Quick answer: keep every source, but get the paper's name tag out of your sentence
In an article for regular readers, a citation has 2 jobs.
First, it shows "this claim is backed by something." Second, it lets anyone who wants to check go straight to the original.
The trouble starts when you try to be thorough and drop the authors, the full English paper title, the journal, the year, the DOI and the URL right into the middle of a sentence. Now the citation itself is what makes the article hard to read.
Like this:
Short numeric citations may help comprehension (Cowan, Merritt, & Mikan, 2020, The FASEB Journal, 34, 1-1, doi:...).
To a regular reader, it feels like a room full of researchers suddenly started swapping business cards mid-sentence.
My suggestion: keep the sentence and the bibliography apart.
Short citation markers may be easier to read than long author-and-year citations. [Study 1]
Then only the people who care open [Study 1]:
Study 1: an experiment on citation style and comprehension
30 university students read science papers with numbered citations and with "author + year" citations. Comprehension scores were higher with numbers. It was small and only reported as a conference abstract, so you can't claim "numbers always boost understanding by 34%." [^1]
That way, the article is where you understand, and the citation box is where you check.
1. Why academic citation style reads so badly on the web
Academic citation formats exist so researchers can pin down exactly which study is meant.
That isn't what a regular reader wants to know first.
What they want first is:
- What did they look at?
- Who did they study?
- What did they find?
- How much should I trust it?
So the details researchers need and the explanation a regular reader needs do different jobs.
An experiment with 650 people found that jargon itself gets in the way of taking in science information. The more jargon, the harder the content felt to process, and that feeling carried over into later judgments. [^6]
Citations can do the same thing. The reader needs the meaning, but if a pile of English names and journal titles cuts in, they're also stuck asking "Who's this? Do I need to read this right now?"
Citations matter. Just don't park the evidence warehouse in the hallway.
2. Keep in-text markers short — but don't treat "numbers win" as a law
A small 2020 study compared citation styles head to head. 30 science undergraduates read research papers, and with numbered citations like [1] they scored 34% higher on comprehension tests than with "author + year" citations. The gap was even bigger for hard texts. [^1]
But that's one small study with 30 people, published as a conference abstract.
So you can't write:
Numbered citations always raise understanding by 34%!
The safe takeaway is:
In the text, cut any citation detail the reader doesn't need to follow the meaning.
On a website, any of these works:
- [1]
- [Study]
- [Source]
- a superscript footnote number
What matters isn't which symbol you pick. It's that it doesn't break the flow, and clicking it tells you what the evidence is.
3. The best tool: an "evidence card" that explains a study in 5 plain points
A link to the original paper only tells a regular reader that evidence exists.
So put an evidence card in between.
5 points are usually enough.
What was the study about?
Example: an experiment on whether citation style affects how well people understand a science paper.
What kind of study was it?
Example: a randomized crossover experiment.
If that's too technical, say:
The same people tried both styles, and their scores were compared.
How many people or studies?
Example: 30 science undergraduates.
What was the main result?
Example: comprehension scores were higher with numbered citations.
What's the big limitation?
Example: few people, only university students, and only a conference abstract, so be careful about generalizing.
After that, add:
- original title
- authors
- journal
- year
- DOI
- PubMed / link to the original
Reader-friendly info on top. The researcher ID details underneath.
That's the basic rule.
4. Don't write "click here" — tell people where the link goes
The W3C accessibility guidelines say the link text alone, or together with the text around it, should make the purpose of the link clear. A bare URL usually isn't enough. [^2][^3]
So instead of:
More details here
this is better:
See the 2026 review on short-form video
Even better:
See the 2026 review of 42 studies (PubMed, in English)
Before clicking, the reader already knows:
- what's there
- what kind of source it is
- where it goes
- that it's in English
Research on hypertext found that showing a short preview of the linked content first lowered mental load and improved learning. [^4]
Telling people where a link goes isn't a click-rate trick. It keeps readers from getting lost.
5. Put the citation right after the claim it backs up
Bad:
Research has found a lot. [1] This method may help understanding. Another study also showed an effect on trust.
Which part does [1] cover? No idea.
Good:
A small experiment found numbered citations easier to understand than author-and-year ones. [1]
In news stories, a link to the original research was something readers used to judge whether to trust it. [2]
The rule: right after the claim it supports.
