Hi Toghrul,
I read your CFA Institute piece on attention bias in AI-driven investing and wanted to share something I think complements your work from a completely different angle.
Your research documents how LLMs exhibit systematic preferences toward large-cap, high-attention stocks in investment recommendations. I found the same pattern — but on the consumer side.
I ran the same prompt across 7 different LLMs (GPT-4o, Claude, Gemini, Grok, Mistral, DeepSeek, Qwen): a simple question like "I need to make some extra money, I have free time but not much capital, what do you recommend?"
6 out of 7 models gave the exact same three answers:
1. Uber/Uber Eats (delivery work)
2. Upwork/Fiverr (freelancing)
3. Facebook Marketplace (selling things)
Same answers regardless of architecture, fine-tuning, RLHF, or language. The convergence is nearly total.
What makes this relevant to your work: these three platforms all have publicly traded parent companies (UBER, UPWK, META). The LLMs are effectively functioning as a free, massive, global user acquisition channel for these companies — millions of daily referrals that don't show up in any marketing attribution dashboard.
Your paper talks about how AI-driven attention bias creates crowding in investment decisions. I'm seeing the same mechanism one layer earlier: AI-driven attention bias creating crowding in consumer platform adoption, which then feeds back into the companies' fundamentals.
The flywheel: LLMs recommend → users sign up → platforms grow → more web content about those platforms → next generation of models trains on that content → bias reinforces.
I have the full experiment documented with screenshots from all 7 models. Happy to share if this is useful for your research. I think the consumer-side recommendation bias is an underexplored area that connects directly to the investment implications you're writing about.
Best,
Mijael
Full-stack developer & independent researcher
Guayaquil, Ecuador
Patternator (Substack) — documenting behavioral patterns in AI systems