A Q&A field guide for business writers who quote research, and would like to keep quoting it next year

Every week a white paper somewhere announces that “studies show” something. Sometimes the study shows it. Often the study shows something smaller, the press release shows something bigger, and the LinkedIn post shows something that would make the original authors wince. Here is how to be the person in the meeting who actually read the thing.

Part 1: Where the Distortion Starts

Q1. If the headline got the science wrong, was it the reporter’s fault?

• Usually the error was already baked in upstream. Petroc Sumner and Chris Chambers at Cardiff University checked 462 press releases from 20 leading UK universities against the papers they described and 668 news stories (BMJ, 2014).
• 40% of the press releases contained exaggerated advice, 33% exaggerated causal claims, and 36% stretched animal findings to humans. When the release exaggerated, news followed it 58%, 81% and 86% of the time. When the release was accurate, those rates fell to 17%, 18% and 10%.
• The punchline: the hype did not reliably buy more coverage. Universities were inflating claims for, as far as the data could tell, nothing.

Q2. So can I cite the press release if I’m in a hurry?

• Please don’t. A 2012 PLoS Medicine cohort (Yavchitz et al.) followed 70 randomized trials. Spin showed up in 40% of abstract conclusions, 47% of press releases and 51% of news items. Spin in the abstract made spin in the press release 5.6 times more likely.
• Readers of the press release alone overestimated the benefit in about 27% of cases. You are playing telephone, and you are the fourth person in line.

Figure 1 and 2. Exaggeration mostly originates before the journalist types a word.

Part 2: Finding the Actual Paper

Q3. The article says “a Harvard study found.” Where is the study?

• Hunt for the DOI, the string that starts with 10. and pins a paper to one permanent address. Press releases bury it near the bottom. Paste it after https://doi.org/ and you land on the version of record.
• No DOI? Search the author’s name plus two distinctive words in Google Scholar and click “All versions,” which often surfaces a free PDF. Still stuck? Email the author. Researchers are startlingly happy to hear someone wants to read their work.

Tool	What it does for you	Where
doi.org	Resolves a DOI to the publisher’s official version	doi.org
PubMed / PMC	Free index of biomedical papers; PMC has full text; flags retractions	pubmed.ncbi.nlm.nih.gov
Google Scholar	“All versions” and “Cited by” show free copies and who pushed back	scholar.google.com
Unpaywall	Browser add-on that finds a legal free copy of paywalled papers	unpaywall.org
Retraction Watch DB	Largest retraction database, now open via Crossref	retractiondatabase.org
PubPeer	Post-publication critique; image and data problems often surface here first	pubpeer.com
medRxiv / bioRxiv / SSRN / arXiv	Preprint servers; check for a “now published in” link	medrxiv.org, ssrn.com

Table 1. A free toolkit that covers most of what business writers will ever need.

Part 3: Reading the Methods Section Like You Mean It

Q4. Do professionals really misquote papers, or is this an amateur problem?

• Professionals do it a lot. A 2015 PeerJ meta-analysis of 28 studies (Jergas and Baethge) found 25.4% of citations in medical journals did not support the claim attached to them, with 11.9% rated major errors. A 2017 recalculation put it at 14.5%. The 2025 update, covering 46 studies and 32,000 quotations, landed at 16.9%, half of them major. History journals come in at 24.27%.
• Translation: in a typical paper with 40 references, somewhere between 6 and 10 of them say something other than what the author claims. Copying a citation from another article inherits that risk.

Q5. I opened the methods. What exactly am I looking for?

Question	Why it matters	Red flag
Who or what was studied?	Mice, 40 undergrads and 10,000 employees tell very different stories	“Consumers” who turn out to be 212 online survey takers
How many?	Small samples produce big, fragile effects	Fewer than 100 people and a dramatic headline
Randomized or observational?	Only experiments earn causal verbs	“Causes” or “boosts” from a correlation
Relative or absolute effect?	“Doubles risk” can mean 1 in 10,000 becoming 2 in 10,000	Percent change with no baseline
Preregistered?	Plans filed in advance limit fishing for results	Outcomes that differ from the registered plan
Who paid?	Funding shapes questions and framing	Sponsor also sells the product studied

Table 2. Six methods questions that take ten minutes and prevent most bad citations.

Q6. Why be this fussy? Published results hold up, don’t they?

• Less often than you would hope. The Reproducibility Project: Cancer Biology (eLife, 2021) set out to repeat 193 experiments from 53 high-impact papers. It managed only 50 experiments from 23 papers, partly because zero protocols were described well enough to repeat without help.
• Where it could compare, replication effects were 85% smaller at the median, and 92% came in smaller than the original. Only 46% passed more criteria than they failed. Psychology’s 2015 effort replicated just 36% of 97 significant findings.

Figure 3 and 4. Citations drift from their sources, and flashy findings shrink on a second look.

Part 4: Preprints Versus Peer Review

Q7. Can I cite a preprint?

• Yes, with a label. Preprints are papers posted before peer review, and during COVID they carried roughly 25% of all pandemic science. They are mostly solid: Brierley et al. (PLOS Biology, 2022) found conclusions changed in 7.2% of non-COVID preprints and 17.2% of COVID ones by publication. A study of ICU therapy preprints saw 19% change conclusions.
• Good phrasing: “a preprint posted to medRxiv in March, not yet peer reviewed, found…” Then set a reminder to check for the journal version.

