Use TOK to examine AI-generated claims
Separate fluent output from evidence and investigate how tools shape knowledge and trust.

Identify the claim rather than the polish
An AI-generated response can sound convincing while containing an unsupported or inaccurate statement. Begin by extracting the specific claim you would need to justify. Distinguish factual information, inference and recommendation. Ask what evidence would establish each. The TOK issue is how a tool's presentation influences confidence and what checking process supports a reasonable judgment. Do not evaluate an output only by whether it agrees with you or uses technical language. Fluency is a feature of communication, not sufficient evidence that the underlying claim is sound.
Trace evidence beyond the output
When an output names a source, find and inspect the source yourself. Check that it exists, that the relevant passage or data actually supports the claim and that its context has not been changed. If no source is provided, decide what independent material would be needed. Keep a record of the claim, check and resulting judgment. This process can reveal missing evidence as well as errors. An unverified claim should remain unverified; repeated generation does not become independent confirmation merely because the wording changes.
Examine how the tool shapes attention
Consider what information the response selects, omits or frames as important. A summary may privilege familiar explanations while excluding a less prominent but relevant perspective. Compare the output with material from appropriate sources and ask which assumptions guide its organization. Avoid concluding that every omission is deception; limited space and purpose also matter. The analytical question is whether the selection is justified for the task and what a reader would reasonably infer. Connect this issue to methods, tools and perspectives in the current framework.
Try a worked verification scenario
Imagine a student receives an AI summary claiming that a school transport change reduced journey times for everyone. The student inspects the cited report and finds an average reduction, with increased times on some routes. The revised judgment distinguishes a network average from individual experience. The TOK analysis asks how compression into a simple statement altered the scope of the claim and encouraged overconfidence. This is an illustrative scenario, not a reported study. Its value lies in showing a checking decision and the knowledge distinction that follows.
Consider responsibilities when using the result
Ask who could be affected by an inaccurate or biased claim and what level of checking the use requires. A casual brainstorming idea and a consequential decision may require different scrutiny. In assessed work, follow school rules and the IB's academic integrity policy. If AI-generated material is included, it must be acknowledged appropriately in the text and bibliography; acknowledgement does not replace student authorship or task requirements. Keep tool use transparent and verify factual material through suitable sources. Do not present generated reasoning as your own independent analysis.
Make a reasoned judgment about reliance
State what you can now trust, why and within which limits. Perhaps the tool is useful for generating search terms but unsuitable as evidence for the claim. Perhaps a summary becomes useful after correction against original material. Avoid an automatic verdict that AI is always reliable or always useless. Explain the relationship between the tool, task and verification process. For a TOK assessment, use a specific documented case or object that advances the official prompt or title, with sources and context preserved.
Your next-step checklist
Checklist choices stay in this browser.
Official references
This is an original practical guide from IBvia. The examples are illustrative; your subject guide, assessment year and school instructions determine the requirements.

