The Ethical AI Content Workflow for Law Firms: A 2026 Playbook
A seven-stage operational SOP for legal marketing teams that want AI in the pipeline without exposing the firm to a bar complaint, a malpractice question, or a fabricated citation on a public blog.
Why the ethical AI question sits harder on law firms than on anyone else
A software company that ships an AI-drafted blog post with a wrong statistic has a correction to make. A law firm that ships one with a wrong statute citation, an implied guarantee of outcome, or prescriptive advice to a non-client in a specific jurisdiction has a duty problem. That is the asymmetry that changes how a legal marketing team should treat generative AI. The output looks like any other content deliverable. The consequences of getting it wrong are not.
The bar-adjacent duties stack up quickly. Model Rule 1.1 sets a competence duty and comment 8 has included technology competence for years, so partners cannot claim ignorance of how the marketing team's tools work. Model Rule 1.6 covers any information relating to a client representation, privileged or not, which is why matter facts do not belong in a public model prompt. Model Rule 5.5 forbids unauthorized practice, including by non-lawyers under the firm's supervision, and it is the rule most quietly violated when an AI draft slips from information into prescription. Model Rule 7 governs advertising and solicitation, with a state overlay of ad-review requirements, testimonial restrictions, and outcome-language prohibitions.
ABA Formal Opinion 512, issued in July 2024, ties these rules to generative AI. It is a framework, not a prohibition, reminding lawyers they retain full responsibility for competence, confidentiality, communication, supervision, and reasonableness of fees when AI reduces time on billable tasks. Ethical AI use in legal work is not about picking the right tool. It is about designing the workflow so lawyer judgment is inserted at every point where the output could cause a duty problem.
Marketing content lives inside that same duty envelope. The firm's website is a public statement by lawyers, subject to advertising rules, competence rules, and, for testimonials or client stories, confidentiality rules. A workflow that lets AI draft or distribute firm content without disciplined lawyer touchpoints puts the firm in the position of having advertised, communicated, or advised through a tool it did not adequately supervise.
What ethical AI use in legal content actually means, in operations
Every legal marketing team I have talked to about this has the same first question. They want a rule that tells them where the line sits. The rule is not complicated, but it breaks into three concrete decisions the team makes every day.
The first decision is which stages of the workflow AI touches at all. AI is a drafting and structuring tool. It is not a source of authority, not a verifier, and not an editor of last resort. The stages where AI adds real speed are outline generation, first-draft prose, alt-text and metadata suggestions, and reformatting approved copy for a new surface such as a knowledge-panel answer, an FAQ block, or a schema entry. The stages where AI must not have the final word are fact and citation verification, voice and tone, attorney review, compliance review, and publication.
The second decision is which data goes into the model. This is the confidentiality line, and it is the line firms most often blur without noticing. Public-model prompts never contain client names, opposing party names, docket numbers, settlement figures, minor children's initials, medical records, deposition excerpts, or any factual sequence that a person familiar with the matter could reverse-engineer. That last category is the sneaky one. A prompt like "draft a case study about a plaintiff who fell at a national grocery chain in Cleveland in 2024 and settled for a mid-seven-figure amount" contains no name, but for a specific matter that description may be identifying, and it is now in a vendor's training or logging pipeline. The safer default: nothing about any actual client or matter enters a model, and case studies use only material for which the client has signed a written publication release.
The third decision is disclosure. If AI participated in producing a piece, the firm's editorial policy should say so, name the human editor and reviewing attorney, and confirm that no client facts appear. A four-line note at the foot of the piece is enough. It signals process discipline to sophisticated readers and lines up with the communication duty Opinion 512 walked through in detail.
Mapping the ABA Model Rules to real marketing decisions
A marketing lead who cannot name the rules the workflow protects against will eventually design around the wrong risk. The exercise that grounds this whole conversation is walking through the specific rules that touch AI-drafted content and asking, at each one, what the concrete production decision looks like. The rules below are the ABA Model Rules, which every state adopts with local variation. Consult the state rule where the firm actually practices before treating any of this as final.
Rule 1.1 competence, and the technology-competence comment
Rule 1.1 obligates competent representation. Comment 8, adopted in 2012 and referenced by most state bars, extends that to a duty to keep abreast of the benefits and risks of relevant technology. If a partner cannot describe what happens to a prompt the marketing team pastes into a chatbot, the partner is not currently competent on that tool. The remedy is not that the partner learns to prompt. The remedy is that marketing produces a short internal write-up of every AI tool in the workflow, including data flow, retention posture, and what a leaked prompt would look like, and the responsible partner signs it.
Rule 1.6 confidentiality
Rule 1.6 covers information relating to the representation of a client, which is broader than attorney-client privilege and broader than trial-strategy material. Anything the firm learned because it represents the client is covered, including facts the client shared casually and facts in public filings surfaced through the matter. When marketing drafts a case study on a "recent Cleveland grocery slip and fall," the 1.6 question is whether a reader familiar with the client or matter could identify either from the piece. If yes, the piece needs written informed consent before it appears anywhere and it never enters a public LLM at any stage.
Rule 3.3 candor toward the tribunal
Rule 3.3 does not directly govern marketing, but it becomes marketing's problem when the firm publicizes a court result, quotes a filing, or references a verdict. A blog post that misquotes a filing, misrepresents a verdict, or attributes a favorable dictum to a case that never held it exposes the firm to the same candor question it would face in a brief. AI hallucination on case names and quoted holdings is the direct vector, which is why every quoted line from a case is pulled from the primary source, not from the model.
Rules 5.1 and 5.3 supervision of lawyers and non-lawyers
Rule 5.1 covers a partner's supervisory duty over other lawyers. Rule 5.3 extends the duty to non-lawyer assistants, including outside contractors and, under Opinion 512 and its state analogs, the AI tools they use. The firm cannot outsource its ethics to a marketing agency. If the agency drafts content using AI, the responsible partner remains on the hook. The workflow answer is a signed engagement letter naming the agency's obligations, plus a documented review path that ends at a lawyer's signature before publication.
Rule 5.5 unauthorized practice of law
Rule 5.5 is the rule most quietly violated by AI-drafted marketing. It forbids unauthorized practice by non-lawyers under the firm's supervision. When AI produces "what to do after a workplace injury in Georgia" and the draft slips into prescriptive, jurisdiction-specific, imperative advice, the firm has published a document that reads, to a Georgia reader, as legal advice from a Georgia lawyer. The reader may act on it. The firm may not have intended to accept the responsibility. The workflow answer is the voice-edit stage that rewrites imperatives into informational language, plus attorney review that reads specifically for prescriptive drift.
