AI Writes the Disclosure, Reads the Disclosure, then Votes on the Disclosure

Speaking of quarterly reports (see story above), a new study finds that companies are increasingly facing “AI-induced disclosure pressure – the incentive to write in a way that performs well under algorithmic scrutiny” in quarterly filings.

Why it matters: If public filings are being written for AI, by AI, then judged by AI – how much of them is necessary to begin with?

The use of AI to read MD&As is changing the way we write MD&As, according to research by Hebrew University Business School Professor Keren Bar-Hava. The professor identifies three levels of AI-induced pressure:

• Exposure Pressure. AI flags vague or evasive sounding language, so companies are compelled to sound confident.
• Competitive Pressure. AI benchmarks tone against peers, so companies are compelled to sound strong.
• Reputational Pressure. AI analysis travels fast, from analysts to media and back again.

The result, Professor Bar-Hava says, is companies using AI to write specifically for AI, by emphasizing positive tone, certainty (avoiding terms like “might” or “could”), and strength – regardless of whether this is warranted by performance. The goal is to fool the AI, since it impacts so many things (see a list of how AI analyzes your earning reports here).

Meanwhile, AI is already being used by proxy advisors in information extraction, classification and scoring, according to Professor Masaki Iwasaki of Seoul National University School of Law. Professor Iwasaki believes AI will eventually be used to formulate voting recommendations – and indeed, it seems likely that either proxy advisors or investors will wind up using AI in such a way.

Bottom line: AI is becoming judge, jury and executioner when it comes to public filings – and the CD&A will be no exception. For companies, the burden of producing long filings looks increasingly inefficient if only machines read them. One futuristic option could be to evolve disclosures into layered formats, with essentials for humans and structured data (tagged, machine-readable) for AI. Who can tell what the future brings?