02 / Everyday practice / Lesson 4 of 8
Check before you trust
Find an unsupported claim and correct it using the supplied evidence.
01 / Understand
Compare claims with evidence
Break an answer into statements you can check. A summary may contain a correct date and an invented price in the same sentence. Mark each important statement as supported, contradicted or not established by the information available. An answer that is mostly correct still needs its unsupported parts removed or qualified.
Open the source
A link is a starting point, not proof. If you use an outside source, open it and check that it supports the exact claim. Check who published it, when it applies and whether the answer confuses an estimate with a confirmed result. For a calculation, work it out independently with an appropriate tool rather than asking the same answer to reassure you.
Know when to stop
If you cannot establish a claim, write 'not provided' or leave it out. Do not fill a gap just to make the result look complete. For decisions that affect health, legal rights or important finances, a beginner exercise is not a substitute for relevant expertise.
02 / Worked example
Look closely at the result
A written teaching example, not a live AI response.
Fictional notice: The workshop is on 12 October, from 2 pm to 4 pm. There are 18 places. The fee has not been announced.
The free workshop runs on 12 October, 2-4 pm, with 20 places.
Compare with the tutor’s review
The date and time are supported. 'Free' is not established: no fee has been announced. Twenty places contradicts the notice, which says 18. Correct both before sharing.
03 / Your turn
Put it into practice
Audit the example sentence by sentence. Write a corrected announcement and label the fee as unknown.
Use fictional information. Nothing you write here is sent to a server or an AI. Download your work before leaving; it is not saved automatically.
04 / Check your understanding
One decision before you go
Take this with you
Check each important claim. Preserve uncertainty when evidence is missing.
Complete the practice, review each criterion and check your answer to finish your self-review. Every lesson remains open.
Next: Protect your informationSources & course notes
Reviewed . Examples and exercises are original Aulafy teaching material under MIT. These references inform the course’s approach to reliability and learning; they are not endorsements.
- NIST: Generative AI risk profileBackground on unreliable outputs and evaluating risks.
- UNESCO: Generative AI in educationBackground on human agency, privacy and learning.