Rules 7.1 through 7.5 communications and advertising
Rule 7.1 prohibits false or misleading communications. Rule 7.2 governs advertising, including compensation for referrals and testimonial rules. Rule 7.3 governs solicitation, especially direct solicitation of people known to need legal services. Rule 7.4 covers communication of fields of practice and specialization, including the word "specialist" which many states restrict. Rule 7.5 governs firm names, letterheads, and professional designations. AI reliably produces copy that trips each of these when asked for "punchy homepage copy." Every headline, subhead, call to action, testimonial, outcome reference, and specialty claim runs through the state-specific advertising checklist before publication.
How state advertising rules diverge, and what that means for a multi-state firm
ABA Model Rules 7.1 through 7.5 are the baseline. States adopt with local variation, and the variation is wide enough that a piece that clears advertising review in one state can be a filed complaint in another. A firm marketing across state lines needs a compliance workflow that runs a state-specific pass on every piece, not a single national pass. The four states below illustrate the range.
Florida, and the aggressive filing and testimonial regime
The Florida Bar operates one of the most detailed lawyer advertising regimes in the country. Rules 4-7.11 through 4-7.23 of the Rules Regulating The Florida Bar govern content, and many ad forms require submission to the bar's Ethics and Advertising Department for review, with narrow exemptions. Testimonials are permitted but heavily conditioned. Comparisons to other lawyers, statements characterizing quality, and specific case results all trigger substantive rules. Paid media targeting Florida clients cannot be published on autopilot. Every ad passes through a Florida-specific checklist that flags outcome language, testimonial framing, and characterizations of quality.
Texas, and the specific disclosure obligations
The Texas Disciplinary Rules include detailed advertising rules under Part VII, administered by the Advertising Review Committee. Solicitation rules restrict certain direct contacts. Past-result advertising is permitted but requires specific contextualizing disclosures and firm substantiation. The rule most likely to catch an AI-drafted piece is the requirement that the advertisement identify at least one lawyer responsible for its content. AI does not know to insert that identification. The Texas checklist line for responsible-lawyer identification catches it every time.
California, and the outcome-language restrictions
The California Rules of Professional Conduct, including Rule 7.1 and related sections, prohibit statements that create unjustified expectations about results. The prohibition on guarantees and warranties reaches subtle language a model will happily produce, such as "we get results" framed as a promise. Asked for "confident, benefits-oriented copy," a model reliably drifts into implied guarantees. The California pass rewrites every outcome verb in the passive or conditional and hedges every guarantee-adjacent noun. The checklist asks whether any sentence could be read as guaranteeing a result.
New York, and the retention regime
New York permits broader advertising than Florida but imposes strict retention obligations under Rule 7.1 of the New York Rules of Professional Conduct. Firms must maintain a copy or recording of an advertisement for at least three years after its last dissemination, with longer periods for some formats. That retention duty is where ephemeral chatbot sessions break workflows. The answer is that every piece the firm publishes in a retention-rule state leaves an archived working folder with the model outputs, human edits, and signed reviews.
The multi-state implication
A firm licensed in multiple states does not get to pick the most permissive state and publish under those rules. The controlling rule is generally the state where the target audience sits or where the lawyer is admitted. A firm with offices in Florida, Texas, California, and New York needs four active checklists, and the piece does not publish until it clears every one for every state where it will appear. The compliance stage runs the primary-state checklist and a shorter secondary pass for any state the piece may reach through paid amplification, syndication, or organic. Running national copy under the loosest state's rules is the reliable path to a bar complaint.
The citation problem, and why Mata v. Avianca should be taped to every marketer's wall
The single most reliable way for AI-drafted legal content to embarrass the firm is a fabricated citation. Large language models are pattern completers, and case names, docket numbers, statute sections, and quoted holdings are exactly the structured, plausible-sounding pattern that models produce even when the underlying record does not exist. Every legal marketing team should know the Mata v. Avianca story by heart. In 2023, two attorneys in the Southern District of New York filed a brief citing multiple ChatGPT-generated opinions that did not exist. Judge P. Kevin Castel imposed sanctions of five thousand dollars per attorney and required a corrective notice to each of the judges named in the fabricated opinions. The story lives on because it was not a technology failure. It was a verification failure.
Marketing content is not court filings, but the credibility stakes are similar. A blog post citing a nonexistent Ohio Supreme Court decision, or misquoting a live statute, is a document a prospective client, a referring attorney, or a bar counsel can read and use against the firm. The rule that follows is simple and non-negotiable. Every citation produced by a model, whether case name, statute number, regulation cite, secondary source, or quoted holding, is verified against the primary source before the piece moves to attorney review. Not skimmed. Verified. If the source cannot be found, the citation comes out.
The same discipline applies to statistics. Injury settlement averages, immigration wait times, divorce filing rates, workers' compensation approval percentages, and any other numeric claim gets a source line in the working document and a link to a public authority in the published piece. Numeric hallucinations are harder to spot than case hallucinations because they read as harmless flavor. They are not harmless when the piece is the first Google result and a reader relies on the figure to decide whether to file.
What made Mata a workflow lesson rather than a personal-negligence lesson is the sequence of choices that led to the filed brief. The attorney asked a general-purpose chatbot to find supporting cases. He then asked the same chatbot whether the cases were real. The chatbot said yes. He believed it. Every stage of that sequence was a workflow gap. There was no verification pass against a legal research platform, no independent human read, no supervising attorney who touched the brief before it went to court. Applied to marketing, the sequence is the same one that produces a blog post with a hallucinated case cite, and the workflow controls that prevent it are also the same. Verification lives outside the model. Verification is a signature on a checklist.
Four failure cases the seven-stage workflow prevents
The rule is easier to internalize when the failure is concrete. Below are four failure patterns I have either seen at firms I have consulted with or heard described often enough by legal marketers that they should be treated as recurring risks. Names and identifying details have been changed or generalized. Each walkthrough describes what broke, what should have caught it, and which stage of the workflow is the control.
Failure one. A personal injury blog post that read as legal advice
A regional plaintiff-side firm asked an outside vendor to publish a weekly evergreen piece. The vendor used a general chatbot to draft "What to do in the first 72 hours after a car accident in the state." The model produced a confident, imperative, jurisdiction-specific list. Do not speak to the other insurer. File a report with a specific state department within a specific number of days. Use a specific medical treatment sequence. The piece published without an attorney read. A prospective client followed the medical sequence and later filed suit alleging the published advice caused her to miss an evaluation that would have caught a secondary injury.
