Last updated: July 18, 2026 · Checked against OpenAI's ad policy dated June 4, 2026

Update, July 18, 2026: A StackAdapt representative has told us that lawyer ads can be implemented through their platform. StackAdapt is a confirmed OpenAI ad partner, so account access is real, but partners don't control ad approval; OpenAI's system does, and its content policy still lists legal services as disallowed. We're verifying whether StackAdapt has an actual path to getting personal injury creative approved, or whether this is access to the buying tool without a guarantee ads clear review.

What OpenAI's ad policy says about personal injury

OpenAI's ad policy, updated June 4, 2026, says this: ads for legal advice, representation, or legal services offered to individuals or businesses are not permitted. It then names the examples. Immigration. Personal injury. Legal claims. Document preparation.

What we found running inside ChatGPT

In July 2026, our team observed personal injury ads running inside ChatGPT's free tier. Not legal podcasts. Not LSAT prep. Contingency-fee injury firms offering free consultations and call-now CTAs, from firms in at least two states. We are not publishing the screenshots, because naming the firms is not the point and they have done nothing wrong. Here is what the format looks like.

These offer exactly what the policy prohibits: a consultation, representation, a phone call. Nothing about them resembles a legal podcast or an LSAT prep course, which are the only legal ads the policy's carve-out clearly permits.

And this is not a quiet anomaly. Multiple legal marketing agencies state on their own websites that they are already running live campaigns. One says it stood up an advertiser account and launched for personal injury clients the same night CPC bidding went live. Another claims the first lawyer ad ever placed inside ChatGPT. Several now sell ChatGPT ads to law firms with the flat assertion that legal services are an eligible category, which is the direct opposite of what the Disallowed Ads section says. These claims are easy enough to find. We are not linking to them, for reasons that will be obvious by the end of this article.

So what is going on, and what should your firm actually do about it?

Why these ads might be running

One possible explanation is sitting in the same policy document that appears to forbid them.

In Section 2, the Ad Content Policy, OpenAI says it may approve ads from approved advertisers within the financial services, healthcare and medicine, and legal services categories, and that these categories are rolling out gradually with approvals reviewed manually on a case-by-case basis.

So one section of the policy says legal services ads are not permitted. Another says legal services advertisers may be approved case by case. Both are live on the same page. Both carry the same June 4 date.

This is not us reading tea leaves. Marketing Brew compared the live policy page against an April 29 version saved on the Wayback Machine and found the case-by-case language was added, not inherited. An OpenAI spokesperson confirmed to them that the placement policy had been updated to be more precise about regulated categories, while reiterating that the company is not widely including ads in legal conversations.

There are three readings, and the difference matters a great deal to your firm.

Reading one: these firms were approved. Manual review, case-by-case, exactly as Section 2 describes. If so, the Disallowed language is stale or is meant to describe the default rather than the ceiling, and the practical path is to apply and get through review.

Reading two: these slipped through. OpenAI states plainly that most ad review is conducted by scaled AI systems with human oversight. Scaled systems have gaps. If these ads are gaps rather than grants, the firms running them are exposed.

The education lane, and what it actually costs

Reading three: they went in through the education door. This is the one nobody talks about, and it is probably the most useful thing in this article.

Remember the carve-out. Ads for general legal education or media may be allowed where no legal services are offered. Now look at what the first publicly claimed lawyer ad inside ChatGPT actually was. In May 2026, a legal marketing agency announced it had placed one, and the campaign promoted educational content about Pennsylvania personal injury law. Not a free case evaluation. Educational content. The same agency lists educational-content-driven advertising for plaintiff-side firms as a core capability.

Two caveats before you get excited, because nobody else will give you these. That placement ran on May 7, which is roughly a month before the June 4 language naming personal injury appeared. It was not a workaround, because at the time there was nothing to work around. And the announcement is a paid press release on a newswire, not reporting. The same release lists "first legal marketing agency in space" among the agency's achievements. Take the claim for what it is.

But the mechanism is real, and it is the most plausible route into this platform right now. Which raises the question worth actually sitting with: what does the education lane cost you?

What the education lane costs you

OpenAI reviews the creative and the landing page together, and its policy says approved creatives cannot point at destinations that introduce disallowed content. So an educational ad pointing at an intake form is not an educational ad. To use this lane honestly you need genuine educational content, hosted on your site, and you have to accept that the click lands on an article rather than a case evaluation. No "injured in a crash? Call now." A guide to Pennsylvania comparative negligence.

Sit with what that means. You are paying per click to send someone to an educational article about their legal problem, hoping they convert later. That is content marketing with a CPC bill attached. And if you are going to build that content anyway, it will earn you organic citations inside ChatGPT's actual answers for free, above the ad slot, without a policy question hanging over it.

The education lane is answer engine optimization with extra steps and an invoice.

Whether that trade is worth it for your firm depends on your numbers, and we cannot answer that from here. Neither can any agency selling you this channel, and you should notice which ones are not raising the question at all.

The enforcement risk nobody is mentioning

Read OpenAI's own enforcement language before you spend a dollar here.

