<- all tokdocs
2026.10.08@angusthenontechnical

The study is real and correctly named on screen, but the video's own Claude run produced a miss and its one negative tone signal was still untestable on the day it posted

AI
Watch on TikTok

View on TikTok ->

A 63.389 s MP4 (QuickTime/MOV container, isom/iso2/mp41 brands), 1080x1920 portrait, HEVC Main profile level 150 video at 30/1 fps and 783,006 bps across 1,901 frames, HE-AACv2 audio at 44,100 Hz stereo and 48,141 bps across 1,369 frames, 6,637,497 bytes total at an 837,684 bps overall rate, format id bytevc1_1080p_837671-1, all confirmed by ffprobe rather than taken from metadata.json. Posted at epoch 1791490814, which is 2026-10-08 20:20:14 UTC, by uploader angusthenontechnical (uploader_id 6783510017912865797) under the channel nickname "Angus the Nontechnical," credited to the track "original sound" by Angus the Nontechnical. The description is 82 characters: "Not financial advice lol but that study is very real #ai #buildwithai #vibecoding ". At capture time (metadata.json mtime 2026-10-19 00:44:03 CDT, epoch 1792388643 = 2026-10-19 05:44:03 UTC) the clip showed 13,200 views, 741 likes, 9 comments, 149 reposts and 462 saves. I read 16 of the 32 extracted frames: 001, 003, 005, 007, 009, 011, 013, 016, 019, 021, 022, 024, 025, 027, 029 and 031, which at the 2 s extraction interval sit at roughly 0, 4, 8, 12, 16, 20, 24, 30, 36, 40, 42, 46, 48, 52, 56 and 60 seconds. The transcript is 245 words across 18 segments. Frame 001 is a talking head composited over a Nike flagship storefront under a white card reading "Claude predicted Nikes downfall." Frame 003 shows a Yahoo-style quote card, "NYSE - Nasdaq Real Time Price - USD / NIKE, Inc. (NKE) / 34.36 -0.25 (-0.71%) / As of 2:08:01 PM EDT. Market Open." with a "Trade Nike on Coinbase" button. Frame 005 shows a Reuters article, "Nike stock plunges as gloomy sales forecast fans growth concerns," by Ananya Mariam Rajesh and Juveria Tabassum, dated June 28, 2024, with "stock plunges" circled in red and the body text "It was the worst day ever for the stock, and the losses wiped out $28.41 billion from the company's market valuation." Frame 007 is a grid of brand logos (Logitech, Shark, Ninja, OXO, Rubbermaid, Tide, Colgate, YETI). Frame 009 is an Amazon search results page of Logitech mice and keyboards with star ratings and review counts. Frame 011 is a Claude conversation titled "Logitech M510 review analysis" containing a table headed "Amazon review research basket" listing Logitech/LOGI, SharkNinja/SN, Helen of Troy/HELE, Hamilton Beach Brands/HBB, Newell Brands/NWL, Spectrum Brands/SPB, Sony/SONY, Procter & Gamble/PG, Colgate-Palmolive/CL, Church & Dwight/CHD, Clorox/CLX and YETI. Frame 013 shows a two-star Amazon review headed "Modern Versions Are Inferior," reviewed in the United States on September 22, 2026, scored "Tone -1," with the fine print "Reference score from the actual Claude run. Jev has not been executed." Frame 016 shows a line chart titled "Review tone over time / Logitech M510 - 94 product reviews / Scored by Claude; one-product sample," axis -2.0 to +2.0 across FY25 Q3 to FY27 Q2, footed by "Scale: -2 strongly negative to +2 strongly positive / Sparse sample. Not a company forecast." Frames 021, 024 and 025 show a bar chart titled "The negative swing / This quarter minus the prior four quarter mean / Logitech M510 - actual sample" with four bars, FY26 Q3 at about +0.25, FY26 Q4 at about +0.70, FY27 Q1 at about +0.19 and FY27 Q2 at about -0.18 in red, set beside a May 05, 2026 press release, "Logitech Announces Q4 and Full Fiscal Year 2026 Results," with "up 7 percent" circled, under the overlay lines "If the signal...", "...but next-quarter revenue rises", "Record the miss." and "Conditional that rule... Keep the company in the evaluation." Frame 022 shows the IDEAS/RePEc listing for "More than words: Can tone of consumer product reviews help predict firms' fundamentals?" by Shunyao (Cynthia) Jin, with the phrase "negative review tone is more informative and useful" circled in red. Frames 027, 029 and 031 return to a plain talking head in a beige hallway. Spoken, the argument is: build a basket of 10 to 20 Amazon-heavy companies, have an AI scrape two years of their reviews, have a smaller model score each review's tone and tag the reason, average the tone per quarter, then line that series up against quarterly SEC filings and historical stock price, because a University of Michigan study says a sudden negative tone "will almost always precede a drop in sales and stock returns." The last transcript segment ends at 63.30 s against a real audio stream duration of 63.389 s, so nothing in the transcript is timestamped past the end of the audio and no Whisper hallucination tail is present.

