How to teach students to verify sources in one lesson

You can teach students to verify sources in one lesson, because the check is short and the evidence for teaching it is good. This 40-minute class activity needs a printed worksheet and a device for each pair, and suits a secondary class or a first-year university seminar.

It has five parts: a warm-up vote on two references, one real and one invented; a four-step method; a worksheet of six references; a debrief; and an exit ticket. The worksheet, its key and the recipe for the invented references are printed here. Near the end the same six references go into EdCitation's free Verify references, so the class watches a tool run, from the publisher's record, the check they have just done by hand.

How do I teach students to verify sources in 40 minutes?

Run six timed blocks, and hold to the timing, or the worksheet takes the whole period.

Nothing needs installing, and no step needs an account.

How do I run the warm-up?

Write two references on the board, one real and one invented, and take a vote before anyone touches a device, so that the class commits to a judgement made on looks alone.

1. Walters, W. H., & Wilder, E. I. (2023). Fabrication and errors in the bibliographic citations generated by ChatGPT. Scientific Reports, 13, Article 14045.
2. Walters, W. H., & Wilder, E. I. (2024). Citation fabrication in student essays written with generative AI. Scientific Reports, 14, Article 22310.

The first is real. The second was invented for this lesson: real authors, real journal, a title that does not exist. We searched the exact title in Crossref and on the web on 22 September 2026 and found nothing.

Count the hands, then ask three or four students why. Expect answers about looks: the article number, the familiar journal, the confident title. Write those up for the debrief. Then check both live, in the four steps, and let the class watch the second fail.

How do I make an invented reference honestly?

Build it from real parts and a title nobody wrote, check that the title does not exist, and keep a written key.

The recipe

1. Pick a real journal your students might meet, and open one issue's contents page.
2. Take two or three real authors from that issue, in the form the journal uses.
3. Write a new title that fits the journal and their field.
4. Search the exact title in quotation marks in Crossref and on the web, and try it in EdCitation's Find sources, which searches about 300 million published works. If anything close comes back from any of them, change it.
5. Give it a year, volume and pages that fit the journal. Leave the DOI off, or, for a harder item, attach a real DOI from a different paper in that journal.
6. Write the key before the lesson.

This is how the real fakes are built: Walters and Wilder (2023) found that most of the fabricated article, book and website citations ChatGPT wrote carried the names of real journals, publishers and organisations. Yours should be as good as the machine's, or the lesson is too easy.

Why this is teaching, and not fabrication

A fabricated reference in a paper asserts that the source exists and that the writer read it. An invented reference in a lesson asserts neither. It is a test item, like a wrong option in a multiple-choice question: the key exists from the start, and the class is told at the debrief which items were invented. Nobody is misled.

Never post invented entries where they could be scraped into a real reference list; those in this article are labelled in the key for that reason.

What is the four-step method?

Resolve the DOI, search the exact title in quotation marks, compare the authors and the year, then open the source and find the claim. Teach it in that order, and have students say the step before they do it.

1. Resolve the DOI. Put it after https://doi.org/. A DOI names one paper: if the page carries a different title or authors, the reference fails. On a projector it can help to show the same test a second way: give the DOI to EdCitation's Cite a source, and the reference it builds from the record names the paper the DOI actually belongs to.
2. Search the exact title in quotation marks. With no DOI, or a failed one, search the full title in quotation marks in Google Scholar, Crossref or PubMed. The exact title nearly always comes back; near misses are not a match.
3. Compare authors and year. Same authors, in the same order? Same year, volume and pages? A real paper with a wrong year is a referencing error; a title no record holds is something else.
4. Open the source and find the claim. A real paper cited for something it does not say is still a broken reference.

Reports, books and web pages are checked at the source itself: a blank result from an index that does not cover them is "could not check", not "no record". The hand method in full is in how to check whether a reference is real; for a class new to DOIs, what a DOI is is a two-minute read.

What goes on the worksheet?

Six references, a verdict column and an evidence column: six is what a pair checks well in thirteen minutes. Print this table.

