OpenAI's own policy pages say two different things about legal advertising, and both are technically still live. Some agencies claim they have a working route into ChatGPT for law firms right now. Others report personal injury ads running inside the app despite an explicit ban. Here is exactly where the line sits today, why it keeps moving, and what a law firm should actually do about it.
The Short Answer
As of today, OpenAI's self-serve Ads Manager, the platform any business can sign into and launch a campaign with a credit card, formally excludes legal services. Ads for legal advice, representation, or legal services offered to individuals or businesses are on the disallowed list, alongside healthcare, financial services and political content, and the ban explicitly names immigration, personal injury and legal claims.
But that is not the whole story, and anyone telling you it is a clean yes or a clean no hasn't read OpenAI's own documentation closely enough. There is a genuine contradiction sitting inside OpenAI's policy right now, a documented enforcement gap, and at least one agency publicly claiming a working side door for law firms through manual approval rather than the self-serve platform. We'll walk through exactly what's confirmed, what's murky, and what we'd actually recommend doing with your budget this quarter.
Don't build a campaign strategy around ChatGPT ads for your firm today. The self-serve platform bans legal services outright, the exceptions are narrow and manually reviewed, and a firm that gets an ad live this month has no guarantee it survives the next policy update, this space has already flipped twice in four months.
The much bigger opportunity, and the one that isn't going anywhere, is earning a citation inside ChatGPT's actual answers rather than paying for a sponsored card beside them. We'll show you exactly what that takes further down.
What ChatGPT Ads Actually Are
OpenAI began testing sponsored placements inside the ChatGPT app in February 2026, then opened a self-serve Ads Manager to US businesses more broadly through the year. A ChatGPT ad appears as a clearly labelled sponsored card that sits below the assistant's written answer, it never sits inside the answer itself and never changes what the model says.
The targeting model is genuinely different to Google Ads. Instead of bidding on keywords or audience segments, advertisers write context hints, short descriptions of the kind of conversation their ad should sit beside. A firm doesn't target "personal injury searchers aged 25 to 54." It describes someone asking what to do after a car accident.
Ads are shown only to people on the free tier and the lower-cost Go tier, not to paying Plus or Pro subscribers. Ad matching happens inside OpenAI's own infrastructure using anonymised signals; personal conversation data isn't shared with advertisers directly.
The Timeline: How We Got to a Contradiction
The confusion isn't accidental, it's the residue of four policy changes in five months. Here's the sequence as best we can reconstruct it from OpenAI's own updates and independent reporting.
- February 2026
OpenAI begins testing sponsored placements inside ChatGPT for Free and Go tier users, in a limited pilot.
- April 2026
OpenAI updates its policy so medical, legal and financial advice conversations are no longer categorically blocked from carrying ads by default, a placement change, not a stated change to advertiser eligibility.
- 5 May 2026
The self-serve Ads Manager opens broadly, with CPC bidding and no minimum spend. During this window, at least one agency, ADSQUIRE, places what's reported as the first lawyer advertisement to run inside ChatGPT, an educational Pennsylvania personal injury campaign.
- 4 June 2026
OpenAI's disallowed-ads list is updated to explicitly exclude legal services, naming immigration, personal injury, legal claims and document preparation directly. The apparent loophole closes, on paper.
- July 2026 onward
Independent reporting notes personal injury ads still appearing inside ChatGPT despite the June update, an enforcement gap rather than a policy reversal. Legal marketing agencies begin publicly advertising "ways in" for law firms through channels other than the self-serve platform.
The Policy Contradiction, in Plain English
Here's the part that actually matters for your decision. OpenAI's ad policy page appears to carry two passages that haven't been reconciled with each other. One, dated from the June update, states plainly that legal services are not permitted. Another, added around the same period, describes a path where OpenAI may approve individual advertisers in regulated categories, including legal services, financial services and healthcare, through manual, case-by-case review, rolled out gradually.
