Generative AI for Earnings Call Scripts: A Reg FD Compliance Walkthrough for IR Teams

Generative AI is now inside the earnings call workflow at most large public companies. EY's 2025 IR Technology Survey found that 61% of IR professionals at S&P 500 companies used at least one AI tool in their earnings preparation process, up from 34% in 2023. The efficiency gains are real. The compliance risks are equally real, and almost no vendor or editorial source explains both in the same place.

This guide is for IR officers, CFOs, and general counsel at public companies who want a concrete, step-by-step answer to the question: how do we use generative AI to draft better earnings call scripts without creating a Reg FD violation, a material misstatement, or a document retention problem?

Key takeaway: The medium of communication does not create a Reg FD exemption. AI-drafted scripts, AI-generated Q&A prep documents, and AI voice-cloned audio are all subject to the same disclosure rules as a live, human-delivered call. The compliance obligation sits with the company, not the tool.

What Reg FD Actually Says About AI-Drafted Earnings Scripts

Regulation FD (17 CFR 243.100-243.103) prohibits selective disclosure of material non-public information to market participants without simultaneous public disclosure. The rule has been in force since August 2000 and has never contained a medium-specific exemption. It applies to earnings calls, pre-recorded remarks, AI-generated scripts, and any investor-specific messaging variants equally.

The public disclosure requirement for an earnings call is typically satisfied by filing a Form 8-K and holding an open, publicly accessible webcast. An AI-drafted script delivered via a public webcast meets that standard in exactly the same way as a live script. The content is what matters, not the drafting method.

Where AI introduces new Reg FD exposure is not in the prepared remarks themselves, but in two adjacent practices that IR teams are adopting without fully thinking through the compliance implications.

The Selective Disclosure Risk in AI-Generated Q&A Prep

AI tools are increasingly used to generate anticipated analyst Q&A documents, compressing what was once a multi-day manual process into hours. The risk: if those documents contain anticipated disclosures of material information that are then shared selectively with certain analysts or investors before the public call, the company may have committed a Reg FD violation before the call even starts.

The SEC has enforced Reg FD in cases where pre-call communications with analysts contained material non-public information. The fact that the prep document was AI-generated does not change the analysis. Treat every AI-generated Q&A prep document as a potential disclosure document subject to Reg FD review before it is shared with anyone outside the core preparation team.

The Audience-Variant Problem

This is the risk that compliance practitioners are not talking about enough. As Matt Kelly at Radical Compliance put it: "The intersection of AI and Reg FD is the sleeper compliance issue of the decade for IR teams. If you use AI to generate investor-specific messaging variants with materially different information, you may have just committed a selective disclosure violation."

AI makes it trivially easy to generate a retail investor version and an institutional investor version of the same earnings narrative. If those versions contain materially different information, that is selective disclosure. The efficiency benefit of AI-generated variants does not survive the Reg FD analysis. Variants must be limited to format and language complexity, never to substantive content.

The SEC's Position on AI-Generated Disclosures

The SEC has been unambiguous: AI does not change your disclosure obligations, and AI-generated errors do not provide a defense.

The SEC's Division of Corporation Finance Staff Bulletin No. 20 warned that AI-generated content in public disclosures must be reviewed for accuracy, and that companies remain fully responsible for the accuracy of all statements regardless of whether AI was used in drafting. The bulletin specifically flagged the risk of AI hallucinations producing inaccurate financial data.

Former SEC Chair Gary Gensler was direct in his June 2023 remarks: "AI does not change your disclosure obligations. If you use AI to make a materially misleading statement, you are still liable under the securities laws."

This position has been maintained under subsequent SEC leadership. The 2023 cybersecurity disclosure rule (effective December 2023) and the SEC's broader rulemaking trajectory signal that the Commission expects companies to disclose material risks from AI use in their operations, including investor communications.

The PSLRA Safe Harbor Problem for AI-Generated Guidance

This is a liability issue that almost no commentary addresses. The PSLRA safe harbor (15 U.S.C. 78u-5) protects companies from securities litigation over forward-looking statements, but only when those statements are accompanied by meaningful cautionary language.

AI-generated guidance language tends toward generic boilerplate. An LLM trained on historical earnings transcripts will produce forward-looking statements that sound plausible but may not reflect the company's actual internal projections, and will attach generic cautionary language that does not meet the "meaningful" standard. KPMG's 2025 report on AI in financial reporting flagged this as one of the highest-risk AI use cases in finance: AI-generated guidance language that lacks specific, tailored cautionary statements may not qualify for the safe harbor, increasing litigation exposure.

Every AI-drafted forward-looking statement needs CFO and General Counsel review specifically for whether the accompanying cautionary language is specific enough to the company's actual risk profile.

