If you run a clinic, a lending business or a law firm, ChatGPT Ads is restricted to you rather than closed. There are two legitimate routes in: apply as an approved advertiser, or structure what you advertise so it is not in the restricted class to begin with. Both are in OpenAI's published policy. Wording around the rule is not a third route, and it is the one that costs accounts.
This is the companion to restricted industries on ChatGPT Ads, which covers what the policy says and who is affected. This page covers what to actually do about it.
Three separate gates, and you need all three
Most confusion here comes from treating this as one yes-or-no decision. It is three, and they are judged independently.
- Advertiser eligibility — whether your business is allowed to buy ads in its category at all.
- The offer, creative and landing page — whether the specific thing you are advertising is allowed, judged end to end.
- Placement — whether ads run in the conversations where you would want to appear.
You can pass the first and fail the second. You can pass both and still find the third quietly limits your reach. Plan for all three or the budget forecast will be wrong.
Route one: the approved-advertiser path
OpenAI's policy names this explicitly:
We may approve ads from approved advertisers within the financial services, healthcare & medicine, and legal services categories. These categories are being rolled out gradually with approvals being reviewed manually on a case-by-case basis.
Three words in there set expectations. Gradually means the category opens on OpenAI's timetable, not yours. Manually means a person looks at it, so it is slower than the few minutes a standard ad review takes. Case-by-case means another business in your category getting approved tells you very little about your own odds.
What the review looks at is business verification first, then the ads themselves. Restricted categories may draw enhanced verification or a manual pass rather than the automated review most ads get. Practically, that means the things which slow an application down are the boring ones: a business entity that does not match the domain, a landing page that does not clearly say who is behind it, missing licence or registration details where the category expects them.
What the path actually looked like for us
Policy tells you the rule. It does not tell you what arrives in your inbox when you apply. Across every advertiser account we have opened in a gated category, the shape has been the same — and it is not the shape the policy text implies.
Every one of them was refused first. Not held, not queued. Refused, with the broadest reason available:
Read literally, that is the channel closed to your business. It is not. The email ends with an Initiate appeal button, and that button is the actual route in.
What went into the appeal was ordinary. Documentation of how the business actually operates, and a plain explanation of what it does to comply with the advertising regulations that apply to it. No legal argument, no reframing of the offer, no attempt to talk the category open. Evidence of the practice, and how it stays compliant.
Every appeal was approved. And the reversal does not say the category opened or a queue cleared:
After reviewing your appeal, we determined that we incorrectly restricted your advertiser account from serving ads.
The first decision was wrong and the appeal caught it. Account opening to restored account ran a few days, and it ran to about the same length every time — this is a repeatable path, not one lucky account. The whole thing happened inside the ChatGPT Ads console and over email. There is no rep to call and no separate form to hunt for.
So the first refusal is a stage in the process rather than the answer to it.
So the plan is not "apply and wait". It is: apply, expect a refusal, and have the compliance documentation ready to appeal the day it lands. OpenAI still publishes no timeline and still reviews case by case, so nobody can promise you a date. But treating that first refusal as the verdict is the mistake that costs a quarter.
Route two: change what you advertise, not how you describe it
This is the route people miss, and it is written into the policy as a set of explicit carve-outs. In each restricted category, OpenAI names what may be permitted alongside what is not.
Healthcare and medicine. Regulated medical products, services and claims involving the prevention, diagnosis or treatment of a condition are currently disallowed, and the policy names clinical care providers, hospitals and prescription services among them. But it also says general health and wellness products that make no medical claims may be permitted, giving fitness equipment and wearable devices as examples.
Financial services. Ads for financial products and services are restricted to approved advertisers, and cryptocurrency, credit repair, debt settlement and debt assistance are disallowed outright rather than gated. The carve-out is for tools that do not promote a financial product or transaction — the policy names budgeting apps, general financial software, and educational material about finance that carries no offer.
Legal services. Advice, representation and services are not permitted, including immigration, personal injury, legal claims and document preparation. The carve-out is general legal education or media where no legal services are offered, and the policy's own examples are legal-themed podcasts and exam preparation courses.
The pattern across all three is the same. What is restricted is the regulated service and the offer attached to it, not the subject matter. A business that genuinely has a non-regulated product — a wellness line with no medical claim, a software tool with no financial transaction, an educational programme with no service attached — is advertising a different thing, and that is a real distinction rather than a reframing.
