← The AI Hype Audit — all 314 verdicts

PARTLY Partly true: the '$4.99B of NYC building permits Claude found me' is a sum of self-reported job costs on free public DOB data anyone can pull with two API calls, the headline '$350M apartment building' is a fat-fingered fitness-center extension, and the poster sells $6,000+ 'lead machines' TikTok · Sep 2026

The claim'Claude Fable 5.1 found me $4,992,775,920 worth of new building permits in NYC.' Overnight, Claude agents 'scraped every building permit in New York over the past 60 days, equating to $4.9 billion,' listed what each job site needs, matched the top contractors for each need with a full outreach kit. Example: 'these guys are about to spend $350 million on one apartment building.' Add 'simple skip tracing plus maybe six to 10 Claude Code agents that live in a cloud server like AWS or Hetzner, then maybe you could bring deals in week in and week out fully autonomously.'

The data is real, public and free. NYC publishes every DOB NOW job filing and every approved permit on data.cityofnewyork.us, with an estimated-cost field, the owner's business name, the permittee (usually the GC), and a plain-English job description. It has a Socrata API, so 'scraped' is generous: two GET requests do it. We ran the numbers on Sep 18, 2026. Filings dated Jul 18 to Sep 16 total 26,752 jobs and $8.51 billion in initial cost; approved permits issued in the same window total $10.07 billion. The poster's $4,992,775,920 is a filtered slice of that, so the magnitude is plausible and, if anything, undersold. But 'found me' is doing all the work. The figure is the sum of numbers applicants typed into their own filings. It is not money, it is not a market, and it was never hidden. The video's marquee example proves the point: the only $350,000,000 permit in that window is a Queens job at a three-story nonprofit building with zero dwelling units, described as 'foundation design to facilitate construction of fitness center as extension.' Sibling filings on the same job are $800,000, $350,000 and $65,000. It is a data-entry error, and the pipeline led with it. Claude Fable 5.1 is real (shipped Sep 1, 2026). The autonomous 'deals every week' tail is hedged with 'maybe' and never shown. The bio link resolves to an agency selling done-for-you lead machines from $6,000 and a $7,500 done-with-you program that lists a 'Contractor Lead Machine,' skip tracing and autonomous outreach. This is a demo reel for that.

What holds up

  • Datasets confirmed live on NYC Open Data via the Socrata API (no scraping needed): 'DOB NOW: Build – Job Application Filings' (w9ak-ipjd; fields initial_cost, filing_date, owner_s_business_name, job_description, per-trade work-type flags), 'DOB NOW: Build – Approved Permits' (rbx6-tga4; estimated_job_costs, applicant_business_name = permittee, owner_business_name, issued_date) and legacy 'DOB Permit Issuance' (ipu4-2q9a). Newest rows are dated Sep 16, 2026.
  • Magnitude test run Sep 18, 2026 with SoQL: filings Jul 18–Sep 16, 2026 = 26,752 rows, sum(initial_cost) = $8,512,673,710; approved permits issued in the same window = 30,678 permits on 24,699 jobs, sum(estimated_job_costs) = $10,074,102,432. Calendar 2025 filings summed to $52.4B, in line with the New York Building Congress's ~$69–74B/yr citywide construction spending. The poster's $4.99B is a plausible subset, not an inflation.
  • Claude Fable 5.1 is a real Anthropic model: announced Sep 1, 2026 (anthropic.com, 9to5Mac, VentureBeat, MacRumors), API id claude-fable-5-1, $10 per million input / $50 per million output tokens, cache reads $0.25 per million.
  • The '$350 million apartment building' traced: the only $350,000,000 permit in the window is job Q00558742 at 72-02 31 Avenue, Queens (Foundation and Earth Work permits issued Aug 25, 2026). The filing shows a 3-story, nonprofit-owned building, 0 proposed dwelling units, description 'foundation design to facilitate construction of fitness center as extension in existing building'; related filings on the same job are $800,000, $350,000 and $65,000. Almost certainly $350,000 with three extra zeros.
  • What is sold: the profile is a TikTok business account; bio link → 'Agentic Solution' → 'I Build Lead Machines. You Own Them.' Done-for-you from $6,000, done-with-you $7,500 for three months, 'Apply for a call.' The offer page lists a 'Contractor Lead Machine,' owner/phone enrichment (skip tracing) and autonomous email outreach, i.e. exactly the pipeline in the video.