What broke was the movement from information to prescription. A jurisdiction-specific imperative directed at a non-client reader crosses the line under Rule 5.5. The voice-edit pass in stage five rewrites imperatives into informational language; the attorney review in stage six reads specifically for prescriptive drift. Neither review happened because the vendor's workflow did not have those stages. The workflow answer: republish with informational framing, a clear disclaimer, and a call to consult counsel; change vendors.
Failure two. A firm bio page that invented a credential
A boutique firm asked a junior marketer to refresh partner bios using a chatbot. The prompt was helpful and specific. Take the partner's existing bio, add professional depth, and expand the education section. The model, on one bio, added a fellowship the partner had never held. It looked plausible because the fellowship name existed and was in the partner's field. The bio published to the firm's site, was picked up by a legal directory, and appeared in a written recommendation the partner submitted for another matter. Opposing counsel spotted the discrepancy in a bar directory check and raised it in a motion.
What broke was the absence of a fact-check pass against the partner's own record. The stage-four verification log would have flagged any credential the model produced that did not appear there; the stage-six attorney review would have caught it if the partner read her own bio. Neither happened, because the marketer treated a bio refresh as low-risk copywriting rather than as a claim about a lawyer's qualifications. The workflow answer: bio pages route through the same fact-check discipline as blog posts, and every credential in every bio traces to a source line. Rule 7.1 does not exempt bios.
Failure three. A paid search ad with an implied outcome guarantee
An immigration firm ran a paid search campaign on high-intent keywords in a state that prohibits statements creating unjustified expectations. The agency used a chatbot to generate ad-copy variants at scale. Asked for punchy, benefits-oriented copy, the model produced a headline stating, in effect, that the firm secured positive outcomes in a large majority of the targeted visa category. The number was plausible and read as a statistic; it was actually a paraphrase of an industry figure, not a firm-specific claim. It ran three weeks, drew a competitor complaint, and cost the firm an ad-review inquiry from the state bar.
What broke was the state-specific compliance pass at stage six. The workflow answer: no paid-search headline runs without the ad copy sitting on the campaign state's compliance checklist, and any numeric outcome claim must have a firm-specific source or come out. Paid media at scale is where the ethical AI workflow either holds or breaks, because the volume of variants makes it tempting to bypass the review that would have caught the single problem line.
Failure four. A testimonial that AI paraphrased into a different meaning
A family law firm collected written client testimonials over the year. The marketing team asked a chatbot to shorten and polish the testimonials for a website refresh. The model tightened the language, and in the process it paraphrased one client's statement in a way that changed the meaning. The original said the firm helped her feel prepared for a hearing. The paraphrase said the firm helped her win her case. She had won her case, but she had not said so in her testimonial, and the state bar's advertising rules treat a testimonial about a specific outcome as a specific-result claim with its own disclosure requirements. The testimonial ran without the required disclaimer.
What broke was the assumption that a testimonial edit is a copy edit. A testimonial edit that changes substance is a new statement attributed to the client, and it needs both the client's confirmation and the state-specific compliance pass. The workflow answer: testimonials are never AI-paraphrased. They are lightly copy-edited for typos, the client signs off on the final language, and the piece clears the state's testimonial-specific compliance check. Stage five voice edit does not apply to quoted client speech.
The seven-stage ethical AI content workflow
This is the operational SOP I run and teach. Each stage has a clear split between what the model produces and what a human owns. The split does not change with the model, the agency, or the practice area. The split is the framework. The rest is craft.
Stage 1. Topic sourcing and attorney SME intake
Every piece starts with a live conversation between the content lead and the attorney who owns the practice area. Thirty minutes on the calendar, a shared doc, a recording only if the attorney consents. The content lead comes in with a proposed topic informed by search demand, competitor gaps, and the firm's editorial calendar. The attorney comes in with the substantive expertise, the current-year updates that matter, the pitfalls other blogs get wrong, and the practical stories, generalized and de-identified, that make the piece worth reading.
The attorney is the human, no exceptions. The content lead can prep the interview with the assistance of a model to summarize the top ranking pages on the target query, extract the subtopics competitors cover, and flag the gaps the firm could fill. That prep runs on public web content only. No firm data, no matter data, no client data enters the model. The output of the prep is a one-page brief the content lead brings to the interview. The interview itself is a human conversation, and the transcript is either produced by an in-firm transcription tool the compliance officer has approved or captured by the content lead in real time.
The QA gate at stage one is a written editorial brief that both the content lead and the attorney sign off on before any drafting starts. The brief names the target reader, the target query, the substantive scope, the pieces of primary source law that will appear, and the practice-area-specific risk flags the piece will need to clear. Without the gate, stage two starts from a model's guess about what the attorney meant, rather than from the attorney's own record.
Example in operation. A trusts and estates partner sits with the content lead to plan an article on the updated state probate exemption. She names the statute section, walks through two common competitor misreads, and describes a de-identified client scenario. The content lead leaves with a brief that includes the section citation, the misread traps, and a signed instruction to build the piece informationally with no advice to the reader on their own estate plan.
Stage 2. Outline generation
The attorney interview transcript, plus the content lead's own notes, becomes the source material for an outline. This is the first place AI earns its keep on a legal piece. A model can take unstructured attorney speech, cluster it into logical sections, propose an ordering that matches search intent, and surface the FAQs the piece should answer. What the model cannot do is decide which of the attorney's points are load-bearing and which are digressions.
The outline prompt is where the firm's editorial voice starts to shape output. A generic "outline this transcript" prompt produces a generic legal blog. A specific prompt that names the target reader, the target search intent, the informational-not-prescriptive constraint, the FAQ candidates the piece should answer, and the length budget produces a much sharper first pass. The prompt should also instruct the model to leave every citation and statistic as a placeholder tag rather than filling in a value the model does not have.
The QA gate at stage two is the attorney's approve-or-adjust pass on the outline. Two rounds is the norm. The content lead reads the outline against the transcript first, restructures for editorial judgment, and then sends the outline back to the attorney with a short note flagging any decisions the attorney should confirm before drafting. Without the gate, stage three drafts against an outline the attorney has never seen, and the substantive re-work at stage six eats every hour saved by the model.
