ChatGPT Ad Policies: What's Allowed, Restricted and Banned in 2026
What you can and cannot advertise on ChatGPT, the restricted categories that need extra care, the copy and image rules that trip people up, and a pre-submission checklist.
OpenAI's advertising rules are tighter than what most performance marketers are used to. Copy that would sail through on social platforms, such as "the best moisturizer you'll ever try" or "only 3 left, order now," is likely to be rejected in ChatGPT.
That strictness makes sense. Ads sit right below an answer from an assistant people trust for advice, so OpenAI is protective of that space. This guide summarizes the rules as publicly described in 2026, organized as banned, restricted and allowed, then gives you a checklist to run before every submission.
A note on accuracy: policies evolve, especially during a beta. This is a practical summary, not legal advice. Always check OpenAI's current advertising policies before launching.
The principle behind the rules
Nearly every rule traces back to one idea: an ad in ChatGPT must not mislead the person reading it or exploit the trust they place in the assistant.
That shows up in three ways:
- Ads are separate from answers. Cards are labelled sponsored and visually separated, and ads do not change ChatGPT's answers.
- Ads cannot pretend to know you. Copy cannot imply the advertiser knows what the user said in the conversation.
- Claims must be true and provable. If you cannot back it up, do not write it.
Prohibited products and services
These cannot be advertised in ChatGPT at all, as of 2026:
- Illegal or recreational drugs and drug paraphernalia
- Weapons, ammunition and explosives
- Tobacco and vaping products
- Counterfeit goods
- Adult content and sexual services
- Gambling
- Hacking tools and services designed to gain unauthorized access
- Payday lending
Political ads are also out of scope for ChatGPT advertising. If your business sits in one of these categories, ChatGPT ads are not an option.
Restricted categories
Restricted means allowed, with extra review and tighter wording. As of 2026 these include:
| Category | What to watch |
|---|---|
| Health, supplements and wellness | No cure or treatment claims, no guaranteed results |
| Weight management | No before-and-after, no promised pounds lost |
| Financial products | Clear terms, no guaranteed returns or income |
| Alcohol | Adults only, local law applies, no excess |
| Dating | Respectful framing, no sexual content |
| Products for minors | Extra care with targeting and claims |
If you sell in any of these, assume every claim will be read closely. A hypothetical example for a magnesium supplement:
- Likely rejected: "Fix your insomnia in 7 days"
- Safer: "Magnesium Glycinate in Easy-to-Swallow Caps" with body "Third-party tested. 60 capsules per bottle. Free shipping on your first order."
The safer version describes the product, its verifiable attributes and the offer. It leaves the outcome to the buyer and the landing page, which must also stay within the rules. Our guide to ChatGPT ads for supplements and wellness goes further.
Copy rules that trip people up
Deceptive or unverifiable claims
"Clinically proven," "#1 rated," "best on the market" and "doctors recommend" all need hard proof you could show a reviewer. If you do not have a published study, an independent ranking or documented endorsements, cut them.
Invented statistics and fake reviews
No made-up numbers ("9 out of 10 customers..."), no fabricated testimonials and no review counts you cannot verify. If you cite a real number, it must match what is on your landing page.
Guaranteed outcomes
You cannot promise results, especially health cures or income. "Make $5,000 a month" or "clear skin guaranteed" are out. A money-back guarantee on the product itself is different, since that is an offer, not an outcome claim, as long as the terms are real.
Exploiting hardship or insecurity
Copy cannot prey on fears or tough situations: debt, illness, appearance anxiety, loneliness. "Embarrassed by your skin?" is the kind of hook to avoid.
Implying you know the conversation
"Saw you were asking about back pain..." is a clear violation. Write cards as a relevant option, not a reply to the user.
Impersonation and trademarks
No impersonating OpenAI, ChatGPT, public figures or other brands, and no using other companies' trademarks. That rules out "Better than [Competitor]," "Works with ChatGPT" phrasing that implies partnership, and celebrity names without permission.
Clickbait, ALL CAPS and fake urgency
"You won't believe this," "LAST CHANCE," and countdown language on an evergreen offer are out. A real, dated sale can be mentioned plainly if it is genuinely time-limited and the landing page shows it.
Image rules
Images follow the same principles:
- No before-and-after transformations, especially for skin, weight or body changes.
