<- all tokdocs
2026.10.27@angusthenontechnical

The video sells a citywide "speed of service" award for Chattanooga med spas; the Claude session it puts on screen says the same data supports no citywide ranking

Marketing
Watch on TikTok

View on TikTok ->

A 68.821995 s vertical clip, 1080x1920 at 30 fps, in an MP4 container that ffprobe reports as mov,mp4,m4a,3gp,3g2,mj2, HEVC video at 596,256 bps alongside stereo AAC at 48,130 bps and 44.1 kHz, 650,989 bps overall, 5,600,302 bytes on disk, with a sidecar audio.mp3 of 68.884898 s at 96,066 bps and 827,193 bytes. Posted at epoch 1793063801, which is 2026-10-27 01:16:41 UTC, by angusthenontechnical, nickname "Angus the Nontechnical", credited to "original sound" by the same account. The description is 45 characters and repeats the title verbatim: "Tis the season #ai #buildwithai #vibecoding ". At capture, 2026-10-27 05:02:07 UTC per the metadata fetch epoch and roughly three hours forty-five minutes after posting, the listing showed 1,822 views, 64 likes, 1 comment, 4 reposts and 57 saves. I read 24 of the 34 extracted frames: 001, 002, 003, 004, 006, 007, 009, 010, 011, 012, 013, 014, 015, 017, 019, 021, 022, 023, 024, 026, 027, 029, 031 and 033, which at the 2 s sampling interval map to 0, 2, 4, 6, 10, 12, 16, 18, 20, 22, 24, 26, 28, 32, 36, 40, 42, 44, 46, 50, 52, 56, 60 and 64 seconds. The transcript is 274 words across 15 segments. The clip opens on a composited Times Square shot with a red Spotify billboard reading "2025 Wrapped" above a SILVERCAST sign, with a white caption card reading "Spotify wrapped is a Trojan horse" (frames 001, 002). Frames 003 and 004 cut to a four-panel grid labelled "DUOLINGO" (a phone showing "2022 Year in Review" and "You are a SUPERSTAR LEARNER / You're in the top 5% of learners worldwide"), "OURA" ("Your scores over the year", Readiness, Sleep, Activity), "STRAVA" (a phone showing "2020 TOTALS", "#YEARINSPORT", "DAYS ACTIVE 242", "1,225", "ELEVATION 41,500", "TOP SPORT") and "APPLE MUSIC" (the Replay panel). Frame 006 shows a Spotify 2025 recap card with "Top Artists 1 shy martin 2 The Chainsmok... 3 Gryffin 4 Hoang 5 Fly By Midnight", "Top Songs 1 Drove You Away 2 Feelings 3 Good Together 4 Softcore 5 I WA...", and "Minutes Listened 37,390". Frame 007 shows a Spotify slide reading "You listened to 459 artists this year." Frames 009, 010 and 031 show the same award card inside a social feed and a Gmail thread titled "Made this for Larimar" with the grey label "MOCKUP" and the body "Hey Mia! Made this little award card concept for Larimar. Thought it would look pretty good on your story. Angus". Frame 011 is a Google Maps result list for "med spas Chattanooga TN" showing "Center MedSpa 4.9 (436) Medical spa 320 E Main St Ste 110", "Alchemy MedSpa + Wellness Center 5.0 (144) Medical spa 200 Manufacturers Rd Unit 105" and "Larimar Medspa 5.0 (355)". Frame 012 is the Chattanooga map with pins for Larimar Medspa, "Milan Laser Hair Removal 5.0 (226)", "Four Bridges Dermatology and Cosmetics - Chattan. 4.9 (1246)", EURO Med Spa, Monarch Beauty Aesthetics, 3 Leaf Med Spa and Center MedSpa. Frame 013 is the Apify store listing for "Google Maps Reviews Scraper", slug compass/google-maps-reviews-scraper, "from $0.30 / 1,000 scraped reviews", "4.9 (199)", developer "Compass", "Maintained by Apify", "764 Bookmarked", "8.6K Monthly active users", "1.3 days Issues response", "6 hours ago Last modified", category "Travel". Frames 014 and 023 show two copyable prompt cards, "COLLECT THE PUBLIC DATA" and "TAG, CALCULATE, MAKE AWARDS". Frames 015, 017, 021 and 022 show a Claude conversation titled "Larimar Medspa review tagging" whose composer footer reads "Opus 5.5 Medium Auto", containing a five-row "Review tags" table ending "Count: 1 of 5 mention speed." and "5 excerpts. No citywide ranking." Frames 019, 024, 027 and 029 show three finished award cards, all stamped "EXAMPLE DATA • CONCEPT AWARD" or "DESIGN CONCEPT". Frame 026 is an Instagram story composer captioned "Angus made this for our page" with the label "MOCKUP" and the handle "@angus.sewell". Frame 033 is a plain talking-head shot in a room with a laptop on a shelf. Spoken, the argument is that year-in-review recaps work because they hand people a flattering, shareable artifact, and that a marketing agency should build one for a prospect's business instead of its own, by scraping one niche in one city, having an AI tag and count the reviews, and shipping an award graphic as cold outreach. The transcript's last segment ends at 68.62 s against an audio duration of 68.884898 s and a video duration of 68.821995 s, so nothing in the transcript is timestamped past the end of audio and no segment is fabricated tail.