OpenAI has not, as far as we can find, publicly clarified which passage governs. Read generously, this is what you'd expect from a fast-moving product team layering policy on top of policy without fully retiring the old language. Read less generously, it's a company keeping a door ajar for revenue from regulated categories it isn't ready to formally reopen.
Either way, the practical result is the same: a firm relying on the self-serve platform is reading an unambiguous ban, while a firm talking to the right agency is being told a manual-approval channel exists. Both of those things can be true at once, and neither tells you what the policy will say in October.
So Who's Actually Running Ads Right Now?
Genuinely, some firms are. This is the part most "ChatGPT ads are banned for lawyers" articles skip past, and it's exactly what prompted us to dig deeper for this piece.
At least one national legal marketing agency has publicly stated it offers a working route into ChatGPT advertising for law firms, not through the self-serve platform, which its own marketing acknowledges is closed to legal, but through what it describes as agency partnerships and a placement pilot with fewer competing firms bidding for the same placements. Separately, personal injury ads have been reported running inside ChatGPT conversations after the June ban took effect, which is either a manual-approval exception working as intended, or enforcement simply not catching every disallowed ad, we can't tell you which from the outside.
What we can tell you is that "some firms are already doing it" is not the same as "OpenAI has confirmed this is allowed." Being first through a gap in enforcement is a real, if fragile, advantage. It is also a position with no guarantee it survives the next policy tightening, and depending on how a campaign is framed, it can sit closer to a compliance problem than a growth channel.
The Narrow Exception That Actually Holds Up
Set aside the enforcement gap and look at what OpenAI's policy actually says is allowed, because there is a genuine, defensible exception underneath all this noise.
Ads for legal education or media, where no legal services are being offered, may be permitted. Think an LSAT prep course, a legal-themed podcast, or genuinely educational content explaining a general area of law without soliciting a client relationship. OpenAI's own framing gives the example of informing someone what they may need to do if they're injured in a car crash, provided the ad and its landing page aren't structured to seek new clients.
That's a narrower needle than most firms want to thread. An ad that names your firm, links to your intake form, and invites a call does not qualify as "educational" under any reasonable reading, no matter how the copy is worded. And even an ad OpenAI has approved as educational won't show inside conversations ChatGPT flags as sensitive, which covers most of the moments where someone has just been in an accident and is actually forming an opinion about who to call.
Why OpenAI Drew This Line at All
This isn't OpenAI being precious. Legal, healthcare and financial services carry the highest regulatory and reputational risk of any advertising category, exactly the categories every ad platform treats most cautiously at launch. Google Ads has run a comparably tight certification and landing-page review process for legal advertisers for years. Meta restricts personal injury and mass tort creative heavily. OpenAI is simply applying the same caution to a brand-new surface, with the added complication that a ChatGPT answer already carries an implied authority a search results page doesn't, users are more likely to trust what the assistant tells them, which raises the stakes on getting a legal ad wrong right next to it.
There's also a simpler commercial logic. Someone typing that they were in a car accident and need a lawyer is, in the current policy, seeing zero ads. The only firms present in that answer are the ones ChatGPT cites organically. That's uncomfortable if you're used to buying your way into search results. It's also, if you read it the other way, the single best argument for investing in AI visibility now rather than waiting for a paid door to open.
The Real Risk of Jumping in Now
If an agency offers you a route into ChatGPT advertising today, ask three questions before you sign anything.
- Is this the self-serve platform, or a manual-approval channel? OpenAI's own documentation says the former excludes legal outright. If it's the latter, get written confirmation of the approval, not a claim that one exists.
- What happens to spend already committed if the policy tightens again? It's tightened once, in June. There's no reason to assume it won't again, especially given the documented internal contradiction.
- Does the creative and landing page actually meet the disclosure, jurisdiction and solicitation rules your state bar or law society already requires? A platform allowing an ad doesn't make it compliant. US firms remain governed by state bar rules and ABA Model Rules 7.1–7.3 on advertising and solicitation regardless of what OpenAI permits; UK firms by the SRA Code; Canadian firms by their provincial law society and the Federation's Model Code; Australian firms by the Legal Profession Uniform Law and their state Legal Services Commissioner.