The policy says that when an ad or advertiser does not meet its standards, enforcement may include rejecting or removing an ad, limiting delivery, requiring edits, or restricting an advertiser's access to ads products. It goes further: severe, repeated, or deceptive violations may result in stronger measures, including suspension or termination of advertiser accounts. And approval is not permanent. OpenAI says it continues monitoring ads and advertiser activity after approval using user feedback, automated metrics, and ongoing quality evaluations.

Translate that into your world. If your PI ads are running because a classifier missed them rather than because a human approved them, you are not an early mover. You are an open enforcement ticket. The downside is not a paused campaign. It is termination of the advertiser account you will want when the category actually opens, at exactly the moment your competitors are getting approved through the front door.

That asymmetry is the whole calculus. The upside of being a few months early on a channel with a $3 to $5 CPC, in an auction with almost no legal competition in it, is modest. The downside is losing your seat.

The bar question is still unanswered

Set OpenAI aside. Your other regulator has not weighed in.

Start with the honest version, because you will hear the confident version from someone else.

Under Model Rule 7.3 and most state analogues, solicitation means a communication directed to a specific person the lawyer knows needs legal services. Contextual and keyword targeting has never been treated that way. Nobody has been disciplined for bidding on "car accident lawyer" in Google Ads, and there is no obvious reason ChatGPT placement would be analyzed differently. On the most natural reading, these are advertisements. That is probably where this lands.

But the format is new in a way that is worth sitting with. Google infers intent from two or three words typed into a box. ChatGPT ads are placed against a conversation in which a person may have described their own accident, their injuries, and their timeline, in detail, moments earlier. That is a difference of degree rather than kind, and degree is exactly the axis regulators tend to notice. No state bar has issued guidance on conversational AI placement. Nobody knows how a disciplinary committee reacts the first time it sees a retainer ad sitting under a transcript of someone describing a crash.

Our read is that this is advertising and will be treated as advertising. Our read is not a legal opinion, and it is not worth your license. Get one in writing from your ethics counsel before you run these, not after. It costs an hour and it settles the question for your firm specifically, which is the only place it actually matters.

The tell is not which answer an agency gives you. It is whether they raise the question at all.

What the platform actually gives you right now

Even setting policy and ethics aside, look at the product, because it changes the math on how hard you should be pushing.

When we first looked at this channel, the missing piece for a PI firm was geography: you could name your state in the headline, but you could not restrict delivery to it. OpenAI closed that gap in May 2026, adding targeting by state, DMA, and ZIP code. A firm licensed in one state can now hold delivery to that state, and ZIP-level control is actually finer than what most local advertisers get in Google by default. If you saw the earlier coverage saying ChatGPT had no geo controls, that is out of date.

What has not changed is how little control you have over which conversations trigger your ad. Geo narrows where an ad can serve; it does nothing to the underlying match, which is contextual. OpenAI decides which prompts are relevant, keywords function as hints rather than levers, and there are no negative keywords to fence off the ones you do not want. You can lock delivery to California and still surface against conversations you would never have bid on in search. For a practice area where the wrong intake is expensive to work, that unpredictability matters more than the geo unlock resolves.

That $3 to $5 CPC is not a bargain either. It is a signal that the auction has almost no serious legal competition in it, because on paper legal is not allowed in the auction. In a market where a competitive legal PPC click runs $200 and up, a $4 click looks like a discount only until you remember price discovery has not happened yet. The moment the category formally opens and every PI firm in the country piles in, that number is going to look nothing like $4.

Meanwhile, the opportunity that is not blocked by anything

When ChatGPT answers a legal question, it does not invent the answer. It pulls from sources and cites them. Getting your firm into those citations is available right now, costs nothing per click, and carries no policy risk or solicitation exposure.

That is answer engine optimization, and structurally it beats the ad slot everyone is fighting over. The citation sits inside the answer. The ad sits below the end of it. A citation carries the weight of having been chosen by the model rather than purchased, which is the exact trust signal someone wants at the moment they are deciding whether to call a lawyer. And unlike a placement that vanishes when spend stops, an earned citation compounds.

The work is not mysterious. Your content, your topical authority, your structured data, and how directly your site answers the questions real claimants are typing. It is the same discipline as good legal SEO, aimed at a different retrieval system.

What to do this quarter

Then, before you spend anything

• Get the ethics opinion in writing. Do it now, while you are not under time pressure. When the category formally opens, the firms holding an opinion will launch in a week and everyone else will spend six weeks in committee.
• Do not move budget from what works. There is no published cost per lead for legal on this platform. Funding an unproven channel by cutting a proven one is not a test, it is a donation.
• If you test anyway, test small and assume you could lose the account. Treat any spend here as the cost of learning, not as a case acquisition line. Do not build intake capacity around it.
• Put the real effort into AEO. Get cited in the answers. Nothing blocks it, no bar question hangs over it, and it compounds.
• Build first-party demand. This whole situation is the argument. A policy line moved and a channel opened or closed depending on who you ask. Google can move an algorithm. Meta can suspend an account. If one gatekeeper owns your case flow, you are one policy update from zero. Rent attention from platforms, funnel it into an audience you own.

Frequently asked questions