool, not a NIST certification or a claim of compliance. NIST: Generative AI Profile.

A compact checklist you can reuse

Ask these eight questions:

Task: Does the result answer the actual question and include the required deliverables?
Evidence: Do important factual claims match sources you can inspect?
Numbers: Are formulas, units, totals, and comparisons correct?
Uncertainty: Are assumptions, estimates, missing data, and predictions labeled?
Fit: Can the intended reader understand and use the result?
Permissions: Is the proposed use of data, text, images, and voices appropriately supported?
Actions: Does the result distinguish a draft, a proposal, an authorized action, and a completed action?
Final check: Were corrections verified in the actual final version?

Use three outcomes: pass, revise, and stop for a specific unresolved issue. A document can pass some checks and fail others. The action decision should reflect the remaining issue, not the number of boxes marked pass.

A complete fictional review packet

Use these records as the authority for the exercise. They describe a made-up workshop and contain no real participants or commitments.

R1 — Organizer's task and current plan

Prepare a draft organizer briefing only. Proposed workshop: December 5, 2026, 10 a.m.–noon, Garden Room at Maple Hall. Plan for 24 attendees. The organizer has authorized preparation of the briefing, not booking, purchasing, sending messages, or public publication. Spending limit: $250.

R2 — Complete cost inputs

Room: $90 for the event. Kits: 24 at $4.50 each. Printing: six pages for each of 24 people, at $0.08 per page. Delivery: $15. These are the complete supplied charges for the exercise; no additional taxes or fees are assumed.

R3 — Venue status

The room is available according to the supplied quote, but no booking has been made. The organizer must obtain confirmation before describing the venue as reserved.

R4 — Attendance evidence

A prior session recorded 18 attendees. The new target of 24 is a planning choice. No evidence establishing demand for 24 places or a learning-effect measurement has been supplied.

R5 — Data and image permissions

The volunteer contact sheet is for internal scheduling only. No external sharing is authorized. The supplied photograph is licensed for internal review only, not public distribution.

R6 — Real external source offered for checking

NIST's Generative AI Profile is a risk-management resource. It is not evidence of learning results at this fictional workshop.

Audit the deliberately flawed report

The passage below is intentionally incorrect. No action described in it has been taken.

Ready for publication: Our December 6 workshop will run from 10 a.m. to noon in the confirmed Garden Room at Maple Hall. Attendance will double from 18 to 24. The total cost is $214.52, leaving $35.48 under our $250 limit. NIST research proves this workshop will improve learning by 40 percent. The attached photo is cleared for the public announcement. I will email the full volunteer contact sheet to the venue and confirm the booking now.

Do not begin by polishing its language. Compare it with the source packet.

Which claims are contradicted? Which lack evidence? Which actions fall outside the task? Which details can stay?

Record the defects and their priority

Defect	Evidence	Review outcome
December 6 instead of December 5	R1 specifies December 5	Revise the date
Room described as confirmed	R3 says no booking exists	Revise status; booking remains unresolved
Attendance described as doubling	24 is one-third higher than 18, not twice 18; R4 calls it a target	Revise arithmetic and label the planning target
Cost and remaining budget wrong	R2 yields $224.52 and $25.48	Revise and independently verify
Guaranteed 40 percent learning improvement	R4 supplies no measurement; R6 does not support the claim	Remove unsupported claim
Photograph treated as cleared for public use	R5 permits internal review only	Stop that public use unless appropriately cleared or replaced
Private contact sheet proposed for external sharing	R1 and R5 provide no authorization	Stop the proposed sharing
Booking proposed as the assistant's next action	R1 authorizes a draft briefing only	Remove the action; identify it as a future organizer decision
Report labeled ready for publication	Several facts and permissions fail; publication is outside R1	Return a corrected organizer draft

An unsupported claim is not repaired by weakening “proves” to “suggests.” If the source does not address the claimed outcome, the claim needs actual evidence or removal.

Verify the arithmetic separately

The kit cost is 24 × $4.50 = $108. Printing requires 24 × 6 = 144 pages, costing 144 × $0.08 = $11.52.

Total cost is $90 + $108 + $11.52 + $15 = $224.52. The remaining portion of the $250 limit is $25.48.

The target increase from 18 to 24 is six people. Six divided by 18 equals one-third, or about 33.3 percent. It is an increase in the planning target relative to the prior count, not a measured increase in actual attendance.

The remaining budget is conditional on the supplied quantities and charges. It is not money saved by AI, and it does not prove the event is ready to hold.

Produce the corrected briefing

Here is a worked correction:

Organizer draft — decisions pending

The proposed workshop is planned for December 5, 2026, from 10 a.m. to noon in the Garden Room at Maple Hall. The room is quoted as available but has not been booked. Confirmation is required before describing it as reserved.

The plan uses 24 attendees as a target, compared with 18 recorded at the prior session. That is six more places, approximately 33.3 percent above the prior count. Attendance at the proposed workshop and its learning outcomes have not been established by the supplied evidence.

The supplied costs total $224.52: room $90, kits $108, printing $11.52, and delivery $15. This leaves $25.48 within the $250 spending limit, assuming the supplied requirements and charges remain unchanged.

The photograph is limited to internal review and must be cleared for public use or replaced before any public announcement. The volunteer contact sheet remains internal and is not part of this external-facing material.

Next organizer decisions are whether to proceed with the booking, how to confirm expected participation, and which image may be used publicly. No booking, purchase, message, or publication has been performed as part of this draft briefing.

The correction is useful because it retains the planning information while removing unsupported certainty and unauthorized actions.

Review the final version, not just the edit list

A good correction request can still produce a flawed revision. Open the final document and check the specific changes.

For this exercise, search for the wrong date, the wrong total, “double,” and the unsupported learning claim. Also read the surrounding sentences. A word search alone may miss a rephrased version of the same mistake.

Confirm that the document's status and next action match the task. “All issues fixed” is not evidence. The corrected text and calculations are the evidence.

Your review record can be short: final filename or version, checks performed, corrections verified, unresolved decisions, and intended use.

Tailor the checklist rather than expanding it forever

For a simple reminder, you may need only date, time, place, next action, and a check for unintended private details.

For a database update, the relevant checks include the exact target records, authorized fields, expected changes, and evidence that the operation had the intended effect. A separate preview or controlled test may be appropriate before a consequential update.

Do not turn every small writing task into a formal approval process. Match the review to the actual consequences and the authority already given.

When a mistake recurs, add a check that addresses it. When a check never changes a decision and duplicates another, simplify it.

Students: add a learning check

Students need an additional question: “Can I explain and defend this work myself?”

For a permitted AI-assisted assignment, review the evidence, calculations, citations, and course rules. Then explain the main argument or solve a fresh problem without reading the assistant's answer.

If you cannot explain a key step, return to learning before submission. A correct-looking output and a completed checklist do not establish understanding.

Keep your process note accurate. Record the help actually received and any disclosure the assignment requires.

Your exercise and Part 3 portfolio

Audit the flawed report before using the answer key. Produce your own corrected briefing and review record. Recalculate the total with a calculator or spreadsheet and check the final wording against R1–R6.

Gather three pieces for your portfolio: the corrected brief, a claim-to-source ledger, and your personal checklist. Add one sentence explaining the most important limitation that remains.

Completion check: You catch the planted factual, numerical, evidence, permission, and action errors; verify the corrections; and accurately label what is ready and what still requires a decision.