The answer key

Hold it back until the debrief.

• 1 is real. The DOI opens the paper; every detail matches.
• 2 is real, with a wrong year. Published in 2019, not 2020: a referencing error, not an invented source.
• 3 is invented. Real authors, real journal, a title that does not exist: searched in quotation marks in Crossref and on the web on 22 September 2026, it returned nothing.
• 4 is invented, with a borrowed DOI. The DOI opens a different paper, entry 6. A DOI names one paper. The title was checked as 3 was.
• 5 is real. No DOI, so it must be found by title, and it is.
• 6 is real. The DOI opens exactly this paper.

What does the research say about teaching this?

Skilled checkers leave the page and look the source up; most students judge by appearance until they are taught otherwise; and a short course of lessons changes that.

Lateral reading

Wineburg and McGrew (2019) watched 45 experienced internet users think aloud as they judged live websites: 10 historians with doctorates, 10 professional fact checkers and 25 Stanford undergraduates. Historians and students often fell for official-looking logos and domain names, and read vertically, staying inside a site to judge it. The fact checkers read laterally, "leaving a site after a quick scan and opening up new browser tabs", and reached better-founded conclusions in a fraction of the time. A reference check is the same move: the reference is judged by the record it points at, not by its looks.

Breakstone et al. (2021) tested 3,446 secondary students in the United States, online, on six tasks, and they struggled on all six: 96% never discovered that a climate website had links to the fossil fuel industry, about two in three could not separate news from adverts on a popular home page, and more than half accepted an anonymous Facebook video, filmed in Russia, as "strong evidence" of US voter fraud. What misled them were weak signs: a site's look, its domain, its About page and the sheer amount it contained. That is the list your warm-up vote will produce.

McGrew et al. (2019) then randomly assigned 67 university students in a critical thinking course to two 75-minute lessons on judging online content (29 students) or to a control group (38). Tested six weeks before and five weeks after, the taught group gained significantly more than the controls. The authors call it a pilot, and it is two lessons, not one; we found no published trial of a single 40-minute lesson, so treat this plan as a first lesson and repeat the check in the next assignment. The Digital Inquiry Group publishes six free lateral reading lessons on its Civic Online Reasoning site; an account is needed to open them.

SIFT

Caulfield (2019) set out four moves for anything met online: stop; investigate the source; find better coverage; and trace claims, quotes and media back to the original context. He suggests "sixty seconds" on where something comes from before reading it. The four-step check is his fourth move made concrete for a reference: trace it to the publisher's record and to the page where the claim sits.

Why the lesson matters now

Walters and Wilder (2023) had ChatGPT write 84 short literature reviews on 42 topics, then checked each of the 636 references: GPT-3.5 had invented 55% of its references and GPT-4 18%, and among the real ones 43% and 24% carried substantive errors. Of the book chapters GPT-4 cited, 70% were fabricated. Spotting them while marking is covered in how to spot fabricated references in student work.

How do I run the debrief and the exit ticket?

Compare the pairs' verdicts with the key, draw out three lessons, and collect an exit ticket that shows whether each student can do the check alone.

The debrief

Take a show of hands for each verdict, reveal the key, and ask:

• What did you open first for entry 4, and what did it show? Entries 4 and 6 share a DOI, and it belongs to 6.
• Was entry 2 fake? No: wrong year, real paper, and markers treat the two very differently.
• Which entry needed a title search? Entry 5. A missing DOI is not suspicious; it is one step more.
• Which warm-up reason would have caught the invented entries? None. Every reason was about looks.

Write up the three lessons: a reference is judged by the record, not by its looks; a wrong detail is not an invented source; "no record" is not "could not check".

The exit ticket

Three questions on a slip, two minutes:

1. A reference has a DOI. What do you do first, and what must match?
2. A reference is not in Google Scholar. Is it invented? Give one reason it might still be real.
3. How would you check a government report?

Good answers: resolve it, and the title and authors must match; no, because reports and older sources may not be indexed; go to the organisation's own website. A "yes" to the second question is the first five minutes of the next lesson.