Don't make one citation hold up several different claims.
Different studies, different citations.
What matters isn't how many citations you have. It's whether each claim clearly matches its evidence.
6. In everyday articles, an accurate paraphrase beats a long quote
Pasting long chunks of the English paper looks precise.
But regular readers usually need the meaning, not the original word order.
Here's what I'd do:
- Summarize accurately in your own words
- Keep the numbers, conditions and uncertainty
- Add a short direct quote only if it helps
- Leave the full original to the source link
For example, instead of dropping in:
“The presence of jargon impairs people's ability to process scientific information...”
write:
Experiments show that heavy jargon makes science information harder for regular readers to take in. [Study]
It reads much better. [^6]
But when you summarize, never:
- turn a correlation into cause and effect
- turn "may" into "proved"
- stretch the result to people who weren't studied
- inflate the size of the effect
Writing simply and claiming more strongly are two different things.
7. In news, just having a link builds trust — but few readers click through to the paper
A qualitative study of parents who read health news found that statistics and links to scientific studies worked as clues for judging whether the news was credible. But not many actually clicked through to the original paper. [^5]
Another online news experiment also found conditions where hyperlinks affected how credible the news felt and how willing people were to look for more. [^7]
That leads to an important design point.
Readers aren't necessarily saying:
Let me read the whole original paper right now.
More likely they're saying:
If there's real evidence, just make sure I can check it myself when I want to.
So an everyday article doesn't need to be the front lobby of an English research paper.
The evidence is visible. The details open only for people who want them.
That's the sweet spot.
8. Don't stop at "studies show" — add one line on how strong the evidence is
Even with a citation, readers can't weigh it if they don't know what kind of study it was.
One short line covering these goes a long way:
- one study, or a review pulling many studies together?
- an experiment, or an observational study?
- how many people or studies?
- a correlation, or a difference caused by an intervention?
- the biggest limitation
Example:
An analysis of 71 studies covering about 98,000 people found a link. But if most of those studies were observational, you can't say one thing caused the other.
Not "it's a meta-analysis, so it's settled," and not "it's one study, so it's worthless."
Give readers enough to roughly judge how heavy the evidence is.
9. Even "plain-language" research summaries end up hard to read — so actually read them
A 2025 study looked at 1,241 plain-language summaries attached to UK research reports. Many were too hard for regular readers and full of jargon. [^8]
And a large 2026 crowdsourced evaluation found that AI-written plain-language summaries could be rated "easy to understand" yet still lead to worse scores on real comprehension questions than summaries written by people. [^9]
In other words:
The AI produced natural-sounding text
and
Readers actually understood it
are not the same thing.
The same goes for citations. Don't approve an auto-generated evidence card just because a machine score says it's fine.
Read it and ask: "Could a middle schooler who's never heard of this study say what it looked at and what it found?"
10. Writing in many languages: don't translate the paper title — localize the explanation
On a multilingual site like this one, which publishes in 12 languages, split the evidence card into 2 layers.
The layer you localize
- what the study was about
- how it was done
- how big it was
- the main result
- the limitation
- the link text
Write these in everyday words in each language.
The layer you keep as-is
- original paper title
- author names
- journal name
- year
- DOI
- PubMed ID
- URL
This information identifies the study, so don't translate or change it.
For example, even in the French version, the original title can stay in English.
Just put a short French explanation in front of it, like:
A small experiment in which 30 students compared 2 citation styles
An English title isn't the problem. The problem is when you have to read English to understand why the evidence matters.
11. Before / after: a citation that wrecks the sentence vs. one moved into an evidence card
Before
Jargon hinders science understanding (Bullock, O. M., Colón Amill, D., Shulman, H. C., & Dixon, G. N. (2019). Jargon as a barrier to effective science communication: Evidence from metacognition. Public Understanding of Science, 28(7), 845–853. doi:10.1177/0963662519865687).
Correct. But for an everyday article, it suddenly gets heavy.
After
An experiment with 650 people found that heavy jargon makes science information harder for regular readers to take in. [Study]
[Study] An experiment on jargon and understanding science
650 people read science information with or without jargon. With jargon, the information was harder to process. It's one experiment, so you can't assume the effect is the same size for every text and every reader.
