# AI Writing Tool Selection Card: One-Page Version

Use this card to compare **two AI writing tools for one recurring writing job**.

## 1. Define the job

**Writing job:** _________________________________________________  
**Finished artifact:** ___________________________________________  
**Starting material:** ___________________________________________  
**Human-led decisions:** _________________________________________  
**Current friction:** ____________________________________________  
**Current time to acceptable final work:** ________________________

## 2. Set the boundaries

**Allowed information:** _________________________________________  
**Information kept outside the tool:** ____________________________  
**Evidence that must remain verifiable:** _________________________  
**Required reviewer:** ___________________________________________  
**Final owner:** _________________________________________________

## 3. Name the finalists

**Finalist A and plan:** __________________________________________  
**Finalist B and plan:** __________________________________________

Verify current plan details, limits, file support, integrations, data terms, controls, and pricing before testing.

## 4. Score WORK Fit

| Dimension | Question | A | B |
|---|---|---:|---:|
| **Work** | Can it handle the actual job? | __ / 3 | __ / 3 |
| **Oversight** | Can a person review it and remain accountable? | __ / 3 | __ / 3 |
| **Repeatability** | Will reusable setup improve this recurring job? | __ / 3 | __ / 3 |
| **Knowledge** | Can approved context be supplied safely and reliably? | __ / 3 | __ / 3 |
| **Total** | Maximum 12 | **__ / 12** | **__ / 12** |

Scoring guide: **0 = poor fit, 1 = weak, 2 = workable, 3 = strong fit.**

## 5. Check hard failures

Remove, restrict, or redesign a candidate if it:

- [ ] requires information that should not enter the tool;
- [ ] produces important claims that cannot be checked;
- [ ] bypasses a required reviewer;
- [ ] leaves no one accountable for the final result;
- [ ] can take consequential action without required approval; or
- [ ] creates more verification and cleanup than it removes.

**A result:** [ ] Continue [ ] Restrict/redesign [ ] Remove  
**B result:** [ ] Continue [ ] Restrict/redesign [ ] Remove

## 6. Run the same four tests

| Test | Finalist A | Finalist B |
|---|---|---|
| Ordinary case |  |  |
| Incomplete case |  |  |
| Conflicting case |  |  |
| Stop-or-escalate case |  |  |

## 7. Count the complete workload

Track factual repair, missing requirements, tone or clarity repair, source verification, formatting and transfer, failed handoffs, and maintenance.

**Manual baseline:** __________________  
**Finalist A time to acceptable final work:** __________________  
**Finalist B time to acceptable final work:** __________________

**Switching or maintenance concern:** _____________________________

## 8. Decide

- [ ] Adopt narrowly
- [ ] Revise and retest
- [ ] Restrict
- [ ] Require human approval
- [ ] Reject

**Selected tool, if any:** ________________________________________  
**Reason:** ______________________________________________________  
**Review trigger:** ______________________________________________  
**Review date:** __________________  **Decision owner:** ___________

> **Choose the tool that makes the finished work easier, not the tool that makes the first 30 seconds look impressive.**