Example in operation. The model proposes an outline that leads with the exemption's history. The attorney says history is competitor content; the piece should lead with the change and its reader context. The outline is restructured, history moves to the end as background, and the revised outline is signed into stage three.
Stage 3. First-draft generation
Only after the outline is attorney-approved does a model produce a first draft. The prompt to the model is the outline plus the transcript notes, and the prompt explicitly scopes the draft to non-privileged, non-matter-specific claims. No client identifiers, no matter numbers, no settlement figures tied to a specific case. The model is instructed to leave any statistic or citation in a bracketed placeholder that reads [VERIFY: primary source needed], so the fact-check pass in the next stage has an explicit checklist.
The tool selection at stage three matters. An enterprise-tier model with a data-processing agreement and a no-training clause is the baseline. A consumer chatbot is not appropriate at this stage even for a piece that contains no client facts, because the firm has no leverage on how a consumer-tier provider handles the prompt or the output. The content lead runs the prompt in the enterprise tool, captures the raw output verbatim, and files the output in the working folder with a timestamp. The raw output is the artifact stage four verifies against.
The QA gate at stage three is a fast structural read by the content lead. This is not a rewrite. The content lead reads the draft end-to-end, flags any paragraph that slipped from informational into prescriptive, marks any section that drifted from the outline, and confirms every citation and statistic is tagged rather than asserted. If the model produced a citation without a placeholder, that citation is either verified immediately or removed and re-tagged. The rewrite belongs at stage five.
Example in operation. The model produces a draft that follows the outline with a placeholder for the current exemption amount, but confidently asserts a percentage of estates that will fall below threshold without tagging it. The content lead flags, removes, and re-tags that number. The draft moves to stage four with a working list of six placeholders.
Stage 4. Fact and citation verification
Every placeholder tag from stage three gets resolved against a primary source before the draft moves. Case citations are pulled and confirmed. Statute sections are checked against the current code. Regulations are checked against the current CFR or state equivalent. Statistics are traced to a public authority. The verification log is a real artifact, tracked in a shared doc or a spreadsheet, and it stays attached to the piece for the life of the article so future updates can start from a known-good baseline. AI has no role here. This is the stage where the firm's professional judgment sits, and it stays with a human.
The verifier is a paralegal, a research staffer, or the content lead trained to work from primary sources. The tool set is a jurisdiction-verified legal research platform, the state code website, the CFR, and the primary-source websites of any agencies referenced in the piece. A general web search is not verification, and a chatbot's confident answer is not verification. If the verifier cannot find the source, the claim comes out. If the verifier finds a source that contradicts the claim, the claim is rewritten and re-verified. Every entry in the verification log carries a URL to the primary source, a date, and the verifier's initials.
The QA gate at stage four is the signed verification log. No log, no move to stage five. The content lead does not accept an unsigned or partially completed log, because the log is the artifact that lets the firm answer any future question about how a specific claim came to appear in a published piece. This gate is also where a compliance officer or an ethics counsel can spot-check the workflow without having to read every draft.
Example in operation. The verifier resolves six placeholders. Four statute and regulation citations check clean. One Treasury figure traces to a public IRS release. One fabricated statistic on estate distribution cannot be sourced anywhere and the associated claim is cut. The log is signed and the draft moves to stage five.
Stage 5. Voice edit
AI first-draft prose has a signature. Anyone who reads a lot of legal content online can spot it inside two paragraphs. It is fluent, it is safe, and it is generic. That last quality is the one that hurts a firm. If the firm has spent years building an editorial voice that sounds like the partners it wants to be known for, publishing generic-model prose flattens that identity. Stage five is where a human rewrites the draft in the firm's voice. Tightening the sentence lengths, cutting the tricolons, replacing the safe verbs with sharper verbs, and rebuilding any paragraph that reads like it could have appeared on any other firm's blog.
Voice edit is also where the informational-not-prescriptive rule gets its second pass. The stage-three read caught the obvious slips. Stage five catches the subtle ones. A sentence that begins with "you should" is a prescriptive drift, regardless of how gentle the rest of the paragraph is. A sentence that names the reader's specific jurisdiction and instructs them to file within a specific window is legal advice, unless the paragraph is explicitly framed as a general summary of statutory deadlines and closes with a clear "consult counsel" line. The voice editor rewrites for both discipline at once, because a piece cannot be both on-brand and non-compliant.
The QA gate at stage five is a comparative read against the raw stage three draft. The editor keeps both drafts open and reads them in parallel, confirming that no new claim has been introduced during the voice rewrite. If a claim has been introduced, that claim gets a new placeholder and loops back through stage four before the piece moves to attorney review. This gate closes a common failure mode where the voice editor, in tightening a paragraph, invents a phrase or a figure that reads well and is not sourced.
Example in operation. The voice editor rewrites twelve paragraphs, tightens the opening, and rebuilds the section on lifetime gifts. One rewritten sentence reads "The rule sets a ceiling of [amount] per person," pulling the verified figure from the log. No new claim, no new placeholder, and the draft moves to stage six.
Stage 6. Attorney review and compliance sign-off
The rewritten draft goes back to the attorney SME for substantive review, and to whoever owns compliance for a rules pass. Substantive review confirms the piece is accurate, current, and does not slip from information into prescriptive advice. Compliance review confirms every headline, call to action, testimonial, and outcome reference follows the state bar's advertising rules and the firm's own house style. Both reviews leave marks in the doc, both reviewers initial the checklist, and any change either reviewer makes goes back through the fact-check log if it introduces a new claim.
The attorney SME pass is not a skim. The SME reads the piece as a partner would read an associate's draft, marks any language that overstates the law, any paragraph that slipped into prescription, any citation that reads current but is stale, and any framing that would embarrass the firm if quoted back at a bar association meeting. If the SME is the same partner who owns the practice area, the review is quick because the material is familiar. If the SME is a different partner filling in, the review takes longer because the reviewer is also learning the substance.
The compliance pass runs the state-specific checklist that the compliance officer has signed off on. The checklist for a Florida piece is different from the checklist for a New York piece. The pass ends with a signed line naming the compliance reviewer, the checklist version, and the date. The QA gate at stage six is both signatures on the same version of the document. If either signature is missing, the piece does not publish.
Example in operation. The SME marks two places where phrasing implies planning advice and asks for rewrites. The compliance reviewer runs the Florida checklist and asks for a disclosure line on an outcome reference in the introduction. Both changes are made, both reviewers re-sign, and the piece moves to stage seven.
