Do Citations Make Nonfiction More Trustworthy? What the Research Shows
Writers add citations to be believed, but how much do they do for the reader? Studies from neighboring fields suggest readers treat them as a quick signal, rarely check them, and gain little from piling them up. This note sets out what the evidence shows, where it runs out, and what a writer can reasonably take from it.


Seventeen of 23 students in one study said references were an important sign that a Wikipedia article could be trusted. Then the researchers replaced the references with ones that had nothing to do with the article's topic. Six students noticed.
That's a small study, and its limits come up below. But it points at a question nonfiction writers rarely ask before deciding how many sources a page needs: what does a citation actually do for a reader who isn't checking it?
Where this evidence comes from, and where it stops
None of these studies tests a book. They're experiments on Wikipedia articles, chatbot answers, textbook passages and persuasive messages, and the ones I found don't measure citation density in book-length nonfiction. What follows shows how readers behave in neighboring settings. Whether it carries over to a full manuscript is a judgment call, and I've marked the places where I'm making one.
Evidence last checked: October 28, 2026.
What the evidence shows
Readers use references as a quick signal, and rarely inspect them
In the Wikipedia study (Lucassen, Noordzij and Schraagen, 2011), the researchers also varied the length of the reference lists. Nine of 23 students noticed. The authors read the pattern as heuristic judgment: readers responded to the presence of a reference list more than to its content, and systematic checking showed up mainly among participants who already distrusted Wikipedia.
A 2025 experiment found the same habit in a different setting. Ding and colleagues had 303 participants ask a chatbot questions, then rate their trust in short answers carrying no citations, one citation, or five. Citations raised trust. Participants checked them, measured by hovering over the link, in fewer than one in ten of the answers that had them (193 of 1,976), though 83 of the 197 participants in the citation groups checked at least once. Participants who did check a citation tended to rate the answer lower, which fits the idea that people who trust an answer see no need to verify it.
One well-chosen citation did about as much as five
In the same chatbot study, average trust on a ten-point scale was 7.7 with no citation, 8.1 with one and 8.2 with five. The gap between one and five wasn't statistically significant. Relevance mattered more than count: answers with randomly assigned citations were rated lower than answers with relevant ones, and when participants did check the random citations, those answers weren't rated significantly higher than answers with none. The authors suggested diminishing returns as one explanation.
Dense in-text citation can cost some readers comprehension
Stiegler-Balfour and colleagues (2020) had people read an expository passage with or without APA-style parenthetical citations, then take a comprehension quiz. Less skilled readers scored lower and read faster. Skilled readers matched the no-citation scores, but only by slowing down. The authors questioned the use of in-text citations in textbooks aimed at general student audiences.
Facing the counterargument beats ignoring it, as long as you answer it
This one comes from persuasion research rather than from studies of citation. O'Keefe's (1999) meta-analysis compared three ways of handling opposing views: one-sided messages; two-sided messages that raise the opposition and refute it; and two-sided messages that raise it and leave it hanging. The refuting version was most persuasive, one-sided messages came next, and the unanswered two-sided messages did worst. The refuting version's advantage didn't hold for advertising topics, where few studies exist. A 1990 experiment by Allen and colleagues, testing an earlier meta-analysis, found consistent effects across 17 messages.
One qualifier matters. A 2021 review by O'Keefe and Hoeken, covering 1,149 studies of 30 message-design choices, found real but small differences between options (median mean effect around r = .10), and the on-average better option didn't win reliably every time. Treat the sidedness result as a tendency, not a rule.
What I take from it
Everything in this section is interpretation. The studies don't say it; I'm reading it off them.
Relevance looks like the lever. Going from no citation to one made a difference; going from one to five didn't; irrelevant citations weakened the effect. A citation seems to earn its place by supporting the exact sentence it's attached to, in a way a reader could confirm.
