Methods · Working with the tool itself · Short guide
K8
Citation chasing
Follow every reference to its source before believing it: killing trust in fluent fake footnotes.
When to use it
- An answer arrives with neat footnotes you have not opened.
- You are building a bibliography and the titles sound almost too perfect.
- A citation supports exactly the sentence you needed, which is a yellow flag.
What to ask AI
Type something like: “Your answer cited: [ ]. I will try to open it. Do not invent replacements if it’s fake; help me with search queries only.”
After you paste what you found, ask whether it supports the use that was claimed.
Keep the hard part on your side: you open the source.
How it helps
Fluent fake footnotes are common enough to deserve a dedicated habit.
Citation chasing kills trust in references that exist only as decorative text.
Even real papers get mis-summarised. Opening the page is the check.
This is epistemic safety more than a memory method, and it still protects your assessed work.
How to run it
- Copy the citation exactly as given.
- Try to open it via library search, DOI, or publisher page.
- If it does not exist, do not ask the model for a replacement citation. Ask for search queries only.
- If it exists, read the relevant section and judge support for the claim.
- Mark unverified citations as unusable in your draft until opened.
Example exchange
You cited Smith and Lee (2019) in the Journal of Learning Sciences on spacing schedules. I’m going to try to open it. If it’s fake, don’t invent a replacement, just help me with search queries.
Try the author names plus “spacing” and the year in your library search, and a DOI lookup if I gave one. I’ll wait for what you find.
No such article. The closest hit is a different review.
Then drop my citation. Use the review you found only if it actually supports your sentence, don’t keep my fake footnote for atmosphere.
Copyable prompt
Your answer cited: [ ].
I will try to open it. Do not invent replacements if it's fake: help me search
queries only. After I paste what I found, judge whether it supports your use.
The Tell
Here is how you know this method has flipped: your bibliography is longer than your opened-PDF folder.
Unopened citations are decoration, not evidence.
Delete or verify before you submit.
Principle evidence
Strength of the underlying learning idea, not a claim about AI products.
The underlying learning idea is rated moderate. Opening cited sources and checking that they support the claim is basic scholarly practice. As a learning outcome, gains are more about epistemic safety than memory. Principle moderate for citation hygiene; AI-era chase-the-fake-footnote drills lack strong trial evidence.
AI delivery evidence
Whether an AI tutor delivers this method well is a separate question.
Models can invent plausible citations. Delivery quality for “help me find the real paper” is uneven and speculative as pedagogy. Your open-and-read step is the method.