Stage 7. Publish and archive with audit-trail metadata
The piece is scheduled and published, and a full audit trail is archived. The archive includes the attorney transcript, the outline, every draft, the fact-check log with primary-source links, the voice-edit diff, and the initialed compliance checklist. This is not paranoia. It is the paper trail that lets the firm answer any future question about how the piece came to exist, whether the question comes from a client, a bar association, a plaintiff considering a defamation claim, or the next content lead six months from now who needs to update the article.
Metadata is the second half of stage seven, and it is where AI earns another small speed win. The model drafts candidate meta titles, meta descriptions, Open Graph copy, Twitter Card copy, FAQ schema entries, and Article schema properties. The site editor reads and approves each one. Metadata is public copy and it counts as advertising in most jurisdictions, so the compliance checklist runs a fast pass on the meta description before the piece publishes. The FAQ schema pulls from the piece's own FAQ section, not from the model's imagination, because a schema entry that does not match the visible page can be treated as a misrepresentation.
The archive is stored where the compliance officer can find it. A folder in the firm's document management system, named by piece and dated, is the standard. The folder retention aligns with the longest applicable obligation, which is the topic of the metrics and audit trail section below. The editorial calendar is updated with the piece's next scheduled review date, because legal content ages, and a piece that was accurate in July 2026 may need a refresh by July 2027.
Example in operation. On the probate exemption piece, the site editor publishes to the firm's blog, the meta description passes a fast Florida compliance check, the working folder is zipped into the document management system with the piece's slug, and the editorial calendar is updated with a six-month refresh date to catch any statute update.
The never-do list
The seven-stage workflow answers what to do. The never-do list is shorter and, in practice, more important. Every firm should be able to recite it from memory.
- Never paste matter facts, client names, opposing party names, or privileged material into a consumer LLM. If a fact sequence would identify a specific client to someone familiar with the matter, treat it as identifying. Even public court records that name a client require the client's written informed consent before use in marketing content, and even with consent, the material does not enter a public model.
- Never publish an AI-drafted case citation without primary-source verification. This is Mata v. Avianca restated for marketing. Every case name, docket number, statute section, and quoted holding is confirmed against the primary record. If the source cannot be found, the citation comes out.
- Never let AI write calls to action or headlines that could constitute prohibited solicitation under state bar rules. Models will happily produce copy that promises outcomes, offers free consultations in restricted-language jurisdictions, or uses testimonial framing that a state bar treats as misleading. Every headline and CTA runs through a state-specific advertising checklist.
- Never ship AI voice unedited in personal-injury, family-law, or any other emotionally charged practice area. A model does not know that the reader of a family-law blog post is often in the worst month of their life. Generic-model empathy reads as fake, and the firm loses trust in the paragraph that was supposed to build it.
- Never treat an AI output as final unless a human has read it in full and initialed the checklist. "Looks good, ship it" is not a review. Skimming is not a review. The signature is the review.
The compliance officer relationship, and how not to become the bottleneck
The compliance officer is the person the workflow is designed to make effective, not the person the workflow is designed to work around. Every failure I have watched at a firm's content operation has one of two shapes. Either compliance becomes a bottleneck that drives marketing to bypass her, or marketing routes around compliance in the name of speed and produces work that reaches her desk as a complaint rather than as a draft. Both trace to the same root cause: no signed standing operating procedure defining what routes through compliance and what does not.
The remedy is a written SOP that the compliance officer, marketing lead, and managing partner sign. The SOP defines three categories. Category A always routes through compliance before publication: paid-media advertising, testimonials, case studies with client-identifying material, homepage revisions, and anything referencing a specific outcome. Category B follows the seven-stage workflow with attorney SME sign-off and no separate compliance review per piece: evergreen educational content in low-risk jurisdictions, standard blog posts on general topics, and internal-linking updates. Category C needs no attorney review at all and is narrow: typo corrections, formatting, non-substantive metadata.
The SOP also defines escalation. Any category B piece where the SME flags a novel question routes to compliance. Any category A piece the compliance officer escalates goes to ethics counsel or general counsel. Any complaint, bar inquiry, or competitor note triggers a defined incident response. Every escalation is logged, and the compliance officer reviews the log quarterly to spot patterns.
The bottleneck failure mode happens when compliance is asked to read every piece. It is unnecessary because the SOP is the compliance officer's own approved framework. Once the SOP and checklists are in place, the officer's day-to-day is category A review, quarterly spot checks of category B, incident response, and annual SOP refresh, a manageable load even at weekly cadence. The bypass failure mode happens when marketing decides compliance is too slow and publishes without the required sign-off. That is the failure that produces the bar complaint. The remedy is not a faster compliance officer; it is a marketing lead who treats a missing signature as a hard stop, backed by a partner in charge who holds the line under pressure.
Selecting AI tools for a legal marketing workflow
Tool selection is where the abstract compliance discussion becomes a purchase order. The right tool for a legal marketing workflow shares four attributes. A written data-processing agreement with real commitments on data handling. A no-training clause on customer content. US data residency options for firms that need them, and equivalent options for firms in other jurisdictions. An audit trail that the firm can retrieve, not just the vendor. Any tool that fails on those attributes is a consumer tool, and consumer tools do not belong in a workflow that touches even hypothetical client-adjacent content.
Anthropic Claude
Claude is available at consumer, team, enterprise, and API tiers. Enterprise and API tiers commit that customer content is not used to train future models by default and offer configurable data residency and retention. Claude's long context window suits stage-two outline generation across long attorney transcripts, and its refusal behavior around legal advice can be tuned via a system prompt that instructs informational output with placeholder tags for citations.
OpenAI GPT-4 and GPT-4o class models
OpenAI offers the GPT-4 family at consumer ChatGPT, Team, Enterprise, and API tiers. Enterprise and API deployments carry no-training-on-inputs commitments and configurable retention. The consumer ChatGPT tier is not appropriate for firm work regardless of individual account settings, because the firm has no leverage on downstream product changes. GPT-4 is strong at first-draft prose and at reformatting approved copy for schema and metadata.
Google Gemini Enterprise
Gemini Enterprise, delivered through Google Workspace and Vertex AI, includes contractual protections around customer data. For a firm already on Google Workspace, Gemini is the low-friction option. Same workflow discipline applies: enterprise tier only, informational system prompt, placeholder tags on every claim, and human verification before publication.
