Evaluate AI Responses

AI can produce useful ideas and explanations, but it can also provide information that is inaccurate, incomplete or biased. It may sound confident even when it is wrong.

You are responsible for checking any AI-generated information you use.

Check claims against reliable sources

Do not treat an AI response as evidence. Verify important claims using trustworthy sources such as:

• Books and library databases

• Government and educational websites

• Established news organisations

• Academic journals

• Museums, universities and recognised institutions

• Original or primary sources

Check names, dates, quotations, statistics and other details carefully.

Watch for AI hallucinations

An AI hallucination occurs when an AI tool produces information that appears believable but is incorrect or invented.

AI may create:

• Facts that are inaccurate

• Events that never happened

• Quotations that no one said

• Statistics without supporting evidence

• Books, articles or websites that do not exist

• References that look genuine but are fabricated

Always open and examine a source before citing it. Confirm that it exists and that it genuinely supports the claim.

Look for errors and omissions

An answer can be misleading even when parts of it are correct.

Ask yourself:

• Is any information inaccurate?

• Has important context been left out?

• Is the explanation oversimplified?

• Does the response confuse fact with opinion?

• Are its examples relevant and appropriate?

• Does the evidence actually support its conclusions?

• Is the information current enough for my purpose?

Identify bias

AI systems learn from large collections of existing information. This material may contain stereotypes, inequalities and biased assumptions, which can appear in AI-generated responses.

Consider:

• Does the response favour one position?

• Does it make assumptions about a person or group?

• Does it use stereotypes or generalisations?

• Does it present an opinion as though it were a fact?

• Does its wording influence how the reader views the issue?

Look for missing perspectives

One AI response cannot represent every experience or point of view.

Ask:

• Whose voices are included?

• Whose voices are missing?

• Are local, cultural or multilingual perspectives represented?

• Are people directly affected by the issue heard?

• How might someone from another community view the question?

• Do reliable sources disagree about the subject?

Seek additional sources that introduce different experiences and interpretations.

Compare and verify

Use this process whenever you evaluate an AI response:

1. Identify the claims that need checking.

2. Search for reliable and relevant sources.

3. Compare the AI response with the evidence.

4. Investigate errors, bias and missing perspectives.

5. Revise your understanding using what you have verified.

6. Cite the reliable sources—not the AI’s unsupported claims.

Questions to ask yourself

Before using information from an AI response, check:

• Can I confirm this information elsewhere?

• Are the supporting sources real and reliable?

• Does each source actually support the claim?

• Is the information accurate and current?

• What important information may be missing?

• Which perspectives are represented?

• Which perspectives should I investigate further?

• Can I explain why I trust this information?

Remember

AI can help you begin an investigation, but it should not be the end of your research.

Question the response, verify its claims and use reliable evidence to reach your own conclusions.