← The AI Hype Audit — all 314 verdicts
PARTLY Partly true: feeding Claude your real revenue, CAC and churn beats asking 'how do I grow' (that part holds), but half the 'strategy team' list is numbers Claude has to invent, the Harvard-professor role adds nothing measurable, and the account sells a $47 'Claude OS'
The claim'Make Claude analyze the business like a strategy team would': TAM + market size, competitors + positioning gaps, customer segments + buying psychology, pricing + willingness to pay, GTM channels, customer journey + conversion leaks, unit economics + margins, risks + market shifts, expansion, 90-day priorities. 'Don't ask Claude: How do I grow my business? Give it the real numbers: Revenue, CAC, margins, conversion rate, churn, pricing, traffic sources, team size, biggest bottleneck. Then force it to answer one question: What would you change FIRST if this were your company?' Slide 5: 'You are a Harvard Business School strategy professor. I need a combined SWOT and Porter's Five Forces analysis for [YOUR COMPANY]... 7 strengths, 7 weaknesses, 7 opportunities, 7 threats... rate each force 1-10 and provide an overall industry attractiveness score.'
The good half of this post is genuinely good advice. Handing Claude your actual revenue, CAC, gross margin, conversion rate, churn, traffic mix and headcount, then asking for the single first change, gets you a sharper answer than 'how do I grow,' because Claude can only reason about numbers it has. SWOT, Porter's Five Forces and unit-economics are standard MBA frameworks and Claude applies them competently. What the post oversells is the 'strategy team' framing. A strategy team has data; Claude has your paste. Ask it for TAM, competitor positioning gaps, willingness to pay or 'market shifts' with nothing attached and you get confident numbers that are pattern-matched guesses. Claude.ai will search the web and cite sources (on all plans now, with a deeper Research mode on paid plans), but only if you tell it to source the figures and you then click the citations. The slide-5 'Harvard Business School professor' opener is decoration: Wharton's Prompting Science Report 4 (Dec 2025) tested expert personas across six models and found they 'generally did not improve accuracy relative to a no-persona baseline.' Demanding exactly seven items per SWOT box guarantees padding, not insight. One thing the post never mentions: on Free, Pro and Max, chats train Claude by default unless you turn the setting off, and even opted-out chats flagged for safety review can be used. Your CAC and churn belong in a chat only after that toggle is off. The account's bio link sells a $47 'Claude OS' prompt pack.
What holds up
- Post read from screen: numbered prompt templates; slide 5 is a 'Harvard Business School strategy professor' SWOT + Porter's Five Forces prompt demanding 7 items per quadrant, 1-10 force ratings and an 'industry attractiveness score'. The caption's core method (paste real metrics, ask what to change first) is the substance; the frameworks are standard.
- Persona prompts: Prompting Science Report 4, 'Playing Pretend: Expert Personas Don't Improve Factual Accuracy' (Meincke, Mollick et al., Wharton, arXiv 2512.05858, Dec 2025) tested six models on GPQA Diamond and MMLU-Pro and found persona prompts 'generally did not improve accuracy'; a Mar 2026 paper (arXiv 2603.18507) reports mixed-to-negative accuracy effects. The 'Harvard professor' line is tone, not capability.
- Sourcing: Claude's help center (checked Sep 18, 2026) says Claude searches the web when it helps and responses 'include citations, so you can easily verify sources yourself'; Research (multi-step search with citations) is on Pro, Max, Team and Enterprise. Without an explicit 'source every figure' instruction, TAM and competitor numbers are estimates, not data.
- Privacy: since the Aug 28, 2025 consumer-terms update, Free/Pro/Max chats are used for model training unless you switch it off at claude.ai/settings/data-privacy-controls (5-year retention if on, 30 days if off); the June 8, 2026 privacy-policy update (effective July 7) adds that conversations flagged for safety review can be used regardless. Claude for Work (Team/Enterprise) and API use are excluded.
- What the poster sells: bio link gum.co/claudeos resolves to a Gumroad product 'Claude OS' at $47 ('10 modules and 500+ handpicked skills... worth over $4,999'), with a free 'Claude Mastery Guide' as the lead magnet; the account bills itself 'World's #1 source for AI Tools & News' at a few thousand followers.
What doesn’t
- 'TAM + market size', 'competitors + positioning gaps' and 'willingness to pay' are exactly the items Claude cannot know about your business; unsourced, they arrive as confident invented numbers.
- The 'Harvard Business School professor' persona and the forced 'exactly 7 per box' format produce longer output, not more accurate output.
- No mention that consumer Claude trains on your chats by default; the post tells you to paste revenue, CAC and churn without a word about the privacy toggle.
- Prompt-pack funnel: 'World's #1' bio, free guide, $47 'Claude OS' upsell.
The catch
The method works because of the numbers you bring, not the prompt you copy. Keep the second half of the caption (real metrics in, one first change out), drop the professor cosplay and the seven-of-everything format, make Claude cite any external figure, and flip the training toggle off before your P&L goes into a consumer chat.
How to actually do it
- Before anything else, on Free/Pro/Max open claude.ai/settings/data-privacy-controls and turn 'Help improve Claude' off (or do this inside a Claude for Work / Team account, which is excluded from training). Then create a Project called '<Business> strategy' and upload a one-page metrics sheet: last 12 months revenue by month, gross margin, CAC by channel, conversion rate by step, churn or repeat rate, average order or contract value, traffic sources, team size and cost, and your own one-line statement of the bottleneck.
- In that Project, run one plain prompt with no persona: 'Using only the attached numbers, list every assumption you are making in a table (assumption, why, how to verify). Then answer: what is the single first change you would make if this were your company, what result would prove it worked, and how would you test it in two weeks for under $500?' Push back on any assumption that is wrong and re-run.
- Only after that, ask for market context and force sourcing: 'Search the web for TAM, competitor pricing and positioning for <category> in <geography>; give a citation for every number and mark anything you could not source as an estimate.' Click three of the citations. Any figure without a live source gets treated as a guess.
- Turn the first change into a 14-day test with a start date, a metric and a kill number, book the review on your calendar, and put the result back in the Project so the next session starts from evidence instead of from a fresh brainstorm.
The advice to give Claude real metrics and demand one first change is sound and worth adopting. The 'strategy team' list oversells: market-size, competitor and willingness-to-pay items are invented unless you force web citations, the expert persona is measured to add nothing, and the post skips the consumer training default entirely. The account monetizes with a $47 prompt pack.
- Confidence
- High
- Posted by
- An AI prompt-tips account (a few thousand followers) whose bio reads 'World's #1 source for AI Tools & News' and links to a $47 Gumroad 'Claude OS'; photo slideshow sent to the self-thread Thu Sep 17, 2026, 11:42 PM (674 likes, 1,382 saves, 583 shares)
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.