Microsoft Copilot for Microsoft 365
Copilot embeds into Word, Outlook, and the rest of M365 with commitments that customer data is not used to train foundation models. For a firm on Microsoft 365, Copilot's draft-in-Word workflow removes friction from stage three because the draft, transcript, and outline can live in the same document. The trade-off is that convenience makes it easier to skip the placeholder rule, so the standing prompt template needs to be explicit about it.
Harvey and Thomson Reuters CoCounsel
Harvey and CoCounsel are legal-specialty platforms designed for legal work rather than for content. They are relevant to a marketing workflow because they can produce sourced summaries of case law and statutes that make reliable inputs to the stage-one attorney interview. If the firm has adopted either platform, marketing can request a sourced brief on the topic the piece will cover. The output is prep material for the human interview, not a first draft.
Self-hosted, cost, and audit considerations
Firms with strong data-residency requirements can self-host open-weight models. This is more operational overhead than most marketing teams want to carry and typically requires a firm-side technical lead. Enterprise tiers cost more per seat than consumer tiers; the math favors enterprise even before compliance, because the risk premium of a consumer-tier leak is orders of magnitude larger than the annual seat cost. On audit, the firm needs the ability to retrieve at least timestamps and prompt-and-output pairs for the accounts marketing uses. If the vendor does not offer that, the workflow captures it locally at the moment of each draft.
Disclosure and attribution templates for four surfaces
A single disclosure line does not cover every surface where the firm publishes AI-assisted content. A blog post, a social post, a case study, and a video each carry different reader expectations and different regulatory constraints. Below are four templates the firm can adapt to house style. Each one names the scope of AI use, names the humans responsible, and closes with the boundary conditions.
Template 1. Blog post editorial note
Editorial note. This article was drafted with assistance from a
generative AI model used at the outline and first-draft stages. All
citations were verified against primary sources by [Editor Name]. The
substance was reviewed and approved by [Attorney Name], licensed in
[State] Bar No. [Number]. This article contains no client facts and
is general information, not legal advice for any specific matter.
For advice on your situation, contact the firm.Signals process discipline, names the responsible attorney with a verifiable bar number, and closes with the general-information disclaimer that rebuts the reader-relied-on-it argument.
Template 2. Social post disclosure line
Written with AI-assisted drafting, reviewed by an attorney at
[Firm Name]. General information only. Not legal advice.Length-constrained. Keeps the two load-bearing pieces: AI participated, an attorney reviewed. If the piece is high-visibility, use the blog template on the linked landing page and keep the social line as a pointer.
Template 3. Case study editorial note
Editorial note. This case study was prepared with assistance from a
generative AI model at the outline stage. All matter facts were
sourced from public court records and from the client's written
publication release dated [Date]. No privileged or confidential
information appears in this piece. The substance was reviewed by
[Attorney Name], licensed in [State], and the client approved the
final language on [Date].Case studies carry the highest confidentiality risk because context can identify the client. The template names the release, the review, and the client's approval. If any of the three is missing, the piece does not publish.
Template 4. Video description note
Script drafted with AI assistance, verified against primary sources,
and reviewed by [Attorney Name], licensed in [State] Bar No. [Number].
Filmed [Date]. General information only, not legal advice.Video is produced faster than written content and more often skips review. The template signals that the same discipline applied. The date helps future viewers see the video was accurate at publication, which matters for evergreen education that ages.
The debate on whether to disclose at all
Not everyone in legal marketing agrees that AI disclosure helps. The counter-argument is that disclosure invites scrutiny of a workflow not strictly required to be disclosed under most current bar rules, and that competitors who do not disclose enjoy a perception advantage. There is some truth to it. My view is that the trend is toward more disclosure, not less. Bar authorities are examining AI use, journalists write about it constantly, and firms that establish a clean disclosure practice early look better than firms pushed to adopt one after an incident. The blog-post template above is short, honest, and reads as process discipline rather than apology. That is the version I would ship.
Scaling the workflow inside an agency serving many law firms
The seven-stage workflow scales to an agency serving dozens of firms, but the scaling has to be deliberate. Three artifacts do most of the work.
First, a per-firm SOP. Every firm on the roster gets a short document naming the attorney SME per practice area, the compliance reviewer, the editorial voice notes, disclosure preferences, and any client-specific never-do items. New agency staff read the SOP before touching a keyboard. This is the difference between an agency that produces uniform slop and one where each firm's content sounds like that firm.
Second, a per-jurisdiction fact-check appendix. State statutes, ad rules, and evolving case law change often enough that the agency maintains a shared research library organized by state, with dated entries and primary-source links. A Georgia workers' compensation piece starts from the current jurisdiction file; the fact-check pass either confirms entries or flags refresh candidates.
Third, per-practice-area review depth. Not every piece needs the same attorney touch. A "what is estate planning" evergreen in a low-risk state differs from "how to challenge an ICE detainer in the Third Circuit." The editorial calendar tags every piece from light to heavy at intake, so no one discovers on publication day that a piece needed more review than the schedule allowed.
Where marketing meets client-facing AI adoption at the firm
The scope of this article is marketing. Where AI touches actual client matters is out of scope, because that is a legal-work question that runs through the firm's practice leadership, general counsel, and ethics counsel rather than through the content team. There is a boundary case the marketing team needs to understand. When the firm adopts AI tools for legal work, marketing wants to publicize the adoption, and the language of that publication is regulated in ways the marketing team may not intuit.
Four issues sit under any public statement about the firm's own AI use. Confidentiality under Rule 1.6, because a statement that references a specific matter needs the client's consent before it publishes. Competence and truthful communication under Rule 7.1, because a claim that AI produces faster or better outcomes has to be substantiable. Fee reasonableness under the framework Opinion 512 describes, because efficiency claims interact with how the firm charges. Supervision under Rules 5.1 and 5.3, because any external statement about AI has to be paired with a statement that attorneys remain responsible for the work.
The safer framing describes practice, not outcomes, and never touches a specific matter. "The firm uses enterprise AI tools with human attorney oversight for research, drafting, and internal knowledge management" is a description a bar counsel can read without concern. "AI helps us win faster" is not. The workflow answer is that any marketing content describing the firm's AI adoption in legal work routes through the general counsel or ethics counsel before publication, even if it would otherwise be a category B piece under the standing SOP.
Metrics, audit trail, and retention policy alignment
A workflow that cannot be reconstructed six months later is not a workflow. It is a story people tell about how the piece came to exist, and stories drift. The audit trail is what lets the firm answer any future question about a specific piece, in a specific way, with specific artifacts. The retention policy is what lets the audit trail survive long enough to matter. Both are more useful and less expensive than they sound.