The spoken award and the on-screen Claude output disagree with each other

At 32.82 s the video says "it can actually calculate. Most reviews that mention the speed of service of every med spa in" and at 37.16 s continues "Chattanooga." The card that illustrates this, fully visible in frame 019 at 36 s, reads "LOCAL LEGENDS / 2026", "34 reviews mentioned speed of service", "CHATTANOOGA WINNER", "SPEED OF SERVI[CE]", "LARIMAR MEDSPA", "CHATTANOOGA, TN".

The number 34 appears nowhere in the work the video shows. The Claude conversation on screen in frames 015, 021 and 022, titled "Larimar Medspa review tagging", ends with "Count: 1 of 5 mention speed." and "5 excerpts. No citywide ranking." The user turn above it reads "Repeat the corrected five-row Review tags table and the count, unchanged. End with '5 excerpts. No citywide ranking.' No other text." The five rows are R1 "right at your appointment time" tagged Yes, and R2 through R5, "incredibly kind and super attentive", "takes time to look at", "made me feel completely at ease" and "left my skin glowing instantly", all tagged No.

Frame 017 at 32 s is more explicit. The visible Claude response says, in numbered point 3, "Count tags in code, not by hand, and set a minimum sample per business in advance; with 5 excerpts and 1 yes, Larimar would fall below any reasonable threshold and stay unranked." Point 4 says "State the collected scope with every result: which businesses, how many excerpts each, the source, and the date. Right now that is 5 excerpts from one Chattanooga medspa, which supports no citywide claim."

So the only tagging run the video displays produced one speed mention out of five excerpts from one business, and the model told the creator that this supports no citywide ranking. The voiceover sells a winner for "every med spa in Chattanooga" over the top of that. The creator did label the artifacts: every card in frames 019, 024, 027 and 029 carries "EXAMPLE DATA • CONCEPT AWARD" or "DESIGN CONCEPT", the email in frame 031 is tagged "MOCKUP", and so is the Instagram composer in frame 026. The labels are on screen and the overclaim is in the audio.

The one number on a card that traces to a visible source is 355, and it matches

Frame 027 at 52 s shows a third card reading "LOCAL LEGENDS 2026", "THE TALK OF THE TOWN.", "355", "public reviews", "LARIMAR MEDSPA", "CHATTANOOGA, TN", footer "DESIGN CONCEPT • COUNT S...". That 355 matches the Google Maps result panel in frame 011 at 20 s, which lists "Larimar Medspa 5.0 (355)". A count of total public reviews is the one claim on any of the three cards that the clip sources on camera, and it is internally consistent.

The other two card numbers are not sourced in the clip. Frame 029 at 56 s shows "THEY CAME BACK. 45 reviews mentioned a return visit." The 45 appears in no visible table, prompt output or scraper result. Treat both 34 and 45 as unverified decoration on a concept mock.