The end-to-end rule is where campaigns actually fail
This is the single most expensive thing to get wrong, because it passes creative review and fails later. The policy is direct:
Ads must be consistent end-to-end: approved creatives may not link to destinations that introduce disallowed content.
Review covers copy, images, video and landing pages. So an approved wellness ad pointing at a page whose primary call to action books a clinical consultation is a clinical care ad, and it will be read as one. The creative was never the problem.
If you are running route two, the landing page has to hold the line the ad set. That usually means a dedicated page rather than the homepage, because the homepage almost always offers the regulated service — that is what the business does. Split the destination before you split the campaign.
What does not work
Rewording. The enforcement section is worth reading in full before anyone gets clever with a headline:
Severe, repeated, or deceptive violations may result in stronger measures, including suspension or termination of advertiser accounts.
Review also continues after approval, using user feedback, automated metrics and ongoing quality evaluation. So an ad that slips through is not a win; it is an unresolved problem with a delay on it. And because enforcement escalates at the advertiser level rather than the ad level, the downside is not a rejected creative — it is the account, including whatever is running in your unrestricted categories.
Set against a channel that is a few months old and currently a small share of most budgets, that is a bad trade. A gated category is worth waiting for. It is not worth the account.
After approval, placement still applies
Worth planning for, because it surprises people who have just been approved: advertiser eligibility and ad placement are different rules. Ads do not run alongside sensitive user contexts, and OpenAI names mental and personal health conversations and emotionally reliant contexts explicitly.
For a clinic in particular, a meaningful share of the conversations you would most want to appear in are ones where no ads run at all. Approval opens the door. It does not put you in every room, and the reach forecast should be built with that discount already applied.
A practical build order
- Decide honestly which route you are on. If the money is made from a regulated service, you are on route one and waiting, or you are building a genuinely separate offer for route two.
- Get the business verification material in order first — entity name matching the domain, clear ownership on the site, licence details where the category expects them.
- Build the landing page before the ad. The page decides which category you are in.
- Keep the restricted offer on a separate campaign and a separate destination from anything unrestricted, so a problem in one does not put the other at risk.
- Apply with the compliance documentation already assembled, and expect to need it. On every account we have opened the first answer was a refusal, and the appeal is where it turned.
- Re-read the policy before each build rather than on a schedule. The categories most likely to change are exactly the ones currently restricted.
What to run while you wait
The answer is usually the channel where intent is already explicit. Someone typing a treatment and a city into a search engine is closer to booking than almost any conversational query, and that has not changed because a new channel launched. Watch cost per lead rather than click volume and the comparison between channels stays honest.
For what the channel costs where it is open to you, we publish what we actually pay by vertical with the sample sizes attached. For the mechanics of building a campaign once you are through, see how to advertise on ChatGPT, and the overview sits on ChatGPT Ads for small business.
Questions we get
Is "restricted" the same as "banned"? No. Restricted means there is a route through — either approved-advertiser review or a genuinely different, non-regulated offer. Some things are disallowed outright rather than restricted, and within financial services that includes cryptocurrency, credit repair, debt settlement and debt assistance.
How long does approval take? OpenAI does not publish a timeline, and because review is manual and case-by-case, any number you read elsewhere is somebody's single data point. In our own accounts the full path — apply, get refused, appeal with documentation, get restored — has run to a few days each time. That is our experience, not a commitment OpenAI has made to anyone.
What do you actually put in the appeal? Documentation of how the business operates, plus a plain explanation of what it does to comply with the advertising regulations for its category. That is what worked on every account we have appealed. It is not a legal argument and not a rewrite of the offer — it is evidence that the practice is a real, compliant one.
Can I advertise the practice rather than the treatment? Clinical care providers are named in the currently-disallowed list, so advertising the practice is the restricted thing. That is the distinction route two turns on, and it is the one worth getting advice on before spending.
Does being approved on Google or Meta help? No. Every network runs its own verification for regulated categories, and each one decides independently.
If my ad was rejected, have I damaged the account? Worth separating two different things. An advertiser account refused at signup is the appealable one described above. A single ad rejected inside a working account is just a rejection. What escalates is a pattern — repeated or deceptive violations are what the policy attaches account-level consequences to. Fix the cause rather than resubmitting variations of the same ad.
Not sure which route your business is on?
Bytown works out where a regulated or restricted business can actually spend, and runs it. Real marketers, one flat fee, full account transparency.
Book a callWe will tell you if the answer is "not this channel yet".
