AI-readable disclosures are 10-Ks, 10-Qs, 8-Ks and other corporate filings drafted so AI engines can extract claims, numbers and risk factors cleanly. They combine legal defensibility with retrieval-friendly drafting: claim-anchored numbers, one name per entity and specific risk descriptions. The aim is to prevent the summary distortion that shapes how analysts, acquirers and reporters frame an issuer when ChatGPT, Claude or Perplexity is their first-pass research tool.
By the EPR Editorial Team. Originally published June 2026. Updated October 2026. Everything-PR and 5W AI Communications share common ownership; Everything-PR reports independently on the communications industry.
Why do disclosures need to be readable by AI engines?
Disclosures need to be readable by AI engines because analysts now use those engines to summarize an issuer before they open the filing. A document written only for legal defensibility can produce a vague or wrong summary, and that gap between the filed record and the machine summary is the exposure. EPR calls it Retrieval Risk.
Most 10-Ks were drafted to survive litigation, not to be summarized in 50 words. Both goals now matter, and the same paragraph can serve both when the claim comes first and the protective language follows.
What are the six drafting rules for AI-readable disclosures?
The six drafting rules apply to every filing, script and investor document an issuer publishes.
Lead with the claim and put the qualifier second. A sentence that opens with its conclusion gets summarized cleanly, while one that buries the conclusion under three protective clauses gets summarized into whatever the model can extract. Keep both the claim and the legal shield in the sentence.
Use one name for the company, one for each segment and one for each product line across the 10-K, 10-Q, 8-K, proxy, earnings script, investor deck and IR page. Inconsistent names create what EPR calls Entity Drift, where each variation competes as a separate reference.
Put the number inside the claim. "Revenue grew 12% to $1.4 billion" anchors better than "Revenue grew, reaching $1.4 billion (a 12% increase)," because the figure and the claim stay together when a model cuts the sentence.
Repeat the same wording for the same claim on every surface. If the 10-Q and the earnings call phrase a result differently, the model learns that the narrative is unstable.
Write risk factors as named, distinct exposures instead of boilerplate. Models pull from the risk section when asked about company risk, and boilerplate comes back as risk-shaped paragraphs that say nothing.
Use the MD&A to state what management thinks is happening. A vague MD&A produces vague summaries, and a specific one produces specific summaries.
Why it works. A language model builds a summary by selecting and compressing passages it can identify as answering the question, so a claim with its number and entity name in one sentence survives compression and a claim split across clauses does not. EPR has not yet published controlled test data for these six rules; the mechanism and the examples come from EPR's editorial testing and should be read as practitioner guidance.
What should a company audit before the next 10-Q?
A company should audit three sections of the next 10-Q: risk factors, MD&A and the segment results discussion. Read each paragraph and ask what a model would extract from it in a 50-word summary. If the answer is vague or wrong, redraft that paragraph before filing.
The same test works on the S-1 for a company heading to market. See IPO communications for how the S-1 risk factors become the phrases engines repeat.
What does Regulation FD cover, and what does it leave open?
Regulation FD, adopted by the SEC in 2000, requires an issuer to disclose material non-public information broadly when it discloses it to market professionals. The rule was written for human conversations, so it does not address how an AI tool combines sources into one summary for a portfolio manager.
That leaves three open questions for general counsel. They are questions, not settled law.
Whether giving paying enterprise users a better-sourced answer than free users counts as selective disclosure.
Who is accountable when an engine keeps returning old guidance for days after an 8-K changes it.
Who is accountable when a model invents a settlement, an executive departure or a guidance number that moves the stock.
The nearest real enforcement example involves a human channel: DraftKings paid a $200,000 SEC penalty after its PR firm posted unreleased growth data on the CEO's accounts, covered in the DraftKings Regulation FD case. Issuers should assume regulators will extend the same logic to new channels, and should document how they monitor what AI engines say about the company.
What should a public company CFO disclose about AI?
A public company CFO should disclose AI strategy, operational use, risk, governance and vendor concentration in specific, structured language. Investors ask about each of these in earnings calls and in AI-assisted diligence, and generic wording leaves the company with a vague answer or none.
Write one to two paragraphs on AI strategy in the annual report, the 10-K business section and the investor relations site, covering how AI changes competitive position.
State where AI is deployed, what it cost and what productivity or margin benefit it produced, with numbers where the company has them.
Separate AI risk into three kinds: risk from building AI, risk from deploying third-party AI, and risk from AI-specific regulation in the company's industry. A line such as "AI may present risks" is too general and can create liability if a material AI risk later appears undisclosed.
Name who at board and executive level owns AI oversight, whether a governance policy exists and what review mechanisms apply.
Disclose material AI vendor relationships, their cost and any concentration risk, and say so when the company uses several vendors.
How does the investor relations website affect AI answers?
The investor relations website is a primary source that engines read for company-specific queries, but many IR sites are not cited at all. In the EPR IR Page Citation Audit published June 29, 2026, financial services IR pages were cited close to 100% of the time and biotech about 92%, while mega-cap tech and junior mining pages were cited close to 0%. Read the findings in the IR Page Citation Audit.
Retrieval Risk is the exposure created when disclosures are written for legal defensibility alone. AI engines then produce vague or distorted summaries that shape how analysts, acquirers and reporters describe the company.
What is Entity Drift?
Entity Drift is the loss of consistent recognition that happens when a company uses different names for itself, its segments or its products across filings, scripts, decks and web pages. Each variant competes with the others as a separate reference.
Should risk factors be specific or boilerplate?
Risk factors should be specific. Models pull from the risk section to answer questions about company risk, and named, distinct exposures come back as real risks while boilerplate comes back as filler.
Does this apply to private companies?
Yes. Acquirers, partners and investors use AI tools for first-pass diligence on private companies too, so the same rules apply to investor decks and due-diligence answers.
Where do earnings calls fit?
Earnings call transcripts are a major source for AI summaries of an issuer, so script wording should match the filings. See the earnings call communications playbook.
Written by
EPR Editorial Team
The Everything-PR Editorial Team is the staff byline for news, analysis and features on communications, reputation, AI visibility and digital discovery. Everything-PR has published since 2009. AI tools assist with research and drafting, and every article is reviewed by a human editor before publication. Coverage follows the Editorial Policy, and substantive corrections are noted on the article under the Corrections Policy.
