How to Verify a Source Before Citing It: A Checklist

# How to Verify a Source Before Citing It: A Checklist for Checking Tool-Suggested References

A reference that arrives from an AI assistant or a reference manager is a lead, not a source. This checklist sets out six gates. Each one checks a different part of a suggested citation — the identifier, the record, the retraction status — before the entry reaches your reference list. It is written for final-year undergraduate and postgraduate students in Uganda who are writing a first independent study.

Why a Suggested Citation Is Not Yet a Source

Tools that retrieve and summarise the literature produce output that needs checking before you rely on it [S2]. Researchers can now assess how well automated citation output matches real and relevant references, which tells you the output is not automatically correct [S6]. That is not the same as saying tools invent references every time. They do not. The point is narrower: the tool has produced a string that looks like a citation, and you have not yet confirmed that the thing it describes exists as described.

The working rule for this page is simple. A citation produced by a tool is a lead to be checked, not evidence in itself. That rule is this page's own position, not a finding from a study.

Keep one distinction straight. Lang writes about when to cite in scientific publication, not how to verify that a source exists [S1]. That source helps with the first half of the question only. The second half — whether the item is real, correctly described and citable — is what the gates below address.

This checklist does not judge whether a source is the best one for your argument. It judges whether the source is real, whether the record matches the suggestion, and whether you can cite it honestly. Choosing the right source is a separate task, covered in How to Find Credible Academic Sources for a Literature Review.

Before You Start: Keep the Tool's Output Unedited

Copy the full suggested entry into a scratch file before you change anything. Include the DOI or URL, all authors in the order given, the year, the title, the journal or publisher, the volume, issue, pages and article number. Edit the suggestion first and you lose the ability to compare it against the record later.

Record where the suggestion came from: which tool, which prompt, and the date. Automated citation output is something to be assessed rather than trusted [S6], and a note of its origin lets you see later whether one tool is producing more failures than another. The record-keeping step is this page's recommendation; the source supports the background idea, not the procedure.

Note your access limits now. Many journals your tool suggests sit behind subscriptions that your university may not hold. A failed check may mean you could not open the record, not that the source is fake. Write that down so you do not misread your own notes later. No supplied source covers access barriers to scholarly literature in Uganda or East Africa, so this point comes from practice rather than a citation. If you are unsure what your library can reach, ask at the reference desk before you assume a source is unavailable.

Gate 1 — Does the Identifier Resolve?

Take the DOI and resolve it at its registration agency rather than typing the title into a search box. When the landing page opens, compare the title and the first author against the suggestion. If they match, the identifier points at a real record.

Check the identifier's format as well. A DOI that is malformed, or that returns nothing at all, is a stop signal. It is not a formatting problem to be tidied up later.

If this check fails: do not "fix" the DOI by guessing characters or adding digits. Treat the entry as unverified. Move to Gate 2 only if you can find the item by title in a library database.

This gate describes standard DOI resolution. None of the supplied sources covers DOI mechanics or registry procedure, so no citation is attached. If you need an authoritative statement, ask your library for the registry guidance it uses.

Gate 2 — Does the Record Match the Suggestion?

Open the publisher's page or the database record and compare it field by field against the suggestion you saved. The fields to compare are:

• authors, in the order given
• year of publication
• title, word for word
• journal or publisher
• volume, issue, pages and article number

Mismatches are common and usually small. Expect a wrong year, a swapped author order, a journal title confused with a similar one, or a page range that does not appear on the record. No supplied source reports how often these errors occur, so treat that list as the kinds of difference to look for, not as measured frequencies. The nearest supplied source, on assessing how well language-model output cites references [S6], should not be stretched to support specific error types unless you have read what it actually reports.

If this check fails: correct the entry to match the record, not the tool. If the record itself is ambiguous — two similarly titled papers by the same authors in the same year, for example — keep both entries until you have read the abstracts and can tell them apart.

Gate 3 — Is the Source Retracted, Withdrawn or Corrected?

Check three places for a notice: the publisher's article page, the database record, and any linked notice from the publisher or a retraction service. This is the gate most often skipped, and the one where a wrong citation does the most damage. A retracted finding can undermine a paragraph you have already written.

The supplied sources do not name a specific retraction database, and this page will not invent one. Confirm current retraction-checking practice with your university library, which will know which services it subscribes to. If your library subscribes to none, say so in your notes rather than assuming the source is clean.

Learn the difference between the notice types. Each has a different consequence for your text:

• Retraction. The findings should not be relied on. Remove the citation, or cite it only with an explicit note that it has been retracted.
• Expression of concern. The record is under review. Cite with caution, and say so if the claim matters to your argument.
• Correction. The record stands with an amended detail. Cite the corrected version.
• Withdrawal. The item was removed before final publication. Do not cite it as a published source.

If this check fails: remove the citation, or cite it only with an explicit note that it has been retracted. Then check whether any sentence in your draft depends on that source, and revise those sentences too.

No supplied source defines these categories or describes retraction-notice practice, so this gate carries no citation. It is the page's own description of the check.

Gate 4 — Can You Actually Read Enough of It to Cite It?

Confirm that you have opened the source itself, not only its metadata. Abstract-only access is a limit you should state in your notes. It is not a licence to cite the full paper's findings as though you had read them.

For digital editions and online primary sources, check how the edition itself asks to be cited. Bleier addresses this directly for digital editions [S5], and it is the one clean procedural citation in this checklist. Many editions publish a preferred citation form on their own pages; use it.

If this check fails: cite only what you have read, or replace the source. Do not cite a paper you have not opened. This rule is the page's own; no supplied source states it.

Gate 5 — Does the Source Say What You Are Attributing to It?

Find the specific sentence, table or figure that supports your claim. Do not rely on a section heading or an abstract sentence that sounds close enough.