Larimar Medspa itself is real and in Chattanooga. Third-party directory aggregators report different Google review totals at unstated capture dates, for example 328 in one listing, so I cannot retroactively confirm that the live Google count read exactly 355 on the day of filming. Review counts move daily, and a figure captured on camera from Google's own panel is the stronger evidence here.

Spotify did move first, and the grid of imitators is not a launch-order chart

At 5.22 s the video says "every business has their own Spotify wrapped now?" and at 14.62 s "But it turns out the best way to steal this isn't by making one for your customers." The premise that Spotify originated the format holds up, and the specific products in the frame 003 and 004 grid came later.

Wikipedia's Spotify Wrapped entry states the campaign was "first launched in December 2016" and "has been promoted by Spotify every December since that year", preceded by a less developed campaign called "Year in Music" in 2015. Duolingo's own engineering and marketing post, published 8 December 2022 by Jasmine Zhang and Clark Munson, dates its recap to 2019: "In December 2019, we sent the newsletter to our users highlighting their active days, time spent learning, number of words learned", and "In 2020, we took the experience further by moving it into the product." TechCrunch dated 13 November 2019 reports Apple Music Replay as a launch that week: "Apple Music is taking on Spotify with the launch of a new feature, Apple Music Replay, that will allow subscribers to take a look back at their favorite music from 2019", and frames the comparison directly, "The feature is similar in some ways to Spotify's popular year-end review, known as Wrapped."

Two details in the grid are worth separating from the argument. The Duolingo panel shows "2022 Year in Review", three years after the feature's 2019 debut, so the grid is a sample of current artifacts rather than a chronology. And Strava uses the Year in Sport name for two different things: the personal recap the frame shows, carrying "2020 TOTALS" and "#YEARINSPORT", and an aggregate community trend report. Strava's own press release of 7 December 2022 describes the latter as "The highly anticipated annual report concludes the year by revealing global trends in sports and active transportation", which is a different product from the personal card in the frame.

The video's own timing is also ahead of the season it invokes. Wrapped 2025 launched on 3 December 2025, per Spotify's newsroom post of that date, which also states the data window ran "from January until mid November, just a few weeks prior to this year's launch." This clip posted 27 October 2026, roughly five weeks before the window Spotify has used. As of mid-October 2026 Spotify had not announced a 2026 Wrapped date. I attempted to fetch a CelebrityAccess report dated 7 October 2026 about Wrapped artist message uploads and received an HTTP 307 redirect, and the Spotify for Artists Wrapped page returned 404, so I am citing neither.

"Scrape all the Google reviews" collides with Google's own prohibited conduct terms

At 23.68 s the video says "one city. You could scrape all the Google reviews for med spas, which this prompt will do all of" and at 28.2 s "that for you." No terms of service appear anywhere in the clip, and the Apify listing in frame 013 carries no statement about the legality of scraping Google Maps reviews either, which I confirmed by reading the live actor page.

Google's Google Maps and Google Earth Additional Terms of Service, under Prohibited Conduct, bars users from "mass download or create bulk feeds of the content (or let anyone else do so)" and from using "Google Maps to create or augment any other mapping-related dataset (including a mapping or navigation dataset, business listings database, mailing list, or telemarketing list)." The workflow the video describes, building businesses.csv and reviews.csv for every med spa in a city from Google Maps review text, is a business listings database assembled from Google Maps content. The older Google Maps APIs terms are stricter still on storage, with section 10.5(d) limiting caching of Content to temporary copies of "no more than 30 calendar days" and section 10.5(e) stating "You will not use the Service in a manner that gives you or a third party access to mass downloads or bulk feeds of any Content."

The Apify actor details the video shows are accurate against the live listing. The page reports the slug compass/google-maps-reviews-scraper, developer Compass, pricing "from $0.30 / 1,000 scraped reviews", 764 bookmarks and 8,602 monthly active users, which match the frame's "764 Bookmarked" and "8.6K Monthly active users" exactly. The live rating reads 4.86, which the store rounds to the 4.9 shown in the frame. The description matches word for word: "Extract all reviews of Google Maps places using place URLs. Get review text, published date, response from owner, review URL, and reviewer's details."

