Which AI to use, and for what
I would choose the job before choosing the subscription. A tool earns its place when it helps you finish useful work, fits your organization's rules, and does not create more checking than it saves.
What you will leave with
A tool choice based on your work, with the account and access checked.
Practice time: about 30 minutesWrite a one-sentence job description
"Help with marketing" is too broad to compare. "Turn our approved workshop brief into a customer email that our coordinator can review" gives you a task, a source, and a standard.
Next, check the account you are allowed to use. A familiar brand name does not establish permission to upload your files. Record the product, plan, signed-in account, enabled connections, and organizational owner. A personal account and a managed business workspace can have different terms and controls.
Shortlist by where the work happens
- ChatGPT or Claude: Candidates for a separate drafting, analysis, or thinking workspace. Compare them with the same approved material instead of assuming one brand is always the better writer. OpenAI and Anthropic describe specific business-data commitments; those do not automatically apply to every personal account. Read OpenAI's business privacy information and Anthropic's commercial data guidance.
- Gemini in Google Workspace: A candidate when the job lives in Gmail or shared documents. Availability and access depend on your account and administrator settings. Google says Workspace access follows user permissions; that is a reason to review existing sharing, not a reason to assume every file is appropriate. See Google's Workspace privacy and controls.
- Microsoft 365 Copilot: A candidate when the job lives in Microsoft 365. Check the specific product, license, and enabled features with your administrator. Microsoft describes access within existing permissions, which makes the quality of those permissions part of the setup. See Microsoft's security guidance.
For a spreadsheet calculation or a simple rule that moves a record, the tool you already use may be sufficient without generative AI. Keep the task small enough to compare honestly.
Give each candidate the same test
Worked example / Fictional practice material
You need an event announcement. The approved brief has a topic and audience but no confirmed date. Give two approved tools the same brief, tone example, and instruction to mark missing facts.
A useful draft preserves the gap. A draft that invents a date fails the accuracy check, even if you like its voice. Record how long each takes to repair and approve.
Try a routine case, an incomplete case, and a case with conflicting source material. Score each output from 0 to 2 for factual accuracy, completeness, usability, and total effort. Use 0 for unusable, 1 for substantial correction, and 2 for acceptable with routine review. For effort, compare the total time with your usual process: slower, similar, or faster. Keep critical failures separate: a good overall score must not cancel an invented commitment or an unacceptable data practice.
Choose for this job, then revisit
Pick the candidate that clears your account and data requirements and produces useful work at an acceptable total cost. Include review time, administration, and training as well as the subscription. If neither helps, improve the task definition or leave the work where it is.
Write the reason down. Repeat the test when the task, tool, source access, or plan changes. This is a provisional choice for a piece of work, not a permanent ranking of models.
Keep this
Tool comparison sheet
Copy this into your own document and fill in the brackets. Use approved or fictional material when trying it with AI.
JOB / OUTPUT: [One task and its expected result.]
CANDIDATES: [No more than two approved options.]
ACCOUNT CHECK: [Product, plan, account, admin, enabled connections.]
DATA CHECK: [Permitted material, retention, training use, access, deletion.]
TEST CASES: [Routine / missing information / conflicting information.]
SCORE EACH 0 TO 2: [Accuracy / completeness / usability / total effort.]
CRITICAL FAILURES: [Record separately; do not average away.]
TOTAL TIME: [Preparation + drafting + review + corrections.]
DECISION / REASON: [Choose / retest / use existing process.]
OWNER / REVISIT DATE: [Name and date.]Product guidance checked September 5, 2026. Follow the official links above for the terms and controls that apply to your account.
Want help applying this to your organization? Bring the job you have in mind, and we can work out a useful next step.