How IR Teams Are Actually Using Generative AI in the Earnings Workflow

PwC's 2024 CFO Pulse Survey found that 73% of CFOs reported their companies were actively piloting or deploying generative AI in finance functions, with IR-specific use cases among the top five cited applications. The most common use cases, per EY's 2025 survey, are:

1. First-draft script generation -- synthesizing financial results, KPIs, and strategic narrative into a structured prepared remarks draft
2. Analyst question anticipation -- generating likely Q&A based on results, peer calls, and analyst coverage history
3. Transcript summarization -- processing prior quarter transcripts and peer calls for benchmarking
4. Peer benchmarking of earnings language -- analyzing tone, forward-looking language, and disclosure patterns across comparable companies

Each of these use cases carries a different risk profile. First-draft generation and analyst Q&A anticipation carry the highest compliance risk. Transcript summarization and peer benchmarking carry lower direct disclosure risk but raise data governance questions about what financial data is being fed into which AI tools.

The Data Governance Question No One Asks

Before feeding financial data into any AI tool to draft an earnings script, the IR team needs to answer one question: is this tool processing material non-public information on infrastructure that is appropriately secured?

Using a public LLM API with unredacted draft financial results, unreleased KPIs, or internal projections is a data security and confidentiality risk. If that data is used to train or fine-tune a model, it may also constitute a disclosure risk. The answer is not to avoid AI, but to use enterprise-grade, data-isolated deployments and to confirm with legal counsel and IT security that MNPI is not being transmitted to shared model infrastructure.

For the audit trail implications: SEC document requests and litigation discovery can encompass AI prompts and outputs. Version-controlled logs of what was prompted and what was generated are not optional if you want to be able to respond to a regulator inquiry. See Finrep's AI audit trail requirements walkthrough for the document retention framework.

The Hallucination Risk in Earnings Scripts

Academic NLP research on earnings call transcripts, including work from Stanford GSB, has shown that the language of prepared remarks is highly predictive of subsequent stock price movements and analyst forecast revisions. An AI-generated script that inadvertently signals more or less optimism than intended, or that produces a hallucinated revenue figure, can have material market impact.

The scenario IR teams fear most: an AI-drafted script contains a hallucinated EPS figure or revenue number that gets read aloud on a live call. What happens next?

• The company must correct the record promptly. A Form 8-K is the standard vehicle for correcting a material misstatement made in a public forum.
• If the error was in the prepared remarks filed as an exhibit to an 8-K, an amended 8-K may be required.
• The fact that AI generated the error is not a defense under Rule 10b-5. The company reviewed and approved the script.
• The SEC's Office of the Whistleblower has received increasing numbers of AI-related disclosure tips since 2024. Employees and analysts are watching.

The Deloitte 2024 CFO Signals survey found that accuracy of financial figures was the primary concern among IR professionals using AI in earnings preparation, ahead of regulatory compliance. Both concerns are legitimate. The governance framework below addresses both.

For a deeper treatment of the review process for AI-drafted financial commentary, see Finrep's CFO process guide for reviewing AI-drafted financial commentary.

The Governance Framework: Four Steps Before the Script Goes Live

Skadden Arps' February 2025 client memo on AI governance in investor relations recommended a four-step framework. Here is how to implement it in practice.

Step 1: Designate an AI Output Owner for Each Section

Every section of an AI-drafted script needs a named human owner who is accountable for its accuracy. This is not a group review -- it is an individual accountability assignment. The owner for the financial results section is typically the CFO or controller. The owner for the strategic narrative section is typically the CEO's chief of staff or head of IR.

The AI output owner reviews the AI draft against the source financial data, not against their memory of what the numbers should be. Every figure in the script is traced back to the audited or reviewed financial statements and the earnings release draft.

Step 2: CFO and General Counsel Sign-Off on Forward-Looking Statements

Any AI-generated forward-looking statement -- guidance, outlook language, commentary on future performance -- requires explicit CFO and General Counsel sign-off before it goes into the final script. This is not a rubber stamp. The GC review specifically addresses:

• Whether the cautionary language accompanying each forward-looking statement is specific and meaningful under the PSLRA standard
• Whether any statement could be read as inconsistent with internal projections or board-approved guidance
• Whether any statement contains information that has not been publicly disclosed and that would need to be disclosed simultaneously

Step 3: Version-Controlled Logs of AI Prompts and Outputs

Maintain a version-controlled log of every AI prompt used in the script drafting process and every output generated. This log should capture:

• The date and time of each prompt
• The tool and model version used
• The prompt text (including any financial data provided as context)
• The AI output
• The human edits made to the output
• The name of the AI output owner who reviewed that section

This log is a potential SEC inquiry response document. It is also your defense if a hallucinated figure is later discovered: the log shows what the AI produced and what the human reviewer approved.

Step 4: Reg FD Review of the Q&A Prep Document

The AI-generated Q&A prep document must be reviewed by legal counsel before it is shared with anyone outside the core preparation team. The review asks one question: does this document contain any anticipated disclosure of material information that has not yet been publicly disclosed?

If the answer is yes, that information either needs to be included in the public call or removed from the prep document. The prep document itself should be treated as a potential disclosure document for document retention purposes.