How does the class check the whole worksheet at once?

In the last five minutes, paste the whole worksheet into EdCitation's Verify references and let the class watch the six verdicts appear together. It is the best tool for this moment because it does what the lesson taught, at the speed a lesson needs: every entry is looked up in the publisher's record and none is ever written. The worksheet itself shows why that matters. Entries 4 and 6 carry the same working DOI, so a check that asks only whether a DOI opens something passes both; the lesson's check compares the paper found with the whole reference, and so does the tool's. A chatbot cannot stand in here, because asked whether a reference is real it produces more text rather than a lookup.

The verdicts come in three kinds, verified, "check this" (a doubtful match) and not found; a retracted paper would be flagged; and "could not check" stays separate from "not found", which is the distinction the exit ticket just tested. Six references took pairs thirteen minutes by hand. The tool is free, needs no account and nothing installed, and runs in the browser on a phone.

Have the class compare the tool's verdicts with their own before you say anything, and explain any difference from their evidence.

The lesson costs nothing to run. Students who later want more for their own papers will find the plans on the pricing page: Pro is $8 a month and includes References from a file; Max is $24 a month and adds the Library. For a whole school or department, the Institution licence covers every student, with sign-in and LMS integration, training and one invoice.

Quick questions

Is it dishonest to show students an invented reference?

No. An invented reference in a lesson is a test item, revealed at the debrief with a written key, and it asserts nothing to anyone. A fabricated reference in a paper asserts that a source exists and was read.

Does one lesson change how students check sources?

The best evidence we read is for two 75-minute lessons, after which McGrew et al. (2019) found significantly greater gains on tests of online evaluation than in controls. No trial of a single 40-minute lesson was found, so repeat the check in the next assignment.

What if a real reference comes back not found?

Ask which index looked, and whether it should hold that kind of source. Reports, books, theses and articles in small or older journals are often outside Crossref and PubMed and still real, so check them at the source.

Can students do the check on a phone?

Yes. Resolving a DOI and searching a title work in any phone browser, and EdCitation's Verify references runs in the browser with nothing to install and no account.

How should students fix a reference that fails the check?

If the paper is real and a detail is wrong, as with entry 2, they rebuild the reference from the paper's own record, which EdCitation's Cite a source does from a DOI or a title. If the source was invented, the claim it carried has no support until they find a real paper that makes it, read it, and cite that.

References

• Breakstone, J., Smith, M., Wineburg, S., Rapaport, A., Carle, J., Garland, M., & Saavedra, A. (2021). Students' civic online reasoning: A national portrait. Educational Researcher, 50(8), 505-515. https://doi.org/10.3102/0013189X211017495
• Caulfield, M. (2019, June 19). SIFT (The four moves). Hapgood. https://hapgood.us/2019/06/19/sift-the-four-moves/
• Digital Inquiry Group. (n.d.). Teaching lateral reading. Civic Online Reasoning. https://cor.inquirygroup.org/curriculum/collections/teaching-lateral-reading/
• McGrew, S., Breakstone, J., Ortega, T., Smith, M., & Wineburg, S. (2018). Can students evaluate online sources? Learning from assessments of civic online reasoning. Theory & Research in Social Education, 46(2), 165-193. https://doi.org/10.1080/00933104.2017.1416320
• McGrew, S., Smith, M., Breakstone, J., Ortega, T., & Wineburg, S. (2019). Improving university students' web savvy: An intervention study. British Journal of Educational Psychology, 89(3), 485-500. https://doi.org/10.1111/bjep.12279
• Walters, W. H., & Wilder, E. I. (2023). Fabrication and errors in the bibliographic citations generated by ChatGPT. Scientific Reports, 13, Article 14045. https://doi.org/10.1038/s41598-023-41032-5
• Wineburg, S., & McGrew, S. (2019). Lateral reading and the nature of expertise: Reading less and learning more when evaluating digital information. Teachers College Record, 121(11), 1-40. https://doi.org/10.1177/016146811912101102