The prompt cards on screen are stricter than the pitch spoken over them

The video twice says "this prompt will do all of that for you", at 23.68 s and again at 41.16 s. The two prompt cards it flashes contain guardrails the voiceover never mentions.

Frame 014 at 26 s, headed "COLLECT THE PUBLIC DATA", reads in part: "1. Build a public-data list of med spas in Chattanooga, Tennessee. Deduplicate each business by its source place ID. 2. Using connected data tools, collect accessible review text, date, business ID and source URL. Record the fetch date and what you could not access. 3. Save businesses.csv and reviews.csv with stable business_id and review_id fields. Preserve the original text. Never invent missing rows."

Frame 023 at 44 s, headed "TAG, CALCULATE, MAKE AWARDS", reads: "1. Use businesses.csv and reviews.csv. Tag every review for explicit service-speed, repeat-visit and staff-name mentions; keep the evidence quote and row ID. 2. Mark uncertain rows unknown. Audit a sample manually; fix inconsistent tags before calculating. 3. In code, count matching reviews per business over the same date window. Show totals and shares; require at least 30 usable reviews. 4. Name winners only within the collected sample. Report ties, exclusions and source links. No causal claims."

Step 3 on that card requires at least 30 usable reviews before naming a winner. The demo in frames 015 and 021 has five excerpts. Step 4 restricts winners to "the collected sample", which rules out the phrase "CHATTANOOGA WINNER" printed on the card in frame 019 unless the sample is the whole city. The prompts are the honest part of this video and the audio is the part that outruns them.

The hallucination answer is also narrower than the voiceover makes it sound. At 41.16 s the video asks "But wouldn't AI just hallucinate a bunch of these findings? How do we make it good?" and answers "If you have your AI go and tag each row of your data". Row tagging alone does not stop a model inventing a count. The method visible in frame 017 adds four separate controls: stable row IDs "tied to its source URL and retrieval date, so anyone can trace a tag back to the exact text"; tagging "only on explicit wording ('right at your appointment time' = yes; 'glowing instantly' = no)" with the quoted evidence stored beside the tag; counting "in code, not by hand" against a pre-set minimum sample; and stating the collected scope with every result. The frames carry the real recipe, the audio compresses it to one of its four steps.

The composer model reads "Opus 5.5", which is current, and no scraper run appears anywhere

Frame 022 at 42 s shows the Claude composer footer reading "Opus 5.5 Medium Auto". That checks out against Anthropic's live model documentation, which lists Claude Opus 5.5 in the current lineup with Claude API ID claude-opus-5-5, a 1M token context window, 128K max output, and a default effort level of medium. The footer in the frame is a model name plus the default effort setting for that model, not a stray UI string.

What never appears is the pipeline running. Across all 24 frames I read there is no terminal, no Apify run log, no dataset view, no downloaded CSV, no spreadsheet of tagged rows and no count executed in code. Frame 013 is the Apify store listing page with a "Try for free" button and a "Copy setup prompt" button, which is a product page rather than a run. The only model output shown is a five-row table and a four-point method summary. The clip demonstrates the prompt and the design output, and skips the collection and counting steps that the award numbers would have to come from.

One compliance note worth stating plainly, without drawing a legal conclusion. The FTC's Trade Regulation Rule on the Use of Consumer Reviews and Testimonials, 16 CFR Part 465, was announced on 14 August 2024, published in the Federal Register on 22 August 2024 and took effect on 21 October 2024. Section 465.6 prohibits an advertiser from representing "that a website, organization, or entity it controls, owns, or operates provides independent reviews or opinions" about its own category. An agency that mints a "LOCAL LEGENDS 2026" award brand and issues "CHATTANOOGA WINNER" cards to prospects is operating exactly that kind of entity, and the sample size behind the award is the thing that decides whether the claim on the card is accurate. The cards in this video are labelled as concepts and the counts are labelled as example data. A real send would carry neither label.