NIRI's 2024 AI guidance is consistent with this framework: "NIRI recommends all AI-drafted communications undergo legal review, CFO sign-off, and be treated as equivalent to any other draft disclosure for purposes of Reg FD compliance."

AI Voice Cloning for Earnings Calls: What the Compliance Rules Actually Say

Vendors like ViaVid now offer AI voice cloning for pre-recorded earnings call remarks, generating broadcast-quality audio from a text script using a cloned executive voice model (ViaVid recommends a minimum 3-5 minute voice sample). The efficiency case is straightforward: no studio time, instant regeneration if a figure changes at the last minute.

The compliance picture is less tidy. The SEC has not issued specific guidance on AI voice cloning in earnings calls as of September 2026. The general principle under Reg FD and the Securities Exchange Act is that the content of the communication determines compliance, not the method of delivery. A voice-cloned script that contains a material misstatement carries the same liability as a live misstatement.

On disclosure: ViaVid's own marketing acknowledges that "disclosures can be made where required" -- which implies uncertainty about when disclosure is legally required. The honest answer is that no rule currently mandates disclosure of AI voice cloning in earnings calls. But the trend is toward greater transparency. Compliance Week reported in early 2025 that several large-cap companies had begun including voluntary boilerplate noting that portions of their remarks were prepared with AI assistance and reviewed by management and legal counsel.

The Harvard Law School Forum on Corporate Governance raised a non-legal risk worth taking seriously: earnings calls that are entirely AI-generated risk becoming informationally thin and formulaic, frustrating analysts and reducing the quality of price discovery. Their recommendation: use AI for structure and compliance checking, but preserve executive voice for substantive strategic commentary.

AI Use Cases in IR: Risk Comparison

Governance Checklist for AI-Assisted Earnings Script Preparation

Use this before every earnings call where AI tools are part of the drafting workflow.

Before drafting:

• Confirm AI tool is deployed on data-isolated, enterprise-grade infrastructure (no MNPI to shared public models)
• Assign a named AI output owner to each script section
• Confirm version-controlled logging is active for all prompts and outputs

During drafting:

• Every financial figure in the AI draft is traced to the source financial statement or earnings release draft
• No AI-generated Q&A prep document is shared outside the core preparation team before legal review
• Audience-variant scripts are reviewed to confirm no material differences in content

Before final approval:

• CFO sign-off on all financial figures and forward-looking statements
• General Counsel review of all forward-looking statements for PSLRA cautionary language adequacy
• Legal review of Q&A prep document for Reg FD compliance
• Version-controlled log of AI prompts and outputs is complete and archived

After the call:

• If any material error is discovered post-broadcast, assess Form 8-K correction obligation immediately
• Archive the final script, AI draft, and version log per document retention policy
• Consider whether voluntary AI-use disclosure language is appropriate for next quarter

For the broader LLM and MNPI data leakage framework, see Finrep's LLM MNPI data leakage and Reg FD walkthrough. For the Reg FD compliance rules in full, see Finrep's Regulation FD practitioner guide.

FAQ

Does using ChatGPT or another LLM to draft our earnings script violate Reg FD? No, not by itself. Reg FD governs what information is disclosed and to whom, not how the script is drafted. Using an LLM to produce a first draft that is then reviewed, corrected, and delivered via a public webcast does not create a Reg FD violation. The risk arises when AI-generated Q&A prep documents containing material information are shared selectively, or when AI generates materially different messaging variants for different investor audiences.

Does the SEC require us to disclose that we used AI to draft our earnings script? No current SEC rule mandates disclosure of AI use in earnings script drafting specifically. The trend among compliance-conscious IR teams is toward voluntary disclosure, with boilerplate noting AI assistance and human review. The SEC's broader rulemaking direction suggests this expectation will tighten over time.

What happens if an AI-generated script contains a hallucinated financial figure that gets read on the call? The company must correct the record promptly. A Form 8-K is the standard vehicle. The AI origin of the error is not a defense under Rule 10b-5. The governance framework above, particularly the requirement to trace every financial figure to source documents before sign-off, is the primary safeguard.

Can we use AI to generate multiple versions of our earnings messaging for different investor audiences? Only if the versions are limited to format and language complexity. If the versions contain materially different information, that is selective disclosure under Reg FD. This is the highest-risk AI use case in IR and the one most commonly overlooked.

Is AI voice cloning for pre-recorded earnings calls legally compliant? Yes, under current rules, provided the content of the script is compliant. No SEC rule currently requires disclosure of AI voice cloning. Voluntary disclosure is a growing best practice. The liability for any misstatement in a voice-cloned script is identical to the liability for a live misstatement.

Do AI prompts and outputs need to be retained for SEC purposes? Yes, as a practical matter. SEC document requests and litigation discovery can encompass AI prompts and outputs used in the preparation of public disclosures. Version-controlled logs of prompts, outputs, and human edits should be retained under the same policy as other disclosure-related working papers.