What doesn’t

  • 'Found me $4.99B' is the sum of self-reported estimated costs on public filings; nobody found anything, and the true 60-day total is larger, which shows how little the number means on its own.
  • The video's marquee example is a data-entry error: a $350M 'apartment building' that is a fitness-center extension on a 3-story nonprofit building. A pipeline that does not sanity-check outliers will lead your outreach with the wrong targets.
  • The payoff, '6–10 agents on Hetzner bringing deals weekly, fully autonomously,' is hedged ('maybe you could') and never shown; skip tracing plus autonomous calls or texts to cell numbers runs into TCPA, and autonomous email into CAN-SPAM.
  • Demo reel for a $6,000+ lead-machine agency, framed as a personal discovery; the free data and free API are never named.

The catch

Everything in the video that is real is free: the filings, the permits, the cost field, the GC and owner names, and an API that returns them in one request. What the AI adds is classification and a written brief per job, which is useful and cheap. What the AI does not add is money, and it did not catch that the biggest 'deal' on screen is a typo. Before paying anyone $6,000 to build this, run the two queries below yourself; you will have the same $5B spreadsheet in an afternoon and a better one once you throw out the outliers.

How to actually do it

  • Pull the data with two GET requests (get a free Socrata app token to lift rate limits): https://data.cityofnewyork.us/resource/w9ak-ipjd.json?$where=filing_date >= '2026-07-18T00:00:00'&$select=job_filing_number,initial_cost,house_no,street_name,borough,owner_s_business_name,job_type,job_description,filing_date&$limit=50000 and https://data.cityofnewyork.us/resource/rbx6-tga4.json?$where=issued_date >= '2026-07-18T00:00:00'&$select=job_filing_number,work_type,applicant_business_name,owner_business_name,estimated_job_costs,issued_date&$limit=50000. Note initial_cost is stored as text; cast with initial_cost::number in $select or $order. Join the two on the first nine characters of job_filing_number to attach the permittee (GC) and owner to each job.
  • Clean before you believe: dedupe permits to one row per job (30,678 permits vs 24,699 jobs in the last 60 days), drop any filing whose cost is 100x its sibling filings on the same job number, and eyeball the top 25 by hand. That step alone removes the '$350M apartment building.'
  • Have Claude classify each job from job_description plus the dataset's own work-type flags (plumbing, sprinkler, structural, mechanical, foundation, sidewalk shed, solar, demolition) into a short list of trades and materials, and write a one-paragraph brief per job for the top 200 by cleaned cost. Fable 5.1 at $10/$50 per million tokens makes this a few dollars, not a server farm.
  • Outreach is a human's job: email the business names on the filing from your own address with a real opt-out and postal address (CAN-SPAM), do not autodial or text cell numbers pulled from skip tracing (TCPA), and offer one specific introduction per job. Only after this hand-run version produces a paid intro is a $6,000 'machine' worth discussing.

The NYC DOB datasets, the Socrata API, the cost and contractor fields, and Claude Fable 5.1 are all real, and our own 60-day pull came back at $8.5B in filings and $10.1B in issued permits, so the poster's $4.99B is a fair slice. The 'found me' framing, the mis-keyed $350M fitness-center example and the unproven autonomous deal-flow tail are the hype, and the video is a demo for a $6,000+ lead-machine service. Anyone can reproduce the spreadsheet with two API calls and a cleaning pass.

Confidence
High
Posted by
a 'building anything and everything' construction-tech TikToker, ~2.2K followers, business account (Education & Training) whose bio link leads to his lead-machine agency; video posted Sep 16, 2026 and sent to the self-thread Thu Sep 17, 2026, 11:10 PM

See the original claim →

We test hype for free. We build the real thing for a living.

Thirty minutes, no pitch — and you'll leave with something useful either way.

Book a call with Todd or start with the free Business Checkup →

The Verdict Weekly

Three verdicts every Friday. Free forever, unsubscribe anytime, no spam — that would be ironic.

© Schreier Group · schreiergroup.com · See a wild AI claim? Drop it here and we'll test it.