The metadata schema
Every published piece carries a small structured record the firm can retrieve without a full-text search. The record includes title, published URL, publication date, last-modified date, practice areas, target jurisdictions, responsible attorney with bar number, reviewing compliance officer, editor, AI tools that participated and at which stages, the version number of the signed compliance checklist, and the archive location. A spreadsheet works for a small firm. A structured content register is better at agency scale.
Version control and sign-off logs
The working folder for each piece includes every artifact produced. Attorney transcript, pre-brief prep, editorial brief, outline, raw first draft with model name and prompt template, placeholder list, verification log with primary-source links, voice-edit diff, attorney SME markup, compliance markup, final published copy, metadata, and disclosure note. No file is deleted or overwritten during the life of the piece. Every signature on the workflow is captured in a log at the front of the folder with name, date, and version reference. That log is the first artifact a compliance officer or auditor reads.
Retention alignment with bar rules
State bar retention rules for lawyer advertising typically require two to six years depending on jurisdiction. New York requires at least three years for most formats and longer for some. California requires two years for most advertising. Florida imposes retention obligations tied to its filing regime. Texas requires four years for many materials. The safe default is six to seven years from the last-modified date of the piece, which covers every state's minimum and most malpractice statute-of-limitations windows. Retention is per-piece, so a 2029 update restarts the clock rather than letting the piece age out on its 2026 publication date.
Velocity, cost, quality, and turnover as the four metrics
The workflow becomes budgetable when the firm tracks four numbers. Velocity is pieces per month. Cost per piece is total loaded cost divided by pieces shipped. Compliance issue rate is corrections, complaints, or inquiries divided by total pieces. Turnover in the content team is the leading indicator that the workflow is actually working, because ad-hoc AI adoption produces an early productivity spike followed by a rework burden followed by exhaustion. Teams on the seven-stage workflow post lower early gains but flat rework burden and lower churn. Managing partners should look at all four numbers quarterly.
The economic argument for ethical AI workflow discipline
The pitch for the seven-stage workflow inside a firm is not "we should be ethical." Every partner already agrees the firm should be ethical. The pitch is that ethical discipline is the operational choice that produces the strongest business outcome. Content velocity, cost, quality, defensibility, and staff retention all improve when the discipline is in place, and they all degrade when it is not. The business argument is the one that closes the sale with the managing partner.
The velocity math is the easiest. A content lead who used to publish two to three long-form pieces per month can move to six to nine when AI carries the outline and first-draft load, provided the workflow is disciplined. The multiplier is roughly two to three. Firms that skip the discipline see one to one-point-five, because rework from unverified claims and prescriptive drift eats the drafting gain. Firms that refuse AI stay at pre-2023 velocity while competitors iterate, which is not a stable posture.
The cost math depends on how the firm counts editor and attorney time. At scale, cost per finished long-form piece runs one thousand five hundred to three thousand dollars in traditional human-heavy production. With disciplined AI assistance the number can drop to seven hundred to twelve hundred, holding quality constant. Savings come from lower editor hours, not from cutting attorney or compliance hours; the workflow deliberately preserves the review load. Firms that trim attorney or compliance hours as part of the AI transition are the ones that end up paying for a defense.
Quality shows up in the compliance issue rate. Firms running the seven-stage workflow post near-zero post-publication corrections over a rolling twelve-month window. Firms on ad-hoc AI post multiple corrections per quarter and occasional bar inquiries that each consume ten to forty hours of partner and general counsel time. The value of a workflow that prevents even one bar inquiry per year usually exceeds the workflow's total operational overhead.
Defensibility is harder to price but easier to describe. A firm with an archived audit trail answers any inquiry in hours. A firm without one either invents an answer or admits it does not know how the piece was produced. Neither is what a managing partner wants to say to a bar counsel. Staff retention is the last number and, in a competitive market for legal marketers, the one that matters over five years. A team on a written workflow keeps its content leads. An ad-hoc team churns, and the replacement cost is real in both dollars and lost knowledge.
Ethical discipline is a speed advantage, not overhead
The reason to build this workflow is not compliance for compliance's sake. A firm running it ships more legal content per month, at higher trust, than a firm that either refuses AI or uses it without discipline. Refusing AI leaves the firm at pre-2023 speed while competitors iterate. Using AI without discipline sooner or later produces a fabricated citation or a prescriptive paragraph that reads as advice to a non-client, and every hour saved on drafting evaporates the moment one of those pieces surfaces.
The firm that installs the seven-stage workflow gets three things at once. Faster production, because AI carries the drafting and structural load. Tighter compliance, because the review stages are explicit and initialed. A defensible archive, because every piece leaves a paper trail. Three things that in the pre-AI era took three separate initiatives, delivered by a workflow that fits on a single page and can be taught to a new hire in an afternoon.
Legal marketing has always rewarded firms with the discipline to say things carefully. Generative AI does not change that. It raises the ceiling on how much careful content the firm can produce, provided the workflow is real and written down. Everything else is theater.
Frequently asked questions
Does ABA Formal Opinion 512 prohibit lawyers from using generative AI?
No. Opinion 512, issued in July 2024, applies existing duties (competence, confidentiality, communication, supervision, reasonable fees) to AI use. AI use is permitted; the lawyer remains fully responsible for accuracy, client confidentiality, and independent professional judgment. It is a use-with-discipline framework, not a ban.
Can a marketing agency paste client matter facts into ChatGPT if the firm asks for a case study?
No. Matter facts, client names, opposing party names, docket numbers, settlement figures under seal, and privileged material never enter a consumer LLM. Even public court records that identify a specific client require the client's written informed consent before use in marketing, and even then the material does not touch a public AI tool at any stage.
What does Mata v. Avianca teach law-firm marketers?
The 2023 case in which two New York attorneys were sanctioned after filing a brief citing fabricated ChatGPT opinions. The marketing lesson is not that AI cannot draft; it is that every citation a model produces must be verified against the primary source before publication. Case names, docket numbers, statute sections, and quoted holdings are the most common hallucination points.
How do state bar advertising rules affect AI-drafted content?
Every state bar imposes rules on lawyer advertising, and many restrict solicitation language, outcome guarantees, and testimonials. AI does not know these rules and will confidently produce copy that violates them. Every headline, call to action, and testimonial passes through the firm's state-specific advertising checklist before publication. That review is a mandatory stage, not an optional pass.
