In July 2026, Originality.AI ran 5,000 LinkedIn posts through its detector and flagged 81.2% as “Likely AI”. LinkedIn’s answer was not an algorithm change but a button: a “Seems like AI slop” feedback option.
That is the market for LinkedIn posts for lawyers with AI in September 2026: four in five posts read as machine-written, readers can flag them, and everyone has the same tools. The question is how to use ChatGPT or Claude without sounding like the other 81%.
The good news is in Passle’s data: while The Economist found search visits to reference sites down 15%, direct traffic to legal thought leadership on Passle’s platform rose 59%. Tom Elgar: “AI thrives on yesterday’s internet. Lawyers who are winning the content game write about tomorrow’s legal risk.”
The 81% problem: what an AI-likely post looks like
Nobody can reliably detect AI text. OpenAI withdrew its own classifier on 20 July 2023 after it caught 26% of AI-written text and falsely flagged 9% of human writing. The penalty is social: readers notice pattern, not provenance, and stop reading. The tells: an opening that describes the state of the world before saying anything; three abstractions with emojis; a claim with no example or number; the hedge that dissolves the opinion; a closing “Thoughts?” addressed to nobody. Lauren Forbes’s JD Supra guide for 2026 is blunt: “generic AI-generated content can damage credibility”.
Build a voice file from your own writing
The fix is to give the model you to imitate. Collect eight to ten pieces you wrote before you used AI and ask for a description of your voice with evidence, not adjectives. Anthropic’s own prompting guidance says “heavy-handed role prompting is often unnecessary” with current models and that being explicit about what you want works better; ten samples of your own writing are as explicit as it gets.
Here are ten pieces I wrote myself, before I used AI: <samples>[paste]</samples>.
Write a "voice file" I can reuse as standing instructions: (1) sentence length and rhythm, with two quoted examples; (2) how I open and close a post; (3) words I use often and words I never use; (4) how I handle opinions, hedging and humour; (5) how much legal detail I give and how I explain jargon; (6) five things a ghostwriter would get wrong about me. Quote the samples for every claim; do not describe a style you cannot show in them.Save it as the standing instructions of a Claude Project or custom GPT (set-up guide).
The repurposing pipeline: one alert, five formats
The workflow with evidence is not “write a post” but “cut something you already wrote for a client five ways”: upload a transcript or article, ask for the five most shareable insights, generate drafts, then attorney review. One immigration firm (per rankings.io) turned a 45-minute H-1B webinar into LinkedIn and Facebook posts, a newsletter summary, a compliance checklist and an email digest in under an hour instead of two to three days. Start from a verified client alert, so the model has nothing to invent; writing it is covered in AI for law firm business development.
Apply my voice file <voice_file>[paste]</voice_file> to the client alert below and write a LinkedIn post of 150 to 220 words for [in-house counsel at mid-sized manufacturers]. Open with the practical consequence, not the case name; include one concrete example; no hashtags in the body, no bullet lists, no emojis, no "I'm thrilled"; end with one genuine question an in-house lawyer might actually answer. Do not add any fact, figure or authority that is not in the alert. Then give me two alternative first lines and a one-sentence first-reply comment.
Client alert:
[paste]Then the carousel: ask the same Project for a ten-card outline (title, eight one-idea cards of at most 25 words, a closing action), legal statements kept as in the post and [NEEDS SOURCE] where a card needs a fact the post lacks. Gamma builds the deck from pasted text and exports to PDF or PowerPoint (there is a free plan); Canva does the same.
What performs: carousels, specifics, opinions
Buffer’s 2026 analysis of 52 million posts:
| Format | Median engagement rate on LinkedIn | What it means for a lawyer |
|---|---|---|
| Carousel (document post) | 21.77% | A ten-card checklist or “five things this judgement changes” beats every other format by roughly three to one |
| Video | 7.35% | A two-minute explanation of one point; not a post read to camera |
| Image post | 6.52% | Shelton reports posts with images get 40-150% more clicks than text alone |
Source: Buffer, State of Social Media Engagement 2026; Shelton in Attorney at Law Magazine.
The same Buffer data, as summarised by rankings.io, puts the sweet spot at about three posts a week; treat it as a ceiling. For episodic-need practices such as family, M&A and probate, the aim is what strategist Tony Albrecht calls “demand generation”: being the first call when the need arises.
Harrington’s principles for lawyer content
Jay Harrington’s principles are the closest thing to a style guide here.
“Clients don’t want to be impressed. They want to be informed in a way that helps them make better decisions.” — Jay Harrington, Latitude Legal, LinkedIn
“Write for someone, not everyone.” “Clarity beats cleverness.” And: “The best content doesn’t make the lawyer look smart. It makes the reader feel smart.”
His cadence turns posts into work: weekly bullets on industry developments with your perspective; a monthly compilation; then an email to key contacts with a personal note and, selectively, a lunch invitation. It makes you “a commentator and curator of valuable information” rather than the sender of “the dreaded ‘just checking in’ email”, and it is exactly the drafting a model with your voice file does well.
