A colleague forwards a link to a new AI tool. The demo looks impressive and the comments say it changes everything. You now have to decide whether it deserves your time, and all you have to go on is someone else's excitement.
Demos are built to show a tool at its best. Rankings and benchmarks measure things that may have little to do with your job. What you need is a quick way to find out how the tool handles your own work.
This 30-minute test does that. It works on any AI tool, whether it writes, summarises, analyses data or builds slides. You will finish with a score and a clear decision.
Before you start: get your test ready
The test only works if you use real work. A made-up task tells you very little, because you can't judge the result the way you would judge something going to your manager.
Pick one task that meets these three conditions:
- You do it often. A task you repeat every week gives you a real payoff if the tool helps.
- You know what good looks like. You should be able to spot a mistake without looking anything up.
- You have done it before. A finished version from last month gives you something to compare against.
Then prepare a safe version of the input. Replace client names, patient details, salaries and anything else confidential with dummy values.
Picture an HR manager in Bengaluru who writes a monthly attrition summary. She takes last month's data, swaps employee names for codes, and keeps the summary she wrote by hand. That is a complete test kit.
Minutes 0-5: check the data rules first
Do this before you paste anything. If the tool fails here, you can stop and save the other 25 minutes.
Look for answers to three questions on the tool's privacy or data page:
- Is my content used to train its AI models, and can I turn that off?
- Does it connect to my email, files or calendar, and can I limit what it sees?
- Does my company allow this tool for work data?
The first question matters even with well-known tools. ChatGPT, Claude and Gemini may all use chats on personal plans for training unless you opt out, while their business plans generally exclude your content by default. These settings change, so check the current page rather than relying on what you read a while ago.
If you can't find clear answers, treat that as a warning sign. Don't put work material into the tool until someone responsible for IT or compliance has looked at it.
Minutes 5-15: run your real task
Give the tool your task the way you would actually ask for it. Don't study tutorials first, because you want to see what an ordinary working day would feel like.
If you want a consistent starting point, use this prompt:
Role: You are an experienced [your job role] at an Indian company.
Ask: Complete the task below. List any assumptions you made, and tell me
which parts of your answer you are least sure about.
Context: [Describe the task in 2-3 lines, who will read the result, and
what it will be used for.]
Input: [Paste your cleaned-up material here.]
Tone: Clear and professional. Use plain English and no filler.
The line about assumptions does useful work. A tool that tells you where it is unsure saves you checking time, and a tool that sounds equally confident about everything needs closer reading.
Save the first answer before you ask for any changes. The first answer shows how much the tool gets right on its own.
Minutes 15-20: try to break it
Most tools handle a clean request well. Everyday work is messier, so spend five minutes on harder cases.
Try two of these:
- Ask it to change something. Say "make this shorter and move the risks to the top." See whether it keeps what was right or breaks other parts.
- Give it a gap. Remove one piece of information and see whether it asks, flags the gap or quietly makes something up.
- Ask it something it can't know. A question about your internal policy is a good test. The right answer is that it doesn't know.
- Ask for its source. If it quotes a fact, ask where the fact came from, then open the source yourself.
The second and third tests are the most revealing. A tool that fills gaps with confident invention will eventually put a wrong figure in front of your boss or your client.
Minutes 20-25: score it
Compare the result with your own finished version, and score each line out of 5.
| Question | What a 5 looks like |
|---|---|
| Accuracy | No wrong facts, figures or names |
| Editing effort | Usable with only small edits |
| Handling of gaps | It flagged what it didn't know instead of inventing it |
| Ease of use | You could repeat this without a tutorial or help |
| Time saved | Clearly faster, including the time spent checking |
Time saved is the line most people get wrong. Count the checking and fixing, not just how quickly the answer appeared. A draft in ten seconds that takes 40 minutes to correct may be slower than writing it yourself.
Treat accuracy as a hard rule. If the tool scored 2 or lower on accuracy, a great score elsewhere shouldn't rescue it.
Minutes 25-30: make the decision
Add up your score out of 25 and pick one of three outcomes.
- 20 or more: Adopt for this task. Use it on this one task for two weeks, and keep checking its work.
- 13 to 19: Retest once. Try a clearer prompt or a different task. Some tools fit one job and not another.
- 12 or below: Drop it. Close the tab without guilt. These cut-offs are my rule of thumb, so adjust them for how risky your work is.
Finally, write two lines somewhere your team can see: what you tested and what you decided. This stops three colleagues repeating the same test next month.
A clinic manager in Pune, for example, might find that a tool drafts appointment reminders well but stumbles on patient education material. That is a useful result. She can adopt it for reminders, keep a doctor reviewing anything clinical, and move on.
A prompt to review the result
If you want a second opinion on the output, open a different AI tool and ask it to look for problems. Keep the input free of confidential details here as well.
Role: You are a careful reviewer in my field who looks for errors.
Ask: Review the AI-written output below. List any claims that could be
wrong, anything important that is missing, and what I should verify myself.
Context: This output will be used for [purpose] and read by [audience].
Original input: [paste]
Output to review: [paste]
Tone: Direct and specific. Use a short list. Don't rewrite the output.
Use this to decide what to check, not as a final verdict. The second tool can make mistakes too, and the decision stays with you.
FAQs
How many tasks should I test before deciding?
One task is enough for a first decision. If you plan to use the tool widely, repeat the test on two or three different tasks.
Should I test the free version or the paid one?
Start with whichever you can access safely. Features and limits vary by plan, so confirm that the plan you test includes what you would actually use.
What if the tool is great but my company hasn't approved it?
Don't use work data in it. Share your results from dummy data with your IT or compliance team, and ask them to review it.
Try this today
- Choose one weekly task, and find a finished version you have already done.
- Replace confidential details with dummy values, then run the first prompt above in the tool you are curious about.
- Score the result out of 25, and write down your decision in two lines.