Share of IPO Prospectuses Mentioning AI (quarterly)

## What this measures

A Palanor calculation: the share of the Risk Factors and Business sections of IPO registration statements (first S-1 or F-1) that mention AI at least once, by quarter. Patterns: `AI` (case-sensitive), `(?:artificial intelligence|generative ai|machine learning|large language models?|LLMs?)`.

## Why it matters

Shows how fast artificial intelligence moved from a niche word to a standard part of how companies describe their business and its risks when they first ask public investors for money.

## How it is made

The share of prospectuses that mention the topic at least once. Same documents, sections and patterns as the rate series. Each company's first registration statement on Form S-1 (U.S. issuers) or F-1 (foreign private issuers) from 2010, from Palanor's IPO pipeline (ipo.us.registrations.monthly; blank-check companies left out). Palanor fetches the prospectus from EDGAR and finds its "Risk Factors" and "Business" sections by their headings, from the heading to the next major prospectus section; a heading counts only before the financial statements, whose notes carry headings of their own. 7,315 prospectuses were counted; 128 in which neither section could be found are left out. A document's rate is mentions per 10,000 words of those two sections together; a quarter's reading is the mean of its documents' rates, so each company weighs the same, dated by the quarter the registration was filed. The rule that skips the financial statements was added after a hand check of five prospectuses found two whose Business section had been taken from the notes to the statements; every filing was then recounted (counter version 2). Documents under 300 words of section text are left out, and a quarter is published only with at least 30 documents. The topic word lists were fixed and committed on their own (src/lib/filings-text/topics.ts, 2026-10-20) before any counting code existed or any rate was computed; they are counted by regular expressions with no language model.

## How to read it

The series starts in 2010 Q1 at 0.0%. The latest reading, for 2026 Q3, is 73.2%. The high is 74.0% (2026 Q2) and the low 0.0% (2010 Q1). Quarters carry 56 to 219 documents.

## What it does not say

A mention is not exposure: companies list risks defensively, and a word can appear in a risk the company says it does not face. The rate moves with the mix of companies filing (a wave of Chinese or Hong Kong issuers lifts the China rate; a wave of biotech filings lowers the AI rate), not only with what a typical company writes. The quarter is when the registration was first filed publicly, which can trail a confidential draft by months, and many filings never become IPOs; small self-underwritten S-1s count the same as large IPOs. Words inside names count too: a company whose name contains "AI" adds an AI mention each time it names itself. A single passing mention counts the same as a long discussion.

## Source

Palanor calculation from SEC EDGAR filings (U.S. Securities and Exchange Commission; public domain). Palanor reads the filings to count words by fixed rules and never stores or republishes their text.