AgriXiv All articles
Peer Review & Infrastructure

Forty-One Seconds and a Misread Table: Anatomy of a Food Science Preprint's TikTok Summer

AgriXiv
Forty-One Seconds and a Misread Table: Anatomy of a Food Science Preprint's TikTok Summer

Photo: Wesley Fryer from Oklahoma City, Oklahoma, USA, CC BY 2.0, via Wikimedia Commons

The paper went up on a Monday. For twenty-six days, it behaved like most AgriXiv postings in food science: a few hundred downloads, a handful of citations in reference managers, one thoughtful comment from a retired USDA chemist, and silence.

Then, at 7:40 p.m. Eastern on 12 October, a home-cooking creator with 2.1 million followers posted a 41-second video. By Friday it had 8 million views. By the following Tuesday the preprint had 24,000 downloads, three television bookings for the author, and a controversy it was never designed to have.

The paper

Acrylamide in Air-Fryer Potatoes: Batch-Level Measurements Under Domestic Cooking Conditions was posted on 14 September 2026 by a team of food scientists at a Midwestern land-grant university and a state public health laboratory. It reported 96 batches across three potato cultivars, cooked at temperatures from 160 to 200 degrees Celsius for 8 to 25 minutes, with acrylamide measured by liquid chromatography–tandem mass spectrometry.

The numbers were unremarkable to anyone in the field. At 160 degrees for 12 minutes, mean acrylamide came in near 190 micrograms per kilogram. At 200 degrees for 20 minutes, the mean was about 1,180, with one batch at 2,400. The paper's conclusion was that cooking time and surface browning mattered more than the appliance itself, and that consumer guidance emphasizing lighter browning remained sound.

Buried in the discussion, the authors wrote a sentence they would spend six weeks repeating: 'These are measurements, not risk estimates.'

What the video said

The creator read the abstract on camera, paused on the maximum value, and told viewers the potatoes contained 'three times the legal limit.' That sentence is wrong in two directions. There is no US legal limit for acrylamide in food; the Food and Drug Administration publishes guidance for industry on reducing acrylamide formation, not a threshold of legality. The European benchmark levels that the phrase likely gestured at are exactly that — benchmarks, used as a supervisory tool, not safety thresholds, and not applicable in the United States.

The distinction between a hazard and a risk, which is the entire grammar of toxicology, did not survive the edit.

The traffic

The server's usage dashboard tells a clean story. Cumulative downloads sat at 212 on 11 October. They passed 24,000 by 15 October and 61,000 by 1 November. Referral logs showed social platforms accounting for roughly 78 percent of October traffic. The paper's public comment thread, previously dormant, accumulated 140 entries in two weeks, including several from credentialed food chemists attempting corrections and several from viewers demanding to know why scientists had hidden the data.

No one had hidden anything. The dataset was deposited at posting, under an open license, with a README.

What the authors did

On 15 October, the corresponding author posted a two-page plain-language summary as a linked document, stating in the first paragraph that the study measured a chemical's presence and did not estimate anyone's risk. On 17 October, she recorded a three-minute video of her own. On 30 October, the team posted version two, adding a section headed 'What this study does not show,' a revised table that placed reference values in explicit context, and a note explaining that no measurements had changed.

By then the correction had reached perhaps a tenth of the original audience. This is not a failure of diligence. It is the arithmetic of amplification. A correction travels through the same channels or it does not travel at all, and channels move at the speed of the most shareable claim.

What food scientists say

A food toxicologist at a Big Ten institution put the problem bluntly in an interview: the paper's largest value was its least representative one, and the abstract presented it without a comparison class. That is a fair criticism of the manuscript, and it is exactly the kind of criticism peer review tends to surface. But it is also a criticism the server's comment thread surfaced within nine days, where almost no one saw it.

A science communication researcher offered a sharper framing. Preprint servers, she argued, have built excellent infrastructure for discovery and almost none for reception. Journals at least have press offices, embargoes, and a stable publication date that reporters can plan around. A preprint can be discovered by anyone, at any hour, with no intermediary and no context.

What the infrastructure could look like

None of the fixes require gating, and none should be mistaken for censorship.

Plain-language summaries at posting, not after. The authors wrote theirs in three days under pressure. A structured summary field — what was measured, in what system, against what comparison, with what limits — would have been composed in calmer circumstances and read by more people than the abstract.

Author-moderated comment layers. The 140-entry thread contained real expertise. It needed structure: a pinned author response, a distinction between technical and general comments, and a visible marker for entries that had been addressed.

Amplification notes. A short, machine-readable record of where a paper has traveled, maintained by the server and linked from the landing page, so a second-wave reader arrives knowing they are in a second wave.

Creator-facing guidance. Newsrooms get press notes. Independent creators with larger audiences than most newsrooms get nothing. A one-page guide on reading effect sizes and confidence intervals would be cheap to produce and frequently ignored, which is still better than absent.

Public version diffs. The team's version two was a model of transparency. Almost no one could find it.

The uncomfortable part

The temptation, after an episode like this, is to conclude that certain work should have stayed offline until reviewed. That conclusion is wrong on the evidence. The acrylamide measurements were sound, the data were open, and the authors were more candid about uncertainty than the video that carried them. The failure was distributive: the paper reached a mass audience stripped of the caveats its authors had carefully written, and the server had no mechanism designed for that moment because it had never been designed for mass audiences at all.

A preprint server cannot control a 41-second video. It can make sure that the second thing a curious viewer finds is the authors' own voice, in their own words, at the top of the page.

All Articles

Related Articles

The Reviewer's Dilemma: Agricultural Peer Review Is Running on Borrowed Time

The Reviewer's Dilemma: Agricultural Peer Review Is Running on Borrowed Time

She Kept the Receipts: Rachel Carson's Working Files and the Case for Documented Uncertainty

She Kept the Receipts: Rachel Carson's Working Files and the Case for Documented Uncertainty

Thirty-Four Citations, One Without a Journal: How a Rangeland Carbon Preprint Reached BLM Guidance

Thirty-Four Citations, One Without a Journal: How a Rangeland Carbon Preprint Reached BLM Guidance