QWE AI Academy By the end of this tutorial you’ll have a repeatable, four-prompt workflow for pulling apart any viral science headline using ChatGPT (or Claude, or Gemini – same idea). We’ll use the trending brain-aging nasal spray story as the live case study, because it’s everywhere right now and it’s a near-perfect example of hype outrunning the actual paper.
Spoiler: ten minutes with the right prompts and you’ll know the spray works – in 18-month-old mice – and that no human has received a dose. That gap is the whole story. No headline made it obvious.
The 30-second background
In February 2026, a team at Texas A&M led by Dr. Ashok Shetty published findings in the Journal of Extracellular Vesicles (DOI: 10.1002/jev2.70232). Two intranasal doses of tiny particles – extracellular vesicles loaded with microRNAs, derived from human induced pluripotent stem cell neural stem cells – reached aging mouse brains via the olfactory pathway. Effects appeared within weeks. They lasted months. The April 2026 Texas A&M announcement confirmed the work was funded by the National Institute on Aging.
That’s the finding. Real, peer-reviewed, NIA-backed. What happens in your feed is something else: headlines quietly drop the word “mice,” and the press-release metaphors take over.
Two ways people are trying to make sense of this
Method A – scroll the coverage. Read five articles, form a mental average. Every summary borrows the same brain-as-engine metaphor from the Texas A&M press release. The “18-month mouse ≈ 60-year-old human” line appears verbatim across outlets, none of them questioning whether that’s biology or just lifespan arithmetic.
Method B – hand the paper to an LLM. The study is open access. Download the PDF, drop it into ChatGPT or Claude, and skip the game of telephone entirely. You can ask questions no journalist asked, because no journalist had ten minutes and the full document open simultaneously.
Method B is faster for depth. The only catch: you need the right prompts, or the LLM defaults to summarizing what was measured – exactly like a news article.
The 4-prompt workflow
Here’s the exact sequence. Copy, paste, adapt for any study.
Prompt 1 – Get the paper, not the press release
Find the primary source for this claim: "Nasal spray reverses brain aging."
I want the DOI and journal name of the peer-reviewed paper, not
a press release or news summary. If you find it, give me the exact
title and the open-access PDF link.Ask for a DOI and you get one: 10.1002/jev2.70232, Journal of Extracellular Vesicles. Ask vaguely and you get eurekalert.org – a press release aggregator. That distinction is the entire point of Prompt 1. ChatGPT with browsing sometimes surfaces the press release first even when it knows the paper exists; naming the DOI explicitly forces it past that default.
Prompt 2 – Force it to state what the study is not
I've attached the full PDF. In plain English, tell me:
1. What species was tested?
2. Sample size and sex breakdown?
3. Has any human received this treatment?
4. What did the study NOT measure that a reader might assume it did?Question 4 is the one that changes everything. LLMs answer it well – but only if you ask. Without that prompt, they default to summarizing what was measured, and you’d get the same output as a news article. The answer to question 3, by the way, is: no human. Ever.
The phrase worth memorizing: “What did the study NOT measure?” It works on any paper, any LLM, any subject. I’ve used it on GLP-1 studies, gene-editing claims, longevity supplement trials – same pattern every time.
Prompt 3 – Translate the mechanism without the metaphors
Explain the mechanism in one paragraph, no metaphors, no analogies.
Use the actual biological terms. Then in a second paragraph, tell me
which steps are established science and which are novel to this paper.The paper’s actual mechanism: vesicles suppressed the NLRP3 inflammasome, p38/MAPK, cGAS-STING-IFN-1, and JAK-STAT signalling pathways in the aged hippocampus. Single-cell RNA sequencing showed microglial transcriptomes reprogrammed seven days post-treatment – proinflammatory genes silenced, oxidative phosphorylation genes upregulated (per WebProNews summary of the paper, mid-2026). “Brain runs cooler” is the press release version. These are the actual targets.
Prompt 4 – Stress-test the mouse-to-human leap
News coverage says 18-month-old mice are "equivalent to 60-year-old humans."
Explain the limits of that equivalence. What has failed to translate from
mouse models to humans in the past, specifically in neuroscience?
Give me 3 concrete examples.That “equivalent” framing is lifespan arithmetic, not biology. Mouse microglia and human microglia differ in gene expression, density, and response patterns. Ask ChatGPT for the neuroscience-specific failure list and you’ll get Alzheimer’s drug after Alzheimer’s drug – decades of mouse-positive, human-negative results. The equivalence line in every headline is a useful shorthand that became invisible after enough repetition.
Running these four prompts the first time, the thing that genuinely surprised me: the LLM’s answer to Prompt 4 was more sobering than anything I’d read in ten news articles combined. Not because the AI is smarter – because I finally asked the right question.
Four things no viral summary mentioned
The production bottleneck is real. These vesicles come from human induced pluripotent stem cells. Standardized production under good manufacturing practices – GMP-grade, the regulatory baseline for any human therapeutic – isn’t solved yet. That’s not a small engineering footnote; it’s the reason comparable EV therapies have stalled before reaching trial (Doolly analysis, June 2026).
Both sexes, equally. The treatment worked across male and female mice at the same level – Drug Target Review flagged this in May 2026 as unusual. Many neuroscience findings have collapsed when tested in female animals after male-only pilots. This one didn’t. Worth tracking as the work continues.
The dosing schedule is mouse math. Two doses is what worked in 18-month C57BL/6J mice evaluated at 20.5 months. Nobody knows what a human protocol looks like. Assume everything about dose, frequency, and delivery changes.
Patent filed ≠ product coming. Texas A&M filed a U.S. patent. Human trials are years away pending preclinical safety work. If you see a nasal spray for sale with this branding, it isn’t this study.
Why four prompts beat ten articles
Every outlet covering this story worked from the same Texas A&M press release. Same metaphor. Same Shetty quote. Same mouse-age framing, unchallenged. That’s not a criticism – it’s just how science coverage works at speed.
The workflow also generalizes. Swap “nasal spray” for any GLP-1 headline, any gene-editing claim, any longevity supplement – the four prompts still apply. The hype cycle runs the same script every time. The antidote is the same four questions.
FAQ
Can I buy this nasal spray?
No. It exists in a research lab. Only mice have received it.
Which LLM should I use for reading scientific papers?
Anything with PDF upload and a large context window – Claude, ChatGPT Plus, or Gemini Advanced all handle 20-40 page papers. The open-access PDF for this study falls in that range. One practical thing: before you trust a summary, ask the model to confirm how many pages it actually processed. If it ingested 12 of 28 pages and didn’t tell you, the summary has gaps you can’t see.
Will the LLM hallucinate details from the paper?
Yes, it can – especially on specific figures or p-values. The fix is simpler than people expect: ask for direct quotes with page numbers, then spot-check two or three against the PDF. If the quotes are accurate, the interpretation usually is too. If they’re not, you’ll catch it before it matters. This check takes about 90 seconds and it’s the difference between trusting the model and verifying it.
Next step
The open-access PDF is at doi.org/10.1002/jev2.70232. Open a fresh ChatGPT conversation, upload it, run Prompt 2. Ten minutes from now you’ll know more about this study than most people sharing it – and you’ll have a workflow that works on the next viral paper too.