The closing line is also slightly generous to itself. At 63.94 s the video says "follow your boy while you're at it. Cause I'm going to keep posting cool stuff like this for entirely free." The creator's own site, angussewell.com, offers a free semi-weekly newsletter alongside a paid "Member Vault" of pre-built systems. The posts are free. The catalog is not.

Key Takeaways

  • Correction: The video says at 32.82 s that AI can calculate "Most reviews that mention the speed of service of every med spa in Chattanooga", and the card in frame 019 prints "34 reviews mentioned speed of service / CHATTANOOGA WINNER". The Claude session the video itself shows, in frames 015, 017, 021 and 022, ends "Count: 1 of 5 mention speed. 5 excerpts. No citywide ranking." and states that "with 5 excerpts and 1 yes, Larimar would fall below any reasonable threshold and stay unranked."
  • Correction: The workflow proposed at 23.68 s, scraping all Google reviews for a niche to build businesses.csv and reviews.csv, runs against Google's Google Maps and Google Earth Additional Terms of Service, which prohibit users from "mass download or create bulk feeds of the content" and from using Google Maps to "augment any other mapping-related dataset (including a ... business listings database ...)." No terms of service are mentioned in the clip.
  • Partial correction: The prompt card in frame 023 requires "at least 30 usable reviews" and says "Name winners only within the collected sample." The demo run has five excerpts from one business, and the card in frame 019 says "CHATTANOOGA WINNER".
  • Partial correction: At 41.16 s the video answers the hallucination question with row tagging alone. The method visible in frame 017 pairs tagging with stable row IDs tied to source URL and retrieval date, explicit evidence quotes, counting "in code, not by hand" against a minimum sample, and a scope statement with every result.
  • Verified: Spotify Wrapped was "first launched in December 2016" and preceded by "Year in Music" in 2015, Apple Music Replay launched on 13 November 2019, and Duolingo's Year in Review began as an email in December 2019 and moved in-app in 2020. The video's claim at 14.62 s that the format was copied from Spotify holds against the dated sources.
  • Verified: The Apify listing details in frame 013 match the live page, including the slug compass/google-maps-reviews-scraper, developer Compass, "from $0.30 / 1,000 scraped reviews", 764 bookmarks, 8,602 monthly active users and the description text. The frame's "4.9" is the rounded form of the live 4.86 rating.
  • Verified: The "Opus 5.5 Medium Auto" footer in frame 022 matches Anthropic's live documentation, which lists Claude Opus 5.5 with API ID claude-opus-5-5 and default effort medium.
  • Verified: The 355 on the card in frame 027 matches the "Larimar Medspa 5.0 (355)" shown in the Google Maps panel in frame 011. The transcript's last segment ends at 68.62 s against 68.884898 s of audio, so no segment is hallucinated past the end.
  • Unverified: The 34 speed mentions on the card in frame 019 and the 45 return-visit mentions on the card in frame 029 appear in no table, prompt output or scraper result shown in the clip. Third-party directories report Larimar's Google review total as 328 and 334 at unstated dates, so the live count on the filming day cannot be confirmed retroactively.
  • Context: The FTC's rule on consumer reviews and testimonials, 16 CFR Part 465, took effect on 21 October 2024 and bars an advertiser from representing that an entity it controls "provides independent reviews or opinions" in its own category. Every award card in this clip is stamped "EXAMPLE DATA • CONCEPT AWARD" or "DESIGN CONCEPT" and both outreach mockups are labelled "MOCKUP". A real send would not carry those labels.
  • Context: No terminal, run log, dataset view, CSV or code-executed count appears in any of the 24 frames I read. The Apify screen is a store listing page, not a run.
  • Clarification: The closing line at 63.94 s promises content "for entirely free." The creator's site offers a free semi-weekly newsletter plus a paid "Member Vault" of pre-built systems.

Resources

Sources I could not reach and therefore do not cite: a CelebrityAccess report dated 7 October 2026 on Spotify Wrapped artist message uploads, which returned HTTP 307, and the Spotify for Artists Wrapped page, which returned HTTP 404.

Published October 27, 2026. Writeup generated from a favorited TikTok.