- No shocking, violent or hateful imagery.
- No OpenAI or ChatGPT branding or anything that looks like part of the ChatGPT interface.
- No competitor logos or products.
- Keep text in the image minimal, and make sure any text follows the copy rules too.
Product-in-use shots, clean product photography and simple lifestyle scenes are the safest and usually perform well in a small card.
Landing page rules
The landing page must match the ad. If the card says "Free shipping on orders over $50," the page must say the same. If the card shows a specific product, the click should land on that product or a page that clearly features it.
Mismatch is both a policy issue and a performance issue. The message match checker compares your ad and landing page copy for gaps.
Allowed vs not allowed: quick examples
For a hypothetical running shoe brand:
| Not allowed | Allowed |
|---|---|
| The Best Running Shoe Ever Made | Cushioned Trainers for Longer Runs |
| Doctors say these stop knee pain | Extra midsole foam and a wider toe box |
| ONLY 2 HOURS LEFT | Free returns within 30 days |
| Faster than [Competitor] | 8mm drop, 9.1 oz in a men's size 9 |
| Tired of being slow? | Built for new runners training for a 10K |
The allowed versions are specific, verifiable and calm. They also tend to perform better, because a concrete detail answers a question while a superlative just asks for trust.
A pre-submission checklist
Before you submit any chat card, check:
- Title is 3 to 50 characters and body is about 100 characters or less.
- Every claim is true and you can prove it.
- No superlatives ("best," "#1") without documented proof.
- No invented stats, reviews or testimonials.
- No guaranteed health, income or body outcomes.
- No before-and-after imagery.
- No competitor names, trademarks or public figures.
- No reference to OpenAI or ChatGPT that implies endorsement.
- No language that implies you know the user's conversation.
- No hardship or insecurity hooks, ALL CAPS, clickbait or fake urgency.
- The product is not prohibited, and restricted-category rules are followed.
- The landing page repeats the promise and offer exactly.
The free ChatGPT ad policy checker runs a draft through common versions of these checks. For copy that passes and still sells, see our ChatGPT ad copy examples.
What to do if an ad is rejected
- Read the rejection reason in Ads Manager.
- Find the specific word, claim or image that triggered it. It is usually one phrase.
- Rewrite with a concrete, verifiable detail in place of the claim.
- Check that the landing page matches the new version.
- Resubmit. Do not try to slip the same claim through with different wording.
Repeated violations can affect your account, so fixing the root cause matters more than getting one ad approved.
How SecondWin helps
Every ad SecondWin writes is checked against OpenAI's ad policies and SecondWin's own stricter rules: no unproven claims, no competitor names and no made-up facts. We start from the proven promise, proof and offer in the longest-running ads in your niche on Meta's Ad Library, then write original chat cards that make those messages compliant for ChatGPT. Campaigns run in your own OpenAI ad account.
See which messages we would use for your brand in a free URL analysis, or check plans and pricing. SecondWin is independent and not affiliated with OpenAI.
FAQ
Can I advertise supplements on ChatGPT?
Supplements, health and wellness are restricted rather than banned as of 2026, so they can run with extra review. You cannot claim to cure or treat conditions, promise guaranteed results or use before-and-after imagery. Focus copy on verifiable product facts such as ingredients, testing, format and offer, and make sure your landing page follows the same rules as the ad.
Can I mention competitors in ChatGPT ads?
No. OpenAI's policies prohibit using other companies' trademarks and impersonating other brands, which rules out comparison headlines like "better than" a named competitor. Instead, describe what makes your product different in concrete terms, such as materials, specs, price or guarantee, and let the buyer draw the comparison themselves on your landing page.
Can I use urgency in ChatGPT ads?
Only genuine, verifiable urgency, stated plainly. Fake urgency, countdown language on evergreen offers, ALL CAPS and clickbait are not allowed. If you are running a real sale that ends on a specific date, you can mention it calmly, as long as the landing page shows the same offer and end date. When in doubt, lead with the offer itself rather than the deadline.
Why was my ChatGPT ad rejected?
The most common causes are unverifiable claims or superlatives, invented numbers, guaranteed outcomes, before-and-after images, competitor or trademark mentions, copy that implies you know the user's conversation, fake urgency, and landing pages that do not match the ad. Read the rejection reason, find the specific phrase or image, replace it with a concrete fact and resubmit.