What a month of ChatGPT ads taught me in the health space
I ran a four-week ChatGPT ads campaign for a company that sells health tests. Here is what the platform rewards, what its reviewers block, and what people did after they clicked. ChatGPT started showing ads this year, and I wanted real numbers before recommending the channel to anyone. Partway through, I sat in on a […]
I ran a four-week ChatGPT ads campaign for a company that sells health tests. Here is what the platform rewards, what its reviewers block, and what people did after they clicked.
ChatGPT started showing ads this year, and I wanted real numbers before recommending the channel to anyone.
Partway through, I sat in on a Similarweb webinar about advertising inside ChatGPT. Their data lined up closely with what my campaign was showing, so I’ve put both together here.
ChatGPT plays the trusted advisor
People don’t open ChatGPT to browse. They open it to ask a question, and the answer often is a surrogate for the friend, colleague or expert they would have asked before. That puts ChatGPT somewhere in the middle of the funnel, where people research and compare before they commit.
Similarweb shared a few numbers that explain why this matters:
- AI-assisted shopping is growing fast. Shopping sessions that involve AI have grown 2.5x in the past year.
- Brand mentions trigger ads. When someone names a specific brand in a prompt, there’s roughly a 60% chance an ad appears. A person asking about a competitor can see your ad in the middle of their comparison.
- Not everyone sees ads. Only Free and Go plan users see them; premium subscribers never do. About 43% of eligible US users are shown ads, and 40 to 50% of ad-eligible users see at least one ad a month.
- Click-through rates are rising. CTR has nearly tripled since ads launched, even as more advertisers have joined. That suggests the relevance matching is improving and people are getting comfortable with the format.
- It targets intent, not audience. ChatGPT works like Google search, where people are looking for information, guidance or learning. It doesn’t work like Meta or Instagram, where you pick an audience profile.
- Advertisers never see conversations. You get aggregated impressions and clicks, and nothing about what anyone asked.
Source: Similarweb webinar on ChatGPT advertising, fall 2026.
What the best advertisers do
Similarweb’s view of the top advertisers comes down to three habits:
- They run a lot of ads. The leaders run a high volume of distinct creative and landing pages. Experimentation is the norm.
- They lead with an action verb and speak directly to what the person is trying to do.
- They send clicks to a specific page. The landing page matches the question, and it’s never the homepage.
Because ChatGPT matches ads to intent, every kind of question needs its own ad. A few broad ads can’t cover the range of things people ask.
The hard part: getting a health ad approved
Health is a sensitive category, and ChatGPT reviews these ads closely. Only a handful of my ads were approved. The platform doesn’t say why an ad was rejected, so I compared approved and rejected ads line by line. The pattern was consistent. ChatGPT approved education and rejected direct promotion of diagnostic products that make clinical claims, or for content promotion, blogs that make clinical or diagnostic claims.
Rejected Ads:
- Selling the diagnostic claims
- Linking to results help
- A health condition in the destination URL
Approved Ads:
- Linked to general education. Every approved ad went to an educational article or the blog that was informational but didn’t make claims
- Never tried to sell a health product
- Education first ads. The lead verbs were “Explore,” “Learn,” “Insights into” and “Join a webinar.”
- Hedged, third-person science. Phrases like “potential impacts,” “studied for” and “can contribute” helped get ads approved
Two side-by-side comparisons
In four cases only one thing changed between versions of an ad, which shows what the reviewer is screening for.
- The link mattered as much as the copy. An ad was rejected twice while its link ended with a health conditional. With the same copy and a link ending in a neutral topic word, it was approved. Both links led to the same page.
- Rewriting the copy didn’t help. An ad explaining one of the client’s reports was rejected three times. Each rewrite kept the same link to a page about interpreting results.
The playbook: teach in the ad and sell on the landing page. Name tests as topics, never as offers. Keep condition words out of the URLs you submit, and never ask readers about their own health.
Tracking user behavior on the website and how they convert
The ads that worked were top-of-funnel education: explainers, research summaries and a webinar recap. They reached people in the research phase who were reading up on a topic before deciding anything. A webinar recap ad had the best click-through rate in the account, at 1.34%.
Those visitors stayed and explored. Google Analytics recorded 273 sessions from ChatGPT ads with a 95% engagement rate, 24 seconds of average engagement time and 3.56 events per session.
Following individual visitors showed where that interest went. Of seven sessions I traced, four moved from an education article to a product page, and an ordering portal. These were the kinds of paths I saw:
- Education article→Product page→ Report (key conversion)
- Education article→Related blog post→Report→Product page→Ordering portal (key conversion)
- Education article→Spanish-language version→Homepage
- Education article→About the company
- Webinar ad→Related topic page
Your ad is only as good as the conversions you set up
The platform reported 225 conversions from 329 clicks. A 68% conversion rate tied to triggering key events like webinar page signups, registrations and health product page views. Keep in mind if you have high value and low value conversions mixed together it will muddy the waters of what the conversion actually means. Further complicating matters, you’ll have some users that produce multiple conversions while others produce none yet at the end of the day your campaign has 329 clicks and 225 conversions.
If you run ChatGPT ads, report lead actions such as mailing-list signups and completed registrations separately from page views. That’s the only way to see your real cost per lead.
Cutting the budget from $57 to $25 a day
Reserved for update on results of the $25/day test
[Add a paragraph on what happened after the budget cut.]
- Did clicks and impressions drop in line with spend, or did the platform deliver more efficiently?
- What happened to CTR and cost per click?
- Did platform conversions and lead conversions hold their rate?
- What’s the takeaway on budget for a small account limited by ad approvals?
What I tell health brands starting on ChatGPT ads
- Lead with education. Write ads that teach, link them to articles, and let the article lead people to the product.
- Watch your language. Leave out “test,” sample types, “patients,” “your,” diagnostic verbs like “detect” and “pinpoint,” and condition names in URLs.
- Build many intent-specific ads. Every question type and audience needs its own ad, and more approved ads means more delivery.
- Resubmit compliant ads. Review isn’t perfectly consistent.
- Be the answer, too. Premium subscribers never see ads. Publish clear answers on your own site that ChatGPT can cite.
- Measure leads, not page views. Separate signups and registrations from landing-page views.
- Fix the path after the click. Interested visitors will go looking for the next step, so make it easy to find.
ChatGPT ads are new, and the rules are still settling. If you’re in a regulated category and want help planning your first test. Feel free to reach out with questions on your campaign
