Implementation Guide

    Classifying AI-Drafted Messages Under FINRA Rule 2210

    Rule 2210 classifies a message by who receives it and how many retail investors it reaches, not by whether generative AI drafted it. A message distributed or made available to 25 or fewer retail investors in any 30 calendar days is correspondence. More than 25 makes it a retail communication. A message distributed only to institutional investors is an institutional communication. Each category carries different supervision, approval and filing consequences, so firms need a documented classification step before release and a record of how the determination was made.

    A classification workflow to run before release

    1. 1

      1. Identify the audience

      Record whether intended recipients are retail or institutional investors, and who made that determination.

    2. 2

      2. Test for forwarding risk

      If the draft targets institutional recipients, ask whether the firm has reason to believe it or an excerpt will reach a retail investor. If so, it cannot be treated as institutional.

    3. 3

      3. Count retail recipients

      Apply the firm's documented rolling 30-day count for this message and its template family, including planned reuse.

    4. 4

      4. Assign the category

      Mark the message correspondence, retail communication or institutional communication, with a short rationale.

    5. 5

      5. Route to the matching review

      Send it to correspondence supervision, principal pre-use approval or institutional procedures. Hold release until the required step is recorded.

    6. 6

      6. Retain the determination

      Store the rationale, reviewer, approval decision and final message together so they can be retrieved as one record.

    Evidence to retain for each determination

    • The prompt or instruction that produced the draft
    • The generated output and the final released version
    • Audience type, retail recipient count and the 30-day window applied
    • Distribution channel and any template reuse
    • Reviewer identity, approval decision and timestamp
    • Applied policy and the written classification rationale

    Audience and volume decide the category, not AI authorship

    Rule 2210 defines communications as correspondence, retail communications and institutional communications. Correspondence is any written, including electronic, communication distributed or made available to 25 or fewer retail investors within any 30 calendar-day period. A retail communication is distributed or made available to more than 25 retail investors in that period. An institutional communication is distributed or made available only to institutional investors, and it does not include a member's internal communications.

    FINRA states that its rules are technology neutral and apply when a firm uses AI, including generative AI, as they would for any other tool. That holds whether the firm built the tool or uses a third party's, including embedded features in existing products. Which rules apply depends on how the firm deploys the technology. FINRA's guidance on AI-created chatbot communications says they may be correspondence, retail communications or institutional communications depending on the nature and number of recipients, and the firm must supervise them under Rules 2210(a), 2210(b), 3110(b)(4) and 3110.06 through .09.

    The practical consequence is that AI does not create a fourth category. It changes the risk profile. One template can be sent to many recipients and reused across channels, so a message that began as one-to-one outreach can cross the 25-retail-investor threshold without anyone deciding that it should.

    Counting recipients when one template reaches many people

    The definitions turn on a communication being distributed or made available, which makes the retail recipient count over a rolling 30 calendar-day window the relevant measure. The rule text does not say how to count edited variants of a template or reused outputs across advisers, campaigns and channels. Firms need to define their own method and apply it consistently.

    A workable method states what counts as the same communication (for example, a template family with substantive content unchanged), which channels aggregate into one count, and who owns the counter. A per-template rolling tally that flags when cumulative retail distribution approaches 25 gives reviewers time to reclassify and route the message before it crosses the threshold. The method will be examined as closely as any single determination, so it should be written into supervisory procedures and applied the same way each time.

    Institutional drafts need a separate control. Under Rule 2210(a)(4), a member may not treat a communication as distributed to an institutional investor if it has reason to believe the communication or any excerpt will be forwarded or made available to a retail investor. FINRA points firms to policies and procedures reasonably designed to prevent institutional communications from being forwarded to retail investors. Reusing an institutional draft as the basis for a retail message should trigger reclassification, not a quiet edit.

    What each classification triggers

    The category determines the review path. The items below summarize the consequences that classification procedures need to map to.

    • Correspondence: Subject to the supervision and review requirements of Rules 3110(b) and 3110.06 through .09, plus the content standards that apply to all communications.
    • Retail communication: An appropriately qualified registered principal must approve it before the earlier of its use or its filing with FINRA's Advertising Regulation Department. Filing requirements for specified retail communications depend on content.
    • Retail communication exceptions: Regulatory Notice 26-14 describes current rules under which principal pre-use approval does not apply to posts on an online interactive electronic forum, or to retail communications that make no financial or investment recommendation and do not otherwise promote a member product or service. These may be supervised like correspondence.
    • Institutional communication: Distributed only to institutional investors, never treated as such where forwarding to retail investors is reasonably expected. Internal firm communications are outside the definition.
    • All categories: Content must be fair and balanced, based on fair dealing and good faith, and free of false, exaggerated, unwarranted, promissory or misleading statements.

    Where runtime enforcement supports the procedure, and what to monitor

    Written supervisory procedures must reflect a reasonably designed supervisory system tailored to the firm's business under Rule 3110. The firm's drafting-tool inventory should include embedded generative AI features, not only internally built tools, because FINRA rules apply to both.

    Enforcement in the AI execution path can support the routing and evidence steps above. Trussed enforces policy in the execution path for individual interactions and tool calls, with actions to allow, redact, block, or escalate for human approval, and it produces audit-ready evidence from that enforcement. In this context, a policy could hold a draft until a principal approves it, and the interaction record would show the policy applied and the approval outcome. Trussed does not make regulatory determinations. The classification, the recipient-counting method and the approval standard remain the firm's decisions, and a cumulative count across campaigns needs the firm's own tracking.

    As of November 2026, SR-FINRA-2026-004 proposes an amendment to Rule 2210, and Regulatory Notice 26-14 requests comment on modernizing the rule, including firm-written procedures to determine which categories of retail communications require principal pre-use approval. Neither is described here as adopted. FINRA has also said it will consider further guidance on how particular rules apply to specific generative AI use cases, so procedures should be reviewed as that guidance and any rule change appear.

    Classification decision path for an AI-drafted message

    Audience type

    Retail or institutional recipients, including reason to believe of forwarding.

    Recipient count

    Retail investors reached in any 30 calendar-day period.

    Category

    Correspondence, retail communication or institutional communication.

    Review path

    Supervisory review or principal pre-use approval, as the category requires.

    Retained record

    Classification rationale, approval decision and the message itself.

    Turn classification decisions into evidence

    Discuss how policy enforcement in the AI execution path can support approval routing and audit-ready evidence for your supervisory procedures.

    Talk to an Expert
    Next step

    How does your AI governance program compare?

    See where your program has gaps in less than 2 minutes.