Does disclosure of AI use in a blog post help or hurt a law firm?
A short, honest editorial note that describes how AI participated and confirms an attorney reviewed the final copy tends to help. It signals process discipline. The disclosure should name the human editor, the reviewing attorney, and confirm the piece contains no client facts. A vague "AI-assisted" tag with no context looks like a hedge and can hurt trust.
What is the single biggest risk of AI-drafted legal content?
Unauthorized practice of law disguised as helpful blog copy. When a model drafts "what to do after a car accident in Ohio" and the draft slips into imperative, jurisdiction-specific advice without an attorney edit, the firm has published what reads like legal advice to a non-client. That crosses Rule 5.5 and creates duty questions the firm did not intend to accept. Every draft is edited to inform, not advise.
Which ABA Model Rules apply to AI-assisted marketing content?
At least six. Rule 1.1 competence and the technology-competence comment. Rule 1.6 confidentiality of information relating to a representation. Rule 3.3 candor to a tribunal when the piece quotes filings or verdicts. Rules 5.1 and 5.3 supervision of lawyers and non-lawyers, which extends to agency staff and the AI tools they use. Rule 5.5 unauthorized practice, the rule most quietly violated when AI drafts prescriptive advice. Rules 7.1 through 7.5 advertising and solicitation, reaching every headline, testimonial, and outcome reference.
How do state bar advertising rules differ across Florida, Texas, California, and New York?
Florida runs an aggressive regime with formal filing for many ad forms, strict testimonial rules, and detailed content restrictions. Texas requires specific disclosures and administers filings via the Advertising Review Committee, including a responsible-lawyer identification rule. California prohibits guarantees or warranties and regulates statements that create unjustified expectations. New York permits broader advertising but imposes strict retention obligations. A multi-state firm needs state-specific checklists per piece, not a single national pass.
What enterprise AI tools are appropriate for a law firm's marketing workflow?
Enterprise-tier tools with data-processing agreements, no-training clauses on customer content, and US data residency options. Anthropic Claude via API or Enterprise, OpenAI GPT-4 via API or Enterprise, Google Gemini Enterprise, and Microsoft Copilot for M365 all meet the threshold. Specialty legal platforms like Harvey and Thomson Reuters CoCounsel offer tighter audit trails and jurisdiction-aware behavior. Consumer chatbot tiers do not meet the standard because content may be used for training and audit logging is limited.
How long should a firm retain the audit trail for AI-assisted marketing content?
Match the retention period to the longest applicable obligation. Most state bars require at least two years for lawyer advertising, several require three, four, or six. Malpractice and defamation windows can extend it. A safe default is six to seven years from the last-modified date of the piece, retained together with drafts, verification log, and both sign-offs. Store the archive where the compliance officer and future content leads can find it without a full-text search of Slack.
Should a law firm disclose that it uses AI in its own legal work when marketing publicizes an AI-forward practice?
Yes, but only in language the general counsel or ethics counsel has signed off on. Statements about AI in legal work touch confidentiality, fee reasonableness, competence, and supervision. Marketing should confirm what the firm does, how it protects client data, and how attorneys remain responsible. A vague public claim that the firm uses AI without those specifics is worse than saying nothing.
How does the workflow prevent an AI-drafted personal injury blog post from reading as legal advice?
The voice edit stage rewrites imperative sentences into informational ones. Attorney review reads specifically for the slip from "here is how the process generally works" into "here is what you should do." If the draft names a specific jurisdiction, a specific fact pattern, or a specific action a reader should take next, either that language is removed or the piece carries a prominent general-information disclaimer with a call to consult counsel.
What is a realistic content velocity gain from AI when the ethical workflow is in place?
Two to three times, sustained. Content leads who published two to three long-form pieces per month can move to six to nine at similar quality, because AI absorbs outline and first-draft load while human hours concentrate on interview, verification, voice, and review. Without discipline the gain evaporates because rework overwhelms the drafting time saved. Discipline is the multiplier, not the tool.
Does the marketing team need the compliance officer to approve every piece?
No. The compliance officer signs the SOP, and the SOP defines what routes through her. Category A pieces (paid media in strict-solicitation states, case studies naming a client) always route through compliance. Category B pieces (evergreen educational articles in low-risk jurisdictions) follow the SOP with attorney SME sign-off alone. The officer reviews the SOP annually and spot-checks category B quarterly, which prevents both bottlenecks and drift.
References and further reading
Every citation below was verified against the primary source before this article published, consistent with the stage four discipline the article describes.
- American Bar Association. Formal Opinion 512. Generative Artificial Intelligence Tools. Standing Committee on Ethics and Professional Responsibility. July 29, 2024.
- American Bar Association. Model Rules of Professional Conduct. Current edition. Full text and comments available through the ABA Center for Professional Responsibility.
- Mata v. Avianca, Inc., 22-cv-1461 (PKC), 2023 WL 4114965 (S.D.N.Y. June 22, 2023). Opinion and order on sanctions by Judge P. Kevin Castel.
- The Florida Bar. Rules Regulating The Florida Bar, Chapter 4, Rules of Professional Conduct, Subchapter 4-7 (Information About Legal Services).
- State Bar of Texas. Texas Disciplinary Rules of Professional Conduct, Part VII (Information About Legal Services), and Texas Advertising Review Committee guidance.
- State Bar of California. California Rules of Professional Conduct, Rule 7.1 and related sections on communications concerning a lawyer's services.
- New York State Unified Court System. New York Rules of Professional Conduct, Rule 7.1 (Advertising) and related sections on retention of advertisements.
- American Bar Association. ABA Techreport and ongoing coverage from the ABA Journal on generative AI adoption in the legal profession.
- Stanford Institute for Human-Centered Artificial Intelligence and the RegLab. Published studies on hallucination rates in general-purpose and legal-specialty language models. 2024 to 2025.
- Thomson Reuters Institute. Future of Professionals and related reports on generative AI in law firms.
Cross-references on this site: AEO vs GEO vs SEO on how AI-drafted content is discovered by answer engines; the nineteen ranking surfaces; the case study on the law firm AEO retrofit; and the practice-area case studies for personal injury, family law, immigration, and social security disability firms.
Fifteen years in marketing, ten as CMO and Creative Director. Operator of AI-augmented content workflows across regulated verticals including legal, healthcare, and financial services. Write to me at fredericksona.com or connect on LinkedIn.