AI Analysts Can’t Make Sense of Cannabis Stocks Yet. The Reasons Are Structural.

Alan Brochstein has covered cannabis stocks longer than almost anyone: 13-plus years for New Cannabis Ventures, which has been publishing its weekly market newsletter since October 2015. This week he did something worth paying attention to — he tested AI tools on cannabis stock analysis and published what he found.

The verdict, as the headline on his newsletter put it: not quite there yet.

His testing covered ChatGPT and Claude across a range of cannabis stock questions. The results were mixed in an instructive way. The AI tools gave responses that were, in Brochstein’s word, “smart.” The problem is that smart-sounding answers about an industry as structurally abnormal as cannabis can be actively misleading. He used Glass House Farms as an example — not as an attack on the company or its CEO Kyle Kazan, but as an illustration of how AI can produce coherent-sounding analysis in a sector where coherence is harder to achieve than it looks from the outside.

The structural barriers aren’t subtle. Cannabis companies operate under 280E — the federal tax provision that denies normal business deductions to businesses trafficking Schedule I or II controlled substances. This creates GAAP financials that look nothing like any other consumer sector: companies with strong operations can show accounting losses; companies with deteriorating fundamentals can look healthier than they are on metrics that 280E distorts. Standard valuation models weren’t built for this. AI trained on conventional financial data will reproduce conventional financial reasoning — which is exactly wrong for a cannabis P&L.

Add the cross-state legal patchwork (different license structures, different tax frameworks, different inventory rules), the absence of federal banking access, the OTC-trading status of most U.S. MSOs, and the fact that analyst coverage of the sector is thin and inconsistent, and you have an environment where the training data that makes AI useful elsewhere works against it here.

Brochstein’s broader point is also worth noting: AI may help retail investors avoid the worst mistakes — identifying obvious red flags, summarizing disclosed financials — but it is not yet a tool for identifying which cannabis stocks are undervalued. The sector’s value drivers are too idiosyncratic, too contingent on state-level regulatory outcomes, and too dependent on the 280E question (will it end? when?) to be reliably modeled from historical patterns.

That’s probably good news for analysts who actually know the sector. It’s worth a raised eyebrow from any operator currently pitching to investors who are using AI-generated research as their primary due diligence.

Source: New Cannabis Ventures, March 26, 2026.