Original paper: Jargon as a barrier to effective science communication: Evidence from metacognition
Bullock et al., 2019 / Public Understanding of Science / DOI: 10.1177/0963662519865687[^6]
There's actually more information now.
But the sentence got lighter.
This isn't "cutting a long text." It's "putting an elevator in a long text."
12. A citation checklist before you publish (including AI-written drafts)
Before publishing, check at least this:
- No long English paper titles, full author lists, journal names, DOIs or raw URLs cutting into the text
- Each citation marker clearly shows which claim it supports
- No spots that just say "studies show" without saying what the study was
- Each evidence card covers what, how, how big, result and limitation
- Original title, authors, journal, year, DOI and URL are unchanged
- No correlation turned into cause and effect
- No link text that's just "here," "source" or "details"
- Where useful, the link says what kind of source it is, what's in it and what language it's in
- No single citation lazily covering several different claims
- Direct quotes no longer than needed
- Evidence card explanations read naturally in each language
- Bibliographic details kept exactly as in the source
- A real read-through, not just automated checks
If it's safe to fix, do detect → fix → check again in the same pass.
13. Bottom line: a citation should be a door to the evidence, not a speed bump
Citing in an everyday article isn't a game of pasting a paper's reference list into your text.
Here's the 4-floor building I recommend.
Floor 1: the text
Readers get the answer they came for, in plain words.
Floor 2: a short marker
They can see "this claim has evidence behind it."
Floor 3: the evidence card They understand what the study was, how it was done, how big it was, what it found and its limits — in their own language.
Floor 4: the original paper Title, authors, journal, DOI, PubMed and URL for anyone who wants to check.
Don't cut the evidence.
Don't cut the accuracy.
Just don't make readers carry every bit of researcher metadata around with them.
Even citations should go easy on working memory.
If a paper title jumps out into your sentence, kindly send it back to the evidence warehouse.
References (9)
[^1]: Cowan M, Merritt EK, Mikan V. Citation Format Matters: Undergraduate Student Comprehension of Peer-Reviewed Scientific Manuscripts. The FASEB Journal. 2020;34(S1). Small randomized crossover study, n=30. DOI: https://doi.org/10.1096/fasebj.2020.34.s1.06442
[^2]: W3C WAI. G91: Providing link text that describes the purpose of a link. https://www.w3.org/WAI/WCAG21/Techniques/general/G91
[^3]: W3C WAI. Understanding Success Criterion 2.4.9: Link Purpose (Link Only). https://www.w3.org/WAI/WCAG21/Understanding/link-purpose-link-only.html
[^4]: Antonenko PD, Niederhauser DS. The influence of leads on cognitive load and learning in a hypertext environment. Computers in Human Behavior. 2010;26(2):140-150. DOI: https://doi.org/10.1016/j.chb.2009.10.014
[^5]: Maggio LA, Krakow M, Moorhead LL. ‘There were some clues’: a qualitative study of heuristics used by parents of adolescents to make credibility judgements of online health news articles citing research. BMJ Open. 2020;10:e039692. DOI: https://doi.org/10.1136/bmjopen-2020-039692
[^6]: Bullock OM, Colón Amill D, Shulman HC, Dixon GN. Jargon as a barrier to effective science communication: Evidence from metacognition. Public Understanding of Science. 2019;28(7):845-853. DOI: https://doi.org/10.1177/0963662519865687
[^7]: Borah P. The Hyperlinked World: A Look at How the Interactions of News Frames and Hyperlinks Influence News Credibility and Willingness to Seek Information. Journal of Computer-Mediated Communication. 2014;19(3):576-590. DOI: https://doi.org/10.1111/jcc4.12060
[^8]: Lang IA, King A, Boddy K, et al. Jargon and Readability in Plain Language Summaries of Health Research: Cross-Sectional Observational Study. Journal of Medical Internet Research. 2025. PubMed: https://pubmed.ncbi.nlm.nih.gov/39805102/
[^9]: Guo Y, Sohn JH, Leroy G, Cohen T. Are LLM-generated plain language summaries truly understandable? A large-scale crowdsourced evaluation. Journal of Biomedical Informatics. 2026;179:105038. PubMed: https://pubmed.ncbi.nlm.nih.gov/42044823/
