AI Pretender Screen
Test whether an AI narrative has real infrastructure, capability, and investment behind it
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 5
- Confidence
- 93%
The AI Pretender Screen tests whether a company's AI narrative is supported by technological and economic substance. First, compare filings and statements over time for abrupt vocabulary substitutions resembling a global find-and-replace. Next, determine what actually powers the claimed capability: a major third-party model, proprietary infrastructure, or an unspecified internal system. Because frontier model development requires immense capital, data, hardware, and expertise, claims of casually built proprietary AI deserve strong skepticism. Finally, compare the stated capability with disclosed investment, personnel, partnerships, and product evidence. The screen does not reject companies that use external models; it distinguishes honest enablement from claims that imply nonexistent proprietary technology.
Origin
Extracted from Masters in Business during Carson Block's comparison of AI branding with earlier corporate cloud pivots.
Core principles
- 01Sudden vocabulary changes can reveal narrative chasing
- 02Real model development requires substantial scale and capital
- 03AI-enabled products should disclose their technological foundation
- 04Branding is not evidence of proprietary capability
How to run it
- 1
Diff the Narrative
Compare recent filings, presentations, and earnings calls with prior versions to identify abrupt changes in fashionable terminology.
Pro tip Search for repeated passages where only a technology label has changed.
Watch out A terminology change alone is a warning signal, not proof of deception.
- 2
Ask What Runs It
Identify the foundation model, cloud infrastructure, chips, data, and vendors behind the claimed AI capability.
Pro tip Accept credible third-party integration when the company describes it accurately.
Watch out Do not equate using an AI API with developing a proprietary model.
- 3
Check the Scale
Compare the ambition of the claim with spending, compute access, technical staff, and development time.
Pro tip Use leading model developers as rough reference points for the resources required.
Watch out Extraordinary proprietary-model claims need commensurate evidence.
- 4
Look for Product Evidence
Seek demonstrations, customer adoption, measurable improvements, and technically specific disclosures.
Pro tip Favor verifiable outcomes over broad claims that everything is now AI enabled.
Watch out A polished demonstration may still conceal manual work or third-party technology.
- 5
Classify the Claim
Label the company as a credible adopter, credible developer, unclear claimant, or likely narrative pretender.
Pro tip State what additional evidence would change the classification.
Watch out Do not turn skepticism into certainty when the evidence remains incomplete.
In the wild
An analyst compares two annual reports and finds nearly identical product descriptions, except that every reference to cloud technology has become AI. The company names no model provider, discloses no material technical hiring, and shows no new product metrics.
→ The company is classified as a likely narrative pretender pending evidence of genuine capability.
A software company explicitly states that it uses a third-party foundation model, explains its proprietary workflow and customer data controls, and reports measurable reductions in task time. It does not claim to have built the underlying model.
→ The company passes as a credible AI adopter despite lacking proprietary foundation technology.
Common mistakes
Rejecting All Third-Party AI
A business can create real value by integrating an external model. The concern is misleading claims, not the absence of a proprietary foundation model.
Accepting an Internal-Build Claim
Claims of building frontier-quality models internally should be checked against the enormous capital and infrastructure required.
Treating a Red Flag as Proof
Sudden AI language warrants investigation, but analysts still need evidence before concluding that management is deceptive.
Is it for you?
Best for
Analysts reviewing public companies that have recently introduced AI-heavy investor messaging.
Not ideal for
Assessing early private research projects where expenditure and architecture are legitimately confidential.
From the transcript
“I saw some companies that had just done a global find and replaced, like, you know, find this, replace it with cloud.”
“if you see that with AI in a company's filings and in their statements, then yeah, I mean they're they're probably a pretender.”
“these things have cost way too many billions of dollars to develop for you to be able to vibe code.”
From the episode
BONUS: Muddy Waters Capital Founder Carson Block
Carson Block