AI Tool Selection Scorecard

Choose by a repeatable task test, not a universal tool ranking.

Define the test

Task: ___ Information needed: ___ Permitted sources: ___ Sensitivity and approved accounts: ___ Required output format: ___ Reviewer/owner: ___

Use the same permitted input and prompt in two tools. Record tool, version/account where visible, date and settings. Compare like-for-like access where possible.

Record each result

Tool A / Tool B

Accuracy: supported facts ___ / ___; material errors ___ / ___

Correction time in minutes: ___ / ___

Output fits the required format and workflow: ___ / ___

Missing sources or unsupported assumptions: ___ / ___

Total task time including checking: ___ / ___

Winner and reason: ___ Switch trigger: ___ Next review date: ___

Comparison prompt

Complete [task] for [audience] using only [permitted input]. Return [format]. Mark unsupported claims and missing information. Do not fill gaps with invented facts.

Worked example · fictional

Task: draft a weekly update from a fictional tracker. A needs 8 minutes of corrections; B needs 3. Both meet the format, but B invents a deadline. Resolve that factual failure before choosing; speed alone is not the deciding measure.

Three-task test log

1. Routine email: inputs ___ A result ___ B result ___ winner/reason ___

2. Notes-to-plan: inputs ___ A result ___ B result ___ winner/reason ___

3. Research summary: sources ___ A result ___ B result ___ winner/reason ___

These are blank test slots, not Bee’s results. Do not upload confidential material to an unapproved tool.