The study exists, is named correctly on screen, and the circled sentence is verbatim

At 9.38 s the video says "And according to the University of Michigan, these things are more predictable than we think," and at 42.24 s it says "Straight up, according to the University of Michigan study." Frame 022 puts the citation on screen rather than leaving it vague, and the citation checks out.

The paper is "More than words: Can tone of consumer product reviews help predict firms' fundamentals?" by Shunyao (Cynthia) Jin, published in the Journal of Business Finance & Accounting, volume 50, issue 9-10, pages 1910-1942. The Crossref record for DOI 10.1111/jbfa.12680 lists the author affiliation as "College of Business Fairlane Center South University of Michigan-Dearborn Dearborn Michigan USA," gives an online publication date of 2022-12-28 and a print issue date of October 2023, and classifies it as a journal article. Unpaywall reports the published version as open access under a CC BY-NC license with an OA date of 2022-12-07.

The abstract, retrieved verbatim from Crossref, reads in part: "Using a broad sample of consumer reviews posted on Amazon.com, I examine whether the linguistic tone of aggregate consumer product reviews conveys information that is associated with firms' sales, earnings, stock returns and risk. I find that aggregate review tone successfully predicts a firm's forthcoming quarterly sales. Moderating analyses show that this predictability is stronger for firms operating in a highly competitive environment. I further find that review tone predicts a firm's quarterly earnings surprises, abnormal stock returns and risk. A path analysis shows that the effect of review tone on stock prices is partially channeled through its effect on firms' earnings. I finally find that negative review tone is more informative and useful than positive tone in predicting a firm's fundamentals."

The sentence circled in red in frame 022 is the real closing finding of the paper, word for word. The video did not invent the study and did not misattribute it to a different institution in any meaningful way. One clarification: the author is at University of Michigan-Dearborn, a separate campus from the Ann Arbor flagship that most listeners will picture when they hear "the University of Michigan."

I could not retrieve the full text. Wiley Online Library returned HTTP 403 to both the DOI landing page and the direct PDF endpoint, SSRN returned HTTP 403 on the preprint record (abstract id 4291019), and the University of Michigan Deep Blue repository record (handle 2027.42/191627) sat behind a Cloudflare challenge. So the abstract above is the deepest primary text I verified, and I make no claim about the paper's tone-measurement method, its regression specifications or its sample period.

"Almost always precede a drop in sales and stock returns" is not what the paper says

At 42.24 s the video says "if you suddenly see a negative tone out of nowhere, it will almost always precede a drop in sales and stock returns." This is the one place the video genuinely oversells.

The abstract says review tone "successfully predicts a firm's forthcoming quarterly sales" and "predicts a firm's quarterly earnings surprises, abnormal stock returns and risk." That is the language of a statistically significant average association estimated across a panel of firm-quarters, with controls for review rating, review volume and review dispersion. "Almost always" is a claim about hit rate on individual cases, and the abstract contains no hit rate, no accuracy figure and no statement that the relationship holds case by case. The paper also reports that the effect is conditional: "this predictability is stronger for firms operating in a highly competitive environment." A finding that is stronger in some environments than others is by construction not a near-universal rule.

The direction the video gives is right. Tone predicts sales positively, so a tone drop maps to weaker forthcoming sales, and the paper's own closing line says the negative side of the tone distribution carries more information than the positive side. The certainty is wrong. The honest restatement is that negative swings in aggregate review tone are associated with weaker next-quarter sales on average across a large panel, more so in competitive industries.

Worth noting what the video does not do: at 40.50 s it says "see if it correlates with that swing from the average review tone," which is correlational language, and at 50.70 s it says "I am not a financial advisor, even in the slightest. Like, please talk to an actual professional." The video does not flip the correlation into a causal mechanism. It inflates a panel-average association into a near-deterministic rule, which is a different and smaller error than the usual correlation-to-causation reversal.

