Choose an AI assistant by testing the same task

Choose your assistant by comparing the same task

You do not need to rank every AI product before you start. Use a small, fictional task and send precisely the same request to two conversational assistants you can access. Compare their actual answers, not their reputations.

1. 1

   Open a conversational AI assistant

   For example, use ChatGPT, Claude, Gemini or Copilot. Check whether the service is accessible in your region and account.

2. 2

   Send the ready-made request

   Keep the wording identical, including its constraints.

3. 3

   Open a second assistant

   Paste exactly the same request into a new conversation. Save both answers for your comparison.

4. 4

   Compare against your criteria

   Check whether each response obeys the two-sentence limit, avoids inventing personal details and matches the requested tone.

5. 5

   Choose one for your next task

   Write down the reason for your choice, along with one thing you still need to verify.

Write three short possible openings for a farewell speech for a colleague. Keep each opening to no more than two sentences. Use a warm but relaxed tone. Do not invent a name or personal details. End with one sentence explaining which opening would suit the most informal gathering.

Do not let the second assistant see the first answer before you make the comparison. Otherwise you will be testing its ability to rewrite the first answer, not its independent response to the same task.

Compare the responses, not the brands

• Instruction-following: did each answer produce three openings with no more than two sentences each?
• Reliability: did either assistant invent a colleague’s name, career history or personal details?
• Usefulness: which answer would need fewer edits in your real situation?
• Control: can you ask a follow-up question and get a meaningful change without losing the earlier constraints?

Replace the most informal opening with one suitable for a formal gathering. Preserve the two-sentence limit and do not invent names or personal details.

Use the same follow-up separately in both conversations. Compare how well the tools keep your limits when asked to revise. A polished answer may still be unsuitable if it invents facts.

Write down a decision you can explain

I will paste the original request, answer A and answer B below. Compare them on two criteria that can be checked from those inputs: (1) whether each answer follows the original instructions and (2) how usable each answer is for the stated task. Quote the exact passages that support each judgment. Do not assume an assistant is easier to revise unless I also provide a follow-up request and both revised responses. If those follow-up exchanges are absent, say that revision performance is untested. If the original request or either answer is missing, ask me to supply it before comparing. Do not rank brands by reputation.

Make the final choice yourself. For example: “I will use assistant A for this short writing task because it followed the tone and length limits; I still need to check that the text fits my colleague.” This is evidence about one task, not a guarantee of future output.

Check your account before depending on a feature

1. 1

   Open the assistant’s settings or account page

   Find your plan and the features shown as available.

2. 2

   Compare with what you need

   For example, do you need file upload, web access or saved conversation history?

3. 3

   If something is unavailable, find out why

   Look for plan, region, device or organization restrictions. Do not assume all users have identical features.

4. 4

   Record what you verified

   Write down one feature you can actually use and one limitation to revisit later.

A service’s features and prices can change. Use its own current settings and documentation as the source of truth rather than a screenshot, comparison table or AI-generated claim about your account.

Practice without a second account

The exercise below works entirely on this page with two fictional responses based on a short source passage. It lets you check the comparison criteria even if you only have access to one assistant today. You can use the external two-assistant exercise later.