Why checking comes before repeating
Students now meet generated answers everywhere: in search results, in study apps, in chat tools, and in summaries a classmate pasted into a group chat. These answers are often fluent and sometimes correct. The problem is that fluency and correctness look the same on the screen. Responsible AI literacy does not mean refusing to use these tools. It means treating every answer as a draft you have not yet checked, and knowing a short routine for checking it.
This guide gives that routine in four steps. Each step takes a few minutes. Together they turn an answer you are borrowing into an explanation you can stand behind or, just as useful, an answer you decide not to use.
Step 1: Restate the claim in your own words
Close the window and write the main claim in one or two sentences of your own. If you cannot do it without looking, you do not yet understand it well enough to repeat it. That is not a failure; it is information. It tells you to read more slowly, ask a narrower question, or go back to your textbook before going further.
Restating also separates the claim from the decoration. A generated answer may wrap one key idea in several paragraphs of confident framing. Your restatement should keep the idea and drop the framing. For example, a long answer about why the sky is blue might reduce to: “Shorter wavelengths of sunlight are scattered more by air molecules, so more blue light reaches our eyes from all directions.”
Step 2: Find one source you could show a teacher
Look for one source you could put in front of a teacher: a textbook section, a reference work, a page from a recognised institution, or a published article. Check that the source actually says what the answer said, not just something nearby. Note the title and where you found it. If the AI tool provided a link or a reference, open it; generated references are sometimes wrong, incomplete, or invented.
If you cannot find any source that supports the claim, treat it as unverified. You can still mention it as a question you are investigating, but not as a fact.
Step 3: Test it on an edge case
Ask what the claim predicts in a situation slightly different from the example. If shorter wavelengths scatter more, what should the sky look like at sunset, when light passes through more air? If a rule about fractions is true, does it still hold for negative numbers or zero? Trying an edge case is how scientists and mathematicians find the limits of a statement, and it is something students can do with a pencil.
Sometimes the edge case reveals that the original answer was too broad. That is a good result. Your explanation becomes narrower and more accurate, and you learn where the idea stops working.
Step 4: Say what would make it wrong
Finish by writing one sentence that begins “This would be wrong if …”. For a science claim, that might be an observation that contradicts it. For a historical claim, it might be a primary source with a different date. For a summary of a book, it might be a passage that says the opposite. Naming what would make a claim wrong is the clearest sign that you understand what it actually says.
This step also protects you in conversation. When someone challenges you, you already know which kind of evidence would change your mind, and you can discuss it calmly instead of defending a borrowed sentence.
What never to paste in
Part of checking is deciding what you should not ask a tool in the first place. Do not paste in personal information about yourself or others, private messages, school assessment materials you were told to keep confidential, or anything you would not want stored by a company you do not know. When in doubt, ask a teacher or parent first.
Checking is not the same as permission
A checked answer is still not automatically your own work. Your school’s rules on AI use and academic integrity decide what you may submit, and how you must disclose any help. If a teacher says no AI for an assignment, a checked AI answer is still not allowed. If AI help is allowed, say what you used and how, and keep your notes from the four steps. They show the thinking that is yours.
A worked example from homework
Suppose a student asks a chat tool why leaves change colour in autumn and gets a confident paragraph. Step one: the student restates it as “Leaves stop making green chlorophyll as days shorten, so other pigments that were already there become visible.” Step two: the biology textbook’s section on photosynthetic pigments supports the main idea, and the student notes the page. Step three: the edge case is evergreen trees and tropical places with little seasonal change. The answer said nothing about them, so the student narrows the claim to deciduous trees in places with marked seasons. Step four: the claim would be wrong if leaves with no yellow or orange pigment at all still turned yellow in autumn. The final explanation is shorter than the generated paragraph, it has a source, and the student can defend every word of it.
Make it a weekly habit
Try the routine once a week on one answer from any subject. Keep a short log: the claim, your restatement, the source, the edge case, and the “would be wrong if” sentence. After a month, look back. Most students find that the habit changes how they read everything, not only AI answers. For more, see why AI literacy starts with explanation, the parent briefing on what student AI tools cannot do, and noticing when you do not understand.