The video's own Claude run recorded a miss, and it says so on screen

The strongest fact in this clip is one the voiceover never mentions. Frames 021, 024 and 025 show the result of the demo the video is selling, and the result is a failure.

The bar chart "The negative swing / This quarter minus the prior four quarter mean / Logitech M510 - actual sample" has four bars. Three are positive: FY26 Q3 at roughly +0.25, FY26 Q4 at roughly +0.70, FY27 Q1 at roughly +0.19. One is negative: FY27 Q2 at roughly -0.18, rendered in red. Beside it sits the Logitech press release, and the overlay text, read across frames 021, 024 and 025, spells out "If the signal...", "...but next-quarter revenue rises", "Record the miss." and "Conditional that rule on the observed result. Keep the company in the evaluation."

The press release in frame 025 is "Logitech Announces Q4 and Full Fiscal Year 2026 Results," dated May 05, 2026, with "up 7 percent" circled. I verified that document against the primary filing. Logitech International S.A.'s Form 8-K exhibit 99.1 for Q4 FY2026, filed with the SEC, states: "Sales were $1.09 billion, up 7 percent in US dollars and 3 percent in constant currency compared to Q4 of the prior year," and for the year, "Sales were $4.84 billion, up 6 percent in US dollars and 4 percent in constant currency compared to the prior year." The circled figure is accurate.

So while the voiceover at 45.74 s is saying a negative tone "will almost always precede a drop in sales," the visual underneath is logging a case where the signal did not line up with revenue, and captioning it "Record the miss." That is unusually honest production for a finance-adjacent TikTok, and it also undercuts the "almost always" claim more directly than any outside source can.

The one genuinely negative quarter in the demo had not reported results when the video posted

The FY27 Q2 bar is the only negative swing in the video's own sample, and it is the only one that would constitute the "sudden negative tone out of nowhere" the narration describes. It was untestable on 2026-10-08.

Logitech's fiscal year ends in late March, so fiscal 2027 second quarter covers roughly July through September 2026. Logitech issued a press release, "Logitech Announces Date for Release of Second Quarter Financial Results for Fiscal Year 2027," stating that results would be released on Tuesday, October 27, 2026 at 1:00 p.m. Pacific Daylight Time, with a videoconference at 1:30 p.m. PDT. The video posted at 2026-10-08 20:20:14 UTC, nineteen days before that release.

That leaves the demo with four tone observations, three positive swings and one negative swing, one checked comparison that the video itself scores as a miss, and zero checked comparisons involving the negative swing that the entire thesis rests on. The on-screen fine print in frame 016, "Sparse sample. Not a company forecast," and in frame 013, "Reference score from the actual Claude run. Jev has not been executed," are consistent with that. The narration's "you could literally map out over time the average tone for all of their products" at 28.54 s describes a multi-company, multi-product pipeline; what any frame actually shows is 94 reviews of one computer mouse.

The Nike hook is not connected to the Nike demo, because there is no Nike demo

At 0.00 s the video opens "Apparently, personal investors have been predicting the death of Nike by vibe coding," over a card reading "Claude predicted Nikes downfall." No frame I read shows any Claude analysis of Nike, any Nike Amazon listing, any Nike review tone series or any Nike tone chart. The Claude conversation visible in frame 011 is titled "Logitech M510 review analysis," and the twelve-company basket it proposes contains no Nike. The only Nike artifacts in the clip are a stock quote card and a Reuters headline.

Both Nike artifacts check out on their own terms. The Reuters story in frame 005 is real and its figures are confirmed by independent same-day reporting: Al Jazeera's June 28, 2024 piece, "Nike stock plunges on surprise forecast of drop in sales," states "It was the worst day ever for the stock, which slumped 20 percent on Friday, with the losses wiping out $28.41bn from the company's market valuation." That matches the "$28.41 billion" in the Reuters body text in frame 005 exactly. The $34.36 quote in frames 003 and 019 is plausible for the post date: Nike's 52-week low of $31.97 was set on October 2, 2026 and the stock closed at $35.30 on October 14, 2026, so a $34.36 print in the first week of October 2026 sits inside the observed range.

What does not hold is the implied sequence. The Reuters article is dated June 28, 2024. The video posted on October 8, 2026, two years and three months later. Nothing in the clip predicted anything about Nike; it displays a crash that had already happened, then pivots to a workflow demonstrated on a different company. Nike is also a structurally poor example for this specific method, because it stopped selling wholesale on Amazon in 2019 and resumed direct first-party Amazon sales only in May 2025, leaving very little Nike-controlled Amazon review history covering the June 2024 collapse.

