Why verification is the skill that matters
AI assistants produce fluent text whether or not it is right. The risk is not that they are always wrong; it is that wrong answers look as confident as right ones. A short, repeatable check before anything leaves your desk is the difference between a time-saver and a liability.
What can go wrong
| Problem | What it looks like |
|---|---|
| Invented facts | Plausible but false statements, names or figures |
| Fake citations | References to papers, laws or sources that do not exist or do not say that |
| Arithmetic slips | Wrong totals, percentages or conversions |
| Outdated information | Rules, prices or data that have changed |
| Bias | Stereotyped or one-sided framing |
| Privacy leaks | Personal or confidential data entered or reproduced |
| Copyright issues | Text or images that closely copy protected work |
A ten-point checklist
- Is every fact something you can confirm in a primary or reputable source?
- Does every citation exist, and does it say what the AI claims? Open it
- Have you recomputed the numbers that matter?
- Is the information current for your jurisdiction and date?
- Does the language treat groups fairly and avoid stereotypes?
- Has any personal or confidential data been shared with the tool or appeared in the output?
- Could the content copy someone else's work too closely?
- Did you ask the assistant what it assumed or was unsure about?
- Has a person with the right expertise reviewed anything high-stakes?
- Is a named person accountable for what is sent or decided?
Match the check to the task
| Task | Minimum check |
|---|---|
| Email or short draft | Read for tone and facts; confirm names and dates |
| Research summary | Open each source; confirm the claim appears there |
| Numbers or analysis | Recompute headline figures; spot-check data |
| Legal, medical or financial content | Expert review before use; AI is not a substitute |
| Customer-facing content | Brand, accuracy and compliance review |
Build the habit
- Keep a short checklist beside your prompts
- Save examples of errors you catch so your team learns the patterns
- Ask for sources, then verify them rather than trusting them
- Follow your organisation's rules — see our one-page AI use policy
Where it is taught: IITM Pravartak's Applied Generative AI programme covers responsible use and verification from its first module, and returns to ethical research and verifying AI analysis later; see the programme guide. Also read AI governance and agentic AI in cybersecurity.
Explore the programme
See the curriculum, fee and enrolment details on the programme page, or talk to an advisor.
Frequently Asked Questions
Why does AI invent citations?
These systems generate plausible text rather than looking facts up, so they can produce references that look real but do not exist.
What is the quickest check I can do?
Open every source the AI cites and confirm it says what is claimed, and recompute any numbers that drive a decision.
Can I trust AI for legal or medical questions?
No. Treat output as a starting point and have a qualified professional review anything high-stakes.
How do I handle privacy?
Do not enter personal or confidential data into tools your organisation has not approved, and follow data-protection obligations.