The cost is real for some readers. If your audience includes people who struggle with dense text, parenthetical interruptions carry a measurable price, at least in the textbook setting. Moving source detail out of the sentence is a reasonable response, though nobody here compared notes against in-text citations directly, so that part is inference.
Be careful with the line that heavy referencing shows weak thinking. It gets repeated often, and none of this supports it. These studies measured trust ratings and comprehension, not the quality of an author's reasoning. The evidence supports something narrower: beyond a well-chosen source, extra citations bought no additional trust in one experiment, and dense in-text citations slowed or confused some readers in another.
Last, the counterargument finding is worth acting on, modestly. A source that complicates your claim, followed by your answer to it, does a different job from a third source that agrees with you.
Limits to keep in view
The samples are small and specialized: 23 students in one study, 303 online participants in another. The chatbot study was published in the 2025 AAAI conference proceedings. It measured trust with a single self-reported rating, which isn't the same as a reader's willingness to act on what they've read. The persuasion research spans many message types, advertising included, and measures persuasion, not authority in a long-form argument.
Two checks you can run on a chapter
These come out of the findings above. They aren't research results themselves.
Cover the name of the authority in each citation and read the sentence as a claim. If the reasoning around it carries the point and the source supplies lineage, the citation is doing its job. If the sentence stops making sense without the name, the claim was resting on the source rather than on you.
Then go through the citations as the reader who checks them would. Does the source support this sentence? In the chatbot study, answers whose random citations were checked did no better than answers with none.
For a fuller pass, the research and fact-checking section of the Manuscript Readiness Checklist covers sources, quotations and claims that may have changed over time. Step 7 of Before You Start Writing a Book covers deciding what needs a source before you start researching.
A citation is a promise that the claim can be checked. Most readers won't collect on it. The evidence suggests it still pays to be able to honor it, and that a fifth citation adds less than getting the first one right.
Sources
Allen, M., Hale, J., Mongeau, P., Berkowitz-Stafford, S., Stafford, S., Shanahan, W., Agee, P., Dillon, K., Jackson, R., & Ray, C. (1990). Testing a model of message sidedness: Three replications. Communication Monographs, 57(4), 275–291. https://doi.org/10.1080/03637759009376203
Ding, Y., Facciani, M., Joyce, E., Poudel, A., Bhattacharya, S., Veeramani, B., Aguinaga, S., & Weninger, T. (2025). Citations and trust in LLM generated responses. Proceedings of the AAAI Conference on Artificial Intelligence, 39(22), 23787–23795. https://doi.org/10.1609/aaai.v39i22.34550
Lucassen, T., Noordzij, M. L., & Schraagen, J. M. C. (2011). Reference blindness: The influence of references on trust in Wikipedia. Proceedings of ACM WebSci '11. https://research.utwente.nl/en/publications/reference-blindness-the-influence-of-references-on-trust-in-wikip/
O'Keefe, D. J. (1999). How to handle opposing arguments in persuasive messages: A meta-analytic review of the effects of one-sided and two-sided messages. Communication Yearbook, 22, 209–249. https://doi.org/10.1080/23808985.1999.11678963
O'Keefe, D. J., & Hoeken, H. (2021). Message design choices don't make much difference to persuasiveness and can't be counted on, not even when moderating conditions are specified. Frontiers in Psychology, 12. https://doi.org/10.3389/fpsyg.2021.664160
Stiegler-Balfour, J., Jakobsen, K., Stroud, M., & Daniel, D. (2020). APA-style citations can create a roadblock to textbook comprehension for less skilled readers. Teaching of Psychology, 47(2), 147–155. https://doi.org/10.1177/0098628320901384
This resource was written by Victor and published by Focarix as part of the Focarix Knowledge Library.
© 2026 Focarix. All rights reserved. Readers and researchers are welcome to read, study, and share key insights. When quoting, referencing, or citing this research note, please attribute Focarix as the publisher and include a direct link back to the original resource:
Working on something similar?
If you're developing a book, manuscript, or body of knowledge and want help with structure, editing, or publishing, tell us where you are.