Then check the source's scope against how you plan to use it: its sample, its region, its period. This matters when you are writing about Uganda. A study conducted in another country may still be relevant, but you should not present its findings as if they described Uganda. Say where the study was done and let the comparison be explicit. Lang's argument ties citation to the point of use [S1], which is the principle behind this step. The step-by-step procedure is this page's own.

If this check fails: rewrite your sentence to match what the source says, or drop the citation. Agosto and Anderton write about the habit of attributing claims to sources [S7]. That is useful background for why this matters, but it does not supply a procedure, so it is cited here only as background.

Gate 6 — Is the Citation Formatted From the Verified Record?

Build the final reference entry from the checked record, not from the tool's output. Rebuild it in APA 7 using the publisher's page as your source of truth. Two existing pages cover the formatting rules: APA 7 Reference List Rules Students Actually Need and APA 7 Referencing Explained: In-Text Citations, Reference Lists, and Common Errors.

Check whether your department or supervisor requires a different style. APA 7 is common, but it is not universal, and a reference list rebuilt in the wrong style is still wrong. Ask before you format a whole chapter.

Keep a one-line verification note per source: the date you checked it, and what you compared. This is for your own records and for your supervisor if they ask how a reference was confirmed.

If this check fails: a formatting error is fixable; a record error is not. Return to Gate 2 and compare the fields again.

Quick Troubleshooting: Common Failures and What They Mean

Use this table when you arrive mid-problem and need to know which gate to return to.

This table restates the gates above and carries no citation.

What to Do With a Source That Fails Verification

Remove the failed entry from your reference list and from any sentence that depends on it. A reference that fails Gate 1 or Gate 2 has no place in the list, and a sentence built on it has to be rewritten or removed with it.

Replace it with a verified source on the same point. The page on finding credible academic sources sets out how to search for one. Start with your university library's own catalogue and its subscribed databases. A librarian there can also tell you what the library can and cannot reach.

Record the failure in your notes. A pattern of failures from one tool is information about the tool, not about you. Automated citation output is something to assess [S6], and your own log of failures is one way to do that assessment over a project.

Where This Fits in Your Wider Work

Run the gates when a source enters your notes, not when the reference list is assembled. A citation checked at the point of entry is cheap to fix. One checked at the end of a chapter may sit under three paragraphs that need rewriting.

Pair this checklist with Using AI in Academic Research Responsibly and How to Paraphrase Without Plagiarising: A Checklist. The first covers how to use tools without handing them your judgement; the second covers what to do once a source has passed these gates.

A closing note on limits. This checklist reduces the risk of citing a fabricated or misdescribed source. It does not guarantee that a source is appropriate, current or the best available for your question. No supplied source gives Uganda-specific guidance, local repository procedures or a named retraction database, and this page has not implied otherwise. Where a gate rests on this page's own recommendation rather than on a cited source, it says so. Costs, subscription access and the specific services your library holds vary between institutions. This page states no figures for any of them because none are known here.

The general idea that verification is a separate step from production appears in other fields too — in engineering design [S3], in news verification [S4] and in audit [S8]. Those sources are from other domains and are noted here only as an analogy. They say nothing about citation practice, and they are not relied on above.

Sources used

• [S1] Lang, Thomas A. (2023). "Where did this come from? When (not how) to cite sources in scientific publications." European Science Editing. doi:10.3897/ese.2023.e94153
• [S2] Jin, Qiao; Leaman, Robert; Lu, Zhiyong (2023). "Retrieve, Summarize, and Verify: How Will ChatGPT Affect Information Seeking from the Medical Literature?" Journal of the American Society of Nephrology. doi:10.1681/asn.0000000000000166
• [S3] Schock, Alfred (1980). "Design evolution and verification of the general-purpose heat source." doi:10.2172/5355852
• [S4] Butera, Alberto; Staszczak, Julia; Mikalef, Patrick; Gatteschi, Valentina (2025). "Blockchain-Based Source-to-Source News Verification." IEEE Annual Computers, Software, and Applications Conference (COMPSAC). doi:10.1109/compsac65507.2025.00235
• [S5] Bleier, Roman (2021). "How to cite this digital edition?" Digital Humanities Quarterly. doi:10.63744/8tkmuutbe3pu
• [S6] Wu, Kevin; Wu, Eric; Wei, Kevin; Zhang, Angela; et al. (2025). "An automated framework for assessing how well LLMs cite relevant medical references." Nature Communications. doi:10.1038/s41467-025-58551-6
• [S7] Agosto, Denise E.; Anderton, Holly (2007). "Whatever Happened to 'Always Cite the Source?'" Reference & User Services Quarterly. doi:10.5860/rusq.47n1.44
• [S8] Griffith, Emily Elaine; Hammersley, Jacqueline S.; Kadous, Kathryn (2014). "Audits of Complex Estimates as Verification of Management Numbers: How Institutional Pressures Shape Practice." Contemporary Accounting Research. doi:10.1111/1911-3846.12104

Citation check notes: Every bracketed citation ([S1]–[S8]) corresponds to a supplied key, and each is used in a way the source can plausibly support. No citation was removed or rewritten. Two points to flag for your review:

1. S3, S4 and S8 are cited only as an explicit cross-domain analogy, and the text says so. If you would rather keep the reference list strictly to citation-practice sources, delete the closing analogy paragraph and drop S3, S4 and S8 from the list.
2. S5 is described as "the one clean procedural citation in this checklist." That is accurate as written, but check that Bleier's article does in fact give citation guidance for digital editions before publication — the supplied metadata gives the title and identifier only, not the content.

Related reading

• Using AI in Academic Research Responsibly: Accuracy, Citations, Originality, and Academic Integrity
• APA 7 Reference List Rules Students Actually Need