None of that means never test it. It means test it with your eyes open, a small enough budget that a policy reversal doesn't hurt, and your professional-conduct obligations checked before launch, not after.
What Actually Works Right Now: Earning the Placement Instead of Buying It
Every dollar you'd spend chasing an unstable paid channel is a dollar not spent on the one AI visibility lever that isn't going anywhere, and that ChatGPT itself has no ability to gate behind an ads policy: being the firm the model actually cites in its answer.
This is the shift we've written about before. A major structural change has happened in 2026: ChatGPT citations and Google rankings have stopped tracking each other. Only around 7% of ChatGPT citations also rank in Google's top 10 for the same query. Ranking first on Google for "personal injury lawyer in your city" no longer guarantees you're the firm ChatGPT recommends when someone asks it the same question conversationally, which the data shows they now do at a materially higher rate for legal services than almost any other professional category.
What Earns Citation Share in 2026
Citation share isn't won the same way search rankings are. It's earned through a specific, verifiable set of signals that the models weight when deciding who to name.
- Accreditation. Verifiable, third-party credentials like Law Society or state bar specialist accreditation carry real weight, they're a signal the model can check, not a claim it has to take on faith.
- Peer-review directories. Listings like Doyle's Guide or Best Lawyers, pay-to-play as they are, remain some of the most potent external signals currently observed feeding AI recommendations.
- Google review volume and recency. In markets where accreditation and directory listings are thin, the models fall back heavily on review volume, and currency matters as much as count, a hundred reviews from four years ago won't earn a mention.
- Named-attorney content with jurisdictional specificity. Generic practice-area copy performs worse than content tied to a specific lawyer's credentials and a specific state, province or territory's law.
- A clean off-site footprint. Consistent name, address and phone details across every directory and profile, clean schema markup, and a genuine media or press presence all compound.
None of that is exotic. It's the same review-engine, authority-listing and authored-content sequence we walk clients through in our AEO work, it just now carries more weight than ever, precisely because the paid door is shut. It sits alongside the broader AI & technology work we do with firms, including custom GPTs trained on a firm's own knowledge, which increasingly shapes how consistently a firm's expertise shows up across every AI surface, not just ChatGPT.
Run This Diagnostic This Afternoon
You don't need an agency to find out where you currently stand. This takes twenty minutes.
- Ask ChatGPT who the best firm is in your practice area and city.
Use the exact phrasing a prospective client would. If you're not named, ask it directly why not, and ask what would change that.
- Check whether a sponsored card appears at all.
If you see one for a competitor, note the framing, it's a live read on where the current enforcement line actually sits, not just the policy on paper.
- Audit your accreditation and directory footprint.
Are your lawyers' specialist accreditations, Doyle's or equivalent listings, and law society registrations all current, consistent and easy for a model to verify?
- Check review recency, not just count.
If your last batch of reviews is more than twelve months old, that's a gap worth closing before anything else on this list.
- Search for your own name alongside "reviews" and "complaints."
See what a model would find if it went looking for corroborating or contradicting signals about your firm.
What to Actually Do This Quarter
Build your review engine into a repeatable monthly process rather than a one-off campaign, keep your accreditation and directory listings current and consistent, and get named-attorney content live under real bylines with genuine jurisdictional specificity. That's the work that compounds regardless of what OpenAI's ad policy says next month, and it's the same work that will matter if and when the paid door genuinely opens.
If you want to test the paid channel anyway, do it on a small, defensible budget, get the manual-approval status in writing rather than taking an agency's word for it, and have your professional-conduct rules checked against the actual creative before it goes live, not the category in general.
And keep watching. This is a policy that's moved twice in four months. The firm that's actually ready when ChatGPT ads genuinely open to legal services will be the one that spent the interim building the organic visibility that makes any paid spend convert, not the one that sat out the whole conversation waiting for certainty that hasn't arrived yet. See our Practice Proof AI page for how we're applying that thinking across search, content and firm operations right now.