Why can’t law firms advertise on ChatGPT: and what should they do instead?
By The FirmCited team, BizLegal AI Last updated
What exactly happened in June 2026?
When OpenAI launched advertising inside ChatGPT, it drew a hard line around regulated categories: and legal services landed on the excluded list. There is no waiting list, no premium tier, no agency workaround. A personal injury firm with a seven-figure ad budget has exactly the same ChatGPT placement options as a two-lawyer startup: none.
This matters because ChatGPT is not a fringe channel. Legal consumers now start attorney research in AI assistants, asking “do I have a case if…” before they ever type a firm name into Google. When the assistant answers, it frequently names two or three firms. Those names are the new first page.
Why is this different from every previous marketing shift?
Because there is no paid lane. Every prior channel, Yellow Pages, Google, Facebook, TV, let a firm with budget buy its way to visibility on day one. The AI answer channel inverts that: money is useless, and the winners are decided by machine legibility. A firm becomes citable the same way a source becomes credible to a careful researcher: clear identity, verifiable facts, direct answers.
The strategic consequence: this is a land-grab phase. Citations compound: engines that cite a firm keep citing it as its content deepens: and the firms building citability now will be structurally hard to displace once the channel matures. The ad ban froze the shortcut; the head start is what is for sale.
| Channel | Access | Asset type | Competition |
|---|---|---|---|
| Google Ads | Open | Rented: stops with spend | Saturated; USD 100 to 500 CPCs in PI |
| ChatGPT ads | Closed to legal since June 2026 | Not applicable | No inventory at any price |
| AI citations (all engines) | Open: organic only | Owned: compounds | Early, still open to any firm |
| Directories / referrals | Open | Mixed | Mature, incremental |
How does ChatGPT decide which firms to recommend?
Three families of signal. First, entity clarity: the model must resolve your firm as one unambiguous thing: same name, address, and practice areas everywhere it looks, from your site to Google Business Profile to Avvo to the state bar listing. Conflicting data does not weaken you; it deletes you.
Second, source trust: crawlable pages (GPTBot allowed in robots.txt, content in HTML rather than JavaScript), schema markup that mirrors visible text, mentions in directories and press the model already trusts. Third, quotability: pages that answer the user’s actual question in the first hundred words, with specifics, statutes, timelines, fee structures, the model can lift verbatim. Brochureware fails all three tests at once.
What is the playbook, step by step?
Six steps, in strict order: measurement first, plumbing before content. Skipping ahead to blog posts while GPTBot is blocked or your firm has three name variants across directories wastes every writing hour.
| When | Step | What to do | Why it matters |
|---|---|---|---|
| Week 1 | Baseline audit | Run 25 money queries × 4 engines; record every cited firm | You know your real share-of-voice |
| Weeks 1-2 | Entity cleanup | One canonical name/address/practice set across site, GBP, Avvo, Justia, state bar listing | Models resolve your firm as one entity |
| Weeks 2-3 | Technical access | Allow GPTBot, ClaudeBot, PerplexityBot in robots.txt; ship llms.txt and a clean sitemap | AI crawlers can actually read the site |
| Weeks 2-4 | Schema layer | LegalService, Attorney, FAQPage, Article JSON-LD sitewide | Machine-verifiable identity and answers |
| Weeks 3-8 | Answers hub | One page per real client question; 40-60 word direct answer up top; one table each | Quotable content for every money query |
| Monthly | Measurement | Re-run the same prompt panel; track citations per engine | You see real movement, not a hunch |
Steps two through four are one-time infrastructure; step five is where compounding starts. The answers hub deserves emphasis: one page per real client question, “What is my car accident case worth?”, “How much does a divorce lawyer cost in Tampa?”: each opening with a complete 40-60 word answer, each carrying FAQPage schema, a named author, and a visible update date. This is the exact format engines quote, which is why this site is built the same way.
What results should a firm expect, and when?
Honestly: first citations on long-tail questions within one to two months, movement on competitive money queries in three to six, volatility throughout as models update. Anyone promising a specific citation by a specific date is selling something they do not control: we wrote a whole page on why. What is controllable: the deliverables, the timeline of the work itself, and honest monthly measurement against a fixed prompt panel.
The ROI frame is simple. An average consumer case carries USD 15,000 to 50,000 in fees. If this channel produces one additional retained case a year, it outearns most firms’ entire content budget, and unlike ad spend, the asset keeps working after you stop paying for it.
Step one is measurement. We run it for USD 490.
The AI Visibility Audit executes this playbook’s baseline: 25 prompts across up to 4 engines, competitor share-of-voice, and a prioritized fix list: in under 48 hours, credited toward any install.