The workflow is buildable, but the review-acquisition step is the part nobody shows

At 20.12 s the video says "your AI will scrape all of the reviews over the past two years, have a much smaller model, score each of the reviews based on tone, and then tag the reason for that tone score." The scoring half is unremarkable: frame 013 shows a single review scored "Tone -1" against the -2 to +2 scale defined in frame 016, which is a straightforward classification task.

The acquisition half is where the demo stays thin. No frame I read shows a terminal, a script, an API response, a rate-limit error or any artifact of pulling review data at scale. Frame 009 shows an ordinary Amazon search results page in a browser, which is a screenshot, not a pipeline. The sample that did get collected is 94 reviews of one product, which is what manual or lightly automated collection yields, not a two-year sweep across ten to twenty companies. The Claude output in frame 011 concedes the problem: it instructs the user to "check each company's Amazon share of sales and confirm that dated reviews can be accessed," meaning dated review access is a precondition to verify per company rather than an assumption the pipeline can make.

Key Takeaways

  • Verified: The study is real. Shunyao (Cynthia) Jin, "More than words: Can tone of consumer product reviews help predict firms' fundamentals?", Journal of Business Finance & Accounting 50(9-10), 1910-1942, online 2022-12-28, print October 2023, DOI 10.1111/jbfa.12680, author affiliated with the College of Business at University of Michigan-Dearborn.
  • Verified: The sentence circled on screen in frame 022 is verbatim from the published abstract: "negative review tone is more informative and useful than positive tone in predicting a firm's fundamentals."
  • Correction: At 45.74 s the video says a sudden negative tone "will almost always precede a drop in sales and stock returns." The abstract says tone "successfully predicts a firm's forthcoming quarterly sales" and that "this predictability is stronger for firms operating in a highly competitive environment." That is an average, conditional panel association, not a near-universal per-case rule. The abstract reports no hit rate.
  • Correction: The opening claim that "personal investors have been predicting the death of Nike by vibe coding," captioned "Claude predicted Nikes downfall," is not supported by anything in the clip. The Reuters article shown is dated June 28, 2024, two years and three months before the 2026-10-08 post. The Claude conversation shown is titled "Logitech M510 review analysis" and the twelve-company basket on screen contains no Nike.
  • Partial correction: The video's own demo contradicts its voiceover. Frames 021, 024 and 025 caption the one checked comparison "Record the miss." The narration never mentions it.
  • Clarification: The one negative tone swing in the demo falls in Logitech's FY27 Q2, whose results were scheduled for release on October 27, 2026, nineteen days after the video posted. The signal the thesis depends on had no outcome to check against at post time.
  • Clarification: The author is at University of Michigan-Dearborn, not the Ann Arbor campus most listeners will assume from "the University of Michigan."
  • Verified: Logitech's Q4 FY2026 release of May 5, 2026 states "Sales were $1.09 billion, up 7 percent in US dollars and 3 percent in constant currency compared to Q4 of the prior year," matching the circled figure in frame 025, confirmed against the SEC Form 8-K exhibit 99.1.
  • Verified: The Reuters figures visible in frame 005 are accurate. Nike shares fell 20 percent on June 28, 2024, wiping $28.41 billion from its market valuation.
  • Context: Nike exited wholesale on Amazon in 2019 and resumed direct first-party Amazon sales only in May 2025, so Nike's own Amazon review history is thin for the period covering the June 2024 collapse the video opens on.
  • Context: Jin's paper is not the only work in this area. Jiekun Huang, "The customer knows best: The investment value of consumer opinions," Journal of Financial Economics 128(1), 164-182, April 2018, uses Amazon customer ratings rather than review tone and reports a long-short spread portfolio returning roughly 55.7 to 73.0 basis points per month. Huang is at the University of Illinois at Urbana-Champaign, not Michigan.
  • Unverified: Secondary summaries put Jin's sample at 18,794,143 reviews across 6,276 firm-quarters from 292 firms. I could not confirm those numbers against the paper. Wiley returned HTTP 403 on both the landing page and the PDF endpoint, SSRN returned HTTP 403, and the University of Michigan Deep Blue record was behind a Cloudflare challenge.
  • Unverified: The video's claim that an AI can "scrape all of the reviews over the past two years" for ten to twenty companies. No frame shows a terminal, script, API call or error from any collection run. The only sample on screen is 94 reviews of a single Logitech M510 mouse.

Resources

Published October 8, 2026. Writeup generated from a favorited TikTok.