Shelton’s $150,000 post
The tool solves the wrong bottleneck. Fin Wycherley on why most lawyers have not posted in months: “Not because they don’t have insight. But because they don’t have time, clarity, or a system.” AI buys the time; the system and the two years are yours.
Prompts: draft, then critique, then add the detail only you know
Separate drafting from critique: Justia’s prompting guide found that asking for draft, critique and rewrite in one message makes the model “rush the initial draft just to get to the critique” (see legal prompting mistakes). Draft, then open a new message and attack the draft.
Act as a sceptical in-house lawyer who scrolls past most posts. List, as bullets only: every sentence in the draft below that could appear unchanged on any law firm's website; every generic claim; every place it explains what my readers already know; every hedge that makes the opinion disappear; and the one sentence only I could have written. Do not rewrite. Then name the single detail from my own practice that, if added, would make the post impossible to mistake for AI output; ask me for it if you cannot infer it.
Draft:
[paste]The last question is the whole method. Never post the first draft; edit in one detail only you could add, an anonymised wrinkle from a matter or the thing every practitioner knows but the judgement does not say.
Compliance: Rule 7.1 does not care who drafted it
A model given your CV will improve it if you let it: “leading” becomes “the leading”, a settlement becomes a “win”. So:
List every factual claim in the text below in a table: Claim | Type (credential, award, result, testimonial, statistic, comparison) | Source I can point to (or NONE) | Risk under lawyer-advertising rules (High/Medium/Low). Flag any superlative, any case result stated without the facts that produced it, any implied guarantee and any statement about what my AI tools can do. Never invent a source. Finish with the three lines I should delete or soften before this goes live.
Text:
[paste]“AI-assisted and attorney-reviewed” is a reasonable label. Whether you must tell clients is a separate question: do lawyers have to disclose AI use.
A weekly 45-minute routine
- Monday, 10 minutes. Paste last week’s client alert or transcript into the voice-file Project and ask for five shareable insights (prompt in the prompt library).
- Tuesday, 15 minutes. Draft one post from the strongest insight; de-slop pass; add your detail; claims audit if it mentions a result.
- Thursday, 10 minutes. Turn the same insight into a carousel in Gamma or Canva for the following week.
- Friday, 10 minutes. Add the week’s bullet to the monthly compilation; on the last Friday, send it to ten contacts with one personal line each.
Before a conference, the same Project does meeting preparation: one trial lawyer reportedly prepared for twelve GC meetings in two hours instead of twenty to thirty with ChatGPT, Lexis+ and Introhive and “secured new opportunities from 11 of the 12” (a single account published by the Legal Sales and Service Organization; treat it as an anecdote and use public sources only). Then check what the assistants say when someone asks who to hire: getting your firm recommended by ChatGPT and Perplexity.
Where to go next: the business development hub holds the pillar guide and the RFP and GEO pages; ChatGPT prompts for lawyers covers the basics the voice file relies on. The business development, LinkedIn and outreach module of AI Lab for Lawyers runs this repurposing workflow live, on your own material, in a browser.
Frequently asked questions
Can lawyers use ChatGPT for LinkedIn posts?
Yes, and most already do: 8am's 2026 report found 69% of legal professionals use GenAI for work, with drafting correspondence and general research the top uses. The constraints are professional, not technical. Every factual claim in a post is your statement under lawyer-advertising rules, client matters must be anonymised beyond names, and generic AI prose damages credibility. Use the model for structure and first drafts from material you have already verified, then edit in your own voice and detail.
How do I make AI LinkedIn posts sound like me?
Give the model your own writing to imitate rather than adjectives to aim at. Collect eight to ten posts, alerts or emails you wrote unaided, ask the model to describe your rhythm, openings, vocabulary and habits with quoted evidence, and save that voice file as the standing instructions of a Claude Project or custom GPT. Draft from it, run a separate critique pass, and add one detail from your own practice that no model could know.
What LinkedIn content works best for lawyers?
Buffer's analysis of 52 million posts puts carousels at a 21.77% median engagement rate on LinkedIn, about three times video (7.35%) and images (6.52%). Beyond format, Jay Harrington's principle holds: content that helps a specific reader make a better decision beats content that makes the lawyer look clever. Practical consequences of a change, one concrete example and a genuine opinion outperform summaries of the news.
Are AI-written posts against bar advertising rules?
Not as such, but the rules apply to them in full. ABA Model Rule 7.1 treats an AI-drafted post as the lawyer's own communication, so a fabricated award, an inflated case result or an invented testimonial is a violation regardless of who typed it. Florida Opinion 24-1 also bars claiming your AI is superior unless the claim is objectively verifiable. Run a claims audit on any post or profile text that mentions credentials, outcomes or comparisons before it goes live.
How often should a lawyer post on LinkedIn?
Buffer's analysis of 52 million posts, as summarised by rankings.io, puts the sweet spot at about three posts a week, which few practising lawyers sustain. Frederick Shelton, whose one post produced $150,000 of work, recommends at least fortnightly with an 80/20 split of business to personal content, and warns that his first deal came two years in. One good post a week, plus a monthly compilation emailed to key contacts, is a realistic cadence for a busy partner.