Q8. And peer review makes it bulletproof?

• Peer review works like a smoke detector with a weak battery. On May 22, 2020, The Lancet published a hydroxychloroquine study claiming data on 96,032 patients from 671 hospitals on six continents. The WHO paused a trial arm. The data company, Surgisphere, could not produce the data, and the paper was retracted 13 days later.

Part 5: Checking Whether It Still Stands

Q9. How common are retractions, really?

• More common every year. More than 10,000 papers were retracted in 2023, a record, per a Nature analysis. More than 8,000 came from Hindawi journals hit by paper mills. The retraction rate more than tripled in a decade, passing 0.2% of 2022 papers. Those Hindawi papers had already been cited over 35,000 times.

Q10. Once a paper is retracted, surely people stop citing it?

• They keep going. Studies find only 3% to 8% of later citations mention the retraction, and one analysis of 13,000+ biomedical citation contexts found about 5%. Even the 25 retracted papers of anesthesiologist Scott Reuben, a famous fraud, were flagged by only 40% of citers. Wakefield’s 1998 vaccine paper took 12 years to retract.

Q11. Does this happen in business research, or only in medicine?

Case	The claim	What happened
Shu et al., PNAS 2012	Signing honesty pledges at the top of a form cuts cheating	Retracted 2021 after Data Colada showed field data were fabricated; long a staple of nudge talks
Francesca Gino, Harvard Business School	Star research on honesty and rule-breaking	Misconduct found in four papers; tenure revoked May 2025, a first at Harvard since the 1940s. She denies it and is suing
Surgisphere, Lancet 2020	Hydroxychloroquine raised death risk	Data could not be verified; retracted in 13 days
Hindawi / Wiley, 2023	Thousands of routine journal papers	8,000+ retracted; Wiley retired the Hindawi brand
Wakefield, Lancet 1998	Vaccines linked to autism	Retracted 2010; still cited decades later

Table 3. Famous findings that aged badly. Each one appeared in confident business or policy writing.

Figure 5 and 6. Most preprints hold up; most retractions go unnoticed by the people citing them.

Q12. What is the 60-second check before I hit publish?

1. Link the paper’s DOI instead of the press release.
2. Read the abstract, the methods and the limitations paragraph.
3. Write down who was studied and how many.
4. Match your verb to the design: “linked to” for observational, “caused” only for experiments.
5. Give absolute numbers alongside percentages.
6. Label preprints as preprints.
7. Search the DOI in the Retraction Watch database and PubPeer, and check the publisher page for notices.
8. Check the date. A 2009 finding may have been overturned twice since.

Do this and you become the rare writer whose footnotes survive contact with a skeptical reader. In business writing, that is a competitive advantage hiding in plain sight.

Sources

1. Sumner et al., BMJ 2014, exaggeration in press releases: https://doi.org/10.1136/bmj.g7015
2. Bratton et al., follow-up on causal overstatement (28% to 13%): https://pubmed.ncbi.nlm.nih.gov/32500096/
3. Yavchitz et al., PLoS Medicine 2012, spin in press releases: https://doi.org/10.1371/journal.pmed.1001308
4. Jergas & Baethge, PeerJ 2015, quotation accuracy meta-analysis: https://peerj.com/articles/1364/
5. Mogull, PLOS ONE 2017, recalculated quotation error rate: https://doi.org/10.1371/journal.pone.0184727
6. Baethge & Jergas, Res Integr Peer Rev 2025: https://pmc.ncbi.nlm.nih.gov/articles/PMC12285159/
7. Quotation accuracy in history journals, Scientometrics 2023: https://doi.org/10.1007/s11192-023-04755-w
8. Errington et al., eLife 2021, Reproducibility Project: Cancer Biology: https://elifesciences.org/articles/71601
9. Brierley et al., PLOS Biology 2022, preprint to publication changes: https://doi.org/10.1371/journal.pbio.3001285
10. Fraser et al., PLOS Biology 2021, preprints in COVID science: https://doi.org/10.1371/journal.pbio.3000959
11. Morin et al. 2023, ICU therapy preprints vs. publications: https://doi.org/10.1177/08850666231182563
12. Review of 152 COVID RCT preprints vs. journal versions: https://pmc.ncbi.nlm.nih.gov/articles/PMC12543410/
13. The Lancet, retracted Surgisphere study: https://doi.org/10.1016/S0140-6736(20)31180-6
14. Van Noorden, Nature 2023, 10,000+ retractions: https://www.nature.com/articles/d41586-023-03974-8
15. Why retracted articles keep getting cited, Scientometrics 2024: https://doi.org/10.1007/s11192-024-05147-4
16. Teixeira da Silva, Learned Publishing 2025: https://doi.org/10.1002/leap.1667
17. Heibi & Peroni, citations to the Wakefield paper: https://pmc.ncbi.nlm.nih.gov/articles/PMC8338205/
18. Harvard Crimson on Gino tenure revocation: https://www.thecrimson.com/article/2025/5/27/gino-tenure-revoked/
19. Retraction Watch database: https://retractiondatabase.org