Students · The RAFT Framework for Ethical AI Use
Fact checking AI
Practice four checks before relying on an AI output: locate the source, compare the quotation, examine the claim, and recalculate the result.
Open resource · No email required
Exercise 1 A plausible citation
An AI answer claims that a study proves later library hours increase grades. It gives this citation: Chen, M., & Ortiz, L. (2026). Evening library access improves academic achievement. Journal of Campus Learning, 9(2), 40–52.
Exercise 1 — Compare your response with the model
Your submitted response
Model response and explanation
The authors, year, title, publication, volume, issue, and pages differ. The supplied original describes library seat counts and does not measure grades. It cannot support the AI claim about academic achievement.
In a real assignment, search the exact title through your library catalog, the claimed publisher, or a scholarly index. Open the original and match its bibliographic details. Check that it addresses the claim and that any identifier resolves to that same work.
Do not relabel the supplied record as the AI’s citation or invent missing details. An unsuccessful search alone does not prove that a source does not exist. Ask a librarian or instructor when needed; remove unsupported claims before submission.
Compare the reasoning
Does your response distinguish an unverified citation from proof that a source does not exist? Did you identify mismatched details and explain why seat counts cannot support a claim about grades?
This model is one defensible response, not a wording template. Compare evidence and reasoning rather than matching phrases.
Exercise 2 A misleading quotation
AI output: The report concluded that “the workshop caused the increase.”
Exercise 2 — Compare your response with the model
Your submitted response
Model response and explanation
The AI removed “We cannot tell whether,” converting uncertainty about causation into a confident conclusion. Matching a few words is not enough; the quotation must preserve the source’s meaning.
You could also quote the complete second sentence. Keep the population and uncertainty visible. In a real paper, cite the original passage and the location required by your citation style.
Compare the reasoning
Does your corrected sentence retain both the observed rise among volunteers and the uncertainty about its cause? Different wording is fine if the meaning is preserved.
This model is one defensible response, not a wording template. Compare evidence and reasoning rather than matching phrases.
Exercise 3 A claim beyond the evidence
AI output: “The workshop caused a nine-point improvement for all students and will produce the same result in every course.”
Exercise 3 — Compare your response with the model
Your submitted response
Model response and explanation
The mean increased by 61 − 52 = 9 points among these volunteers. The summary does not establish that the workshop caused the change. Other influences and repeat testing could matter. A self-selected group in one setting does not establish an effect for every student or course.
Compare the reasoning
Did you report the nine-point change for these volunteers, avoid a causal conclusion, and avoid generalizing to all students or courses? Point to the study-design detail that supports your caution.
This model is one defensible response, not a wording template. Compare evidence and reasoning rather than matching phrases.
Exercise 4 Percent or percentage points
AI output: “Success rose from 30% to 40%, which is a 10% relative increase. The lesson definitely caused the improvement.”
Exercise 4 — Compare your response with the model
Your submitted response
Model response and explanation
| Calculation | Result |
|---|---|
| Before | 12 ÷ 40 × 100 = 30% |
| After | 16 ÷ 40 × 100 = 40% |
| Percentage-point change | 40% − 30% = 10 percentage points |
| Relative increase | (40 − 30) ÷ 30 × 100 ≈ 33.3% |
The AI’s starting percentages are correct, but it confuses a percentage-point difference with a relative percentage increase. Four additional students succeeded. The counts show a change; without stronger evidence, they do not establish that the lesson definitely caused it.
Compare the reasoning
Compare your denominators, units, and rounding with the model. Did you separate a 10 percentage-point change from a 33.3% relative increase, and avoid claiming that the lesson definitely caused it?
This model is one defensible response, not a wording template. Compare evidence and reasoning rather than matching phrases.
Use these checks on your own work
Locate the original source, preserve its meaning, stay within the evidence, and verify calculations. If a claim cannot be verified, remove it or narrow it to what you can support.
By Dr. Chase Cookson · v2.1 (009019026)
Course, assignment, and institutional rules take priority.