We Tested ChatGPT Ads for Two Weeks: Here's What Happened

We tested ChatGPT ads for two weeks, got early access to the beta ad manager, and ran a real campaign with our own money. This is the honest write-up: about 5,000 impressions, roughly 40 clicks, $200 spent, zero conversions. Not a flattering number in sight. Over the same fortnight, organic and referral traffic from being cited inside ChatGPT sent us more visitors than the paid ads did. So this ChatGPT ads review is less a verdict and more a field report from the first weeks of a very new surface.

TL;DR: In a two-week ChatGPT ads test we spent $200 for ~5,000 impressions, ~40 clicks, and 0 conversions. The beta ad manager auto-charged our card on verification, failed payments repeatedly, and kept nudging us to raise bids until a two-week budget burned in about two days. Over the same window, SEO and GEO referral traffic (getting cited by ChatGPT) out-earned the paid ads. Promising surface, not yet ready to beat organic for us.

If you only read one paragraph: ChatGPT ads work mechanically, the placement is real, and the reporting is clean. But a rushed creative on a brand-new auction is a fast way to donate $200 to OpenAI. The result was not the platform's fault so much as ours. We treated it as an experiment, not a campaign. Still, the friction we hit is worth documenting, because you will hit some of it too.

What ChatGPT ads actually are in 2026

ChatGPT ads are sponsored placements that appear below a ChatGPT answer when the conversation surfaces a relevant product or service. OpenAI announced the test on January 17, 2026, and the pilot went live in the United States on February 9, 2026, for logged-in adult users on the Free and $8-per-month Go tiers. Plus, Pro, Business, Enterprise, and Edu stay ad-free. That is straight from OpenAI's own Testing ads in ChatGPT post and Fox Business's coverage of the announcement.

The unit itself is modest. A product name, a short description, and a link button render as a card, clearly labeled "Sponsored," below the organic answer. OpenAI is explicit that ads do not influence how the model answers and that advertisers never see your chats, history, or memories. Ads sit outside the answer, not inside it.

For anyone who buys media on Meta or Google, the mental model is closer to a search ad than a social feed ad. There is no scroll to interrupt. The user asked a question, got an answer, and the ad rides the intent underneath. That intent signal is the whole pitch.

Which countries can run ChatGPT ads

Availability is limited. As of mid-2026, advertiser access to the self-serve ads manager covers the US, Canada, Australia, and New Zealand, with the UK becoming the first European market where ads appear. OpenAI has named the UK, Japan, South Korea, Brazil, and Mexico as next expansion markets. This matters for targeting: if your buyers sit outside those countries, the surface simply is not open to you yet. Treat the current list as a moving target and check before you plan a launch. It is a beta, and it behaves like one.

How the auction and objectives work

The ChatGPT ads manager in 2026 supports two buying objectives at the base level: Reach (pay per 1,000 impressions, CPM) and Clicks (pay per valid click, CPC). A third, conversion-optimized objective began rolling out in early June for advertisers with tracking installed, per Search Engine Land. Conversion campaigns require OpenAI's ad pixel on your site, and OpenAI encourages sending action data back through its API. Reporting shows impressions, clicks, spend, CTR, average CPC, average CPM, and conversions.

How we set up our two-week ChatGPT ads test

We did the one thing every media buyer would tell you to do: we sent traffic to a dedicated landing page, not the homepage. The homepage answers ten questions at once. A landing page answers one. Ours had a short video, a clear mention of the app, and a single conversion-focused layout built around one action. If you are going to buy intent-heavy clicks, do not spend them on a page that makes people choose what to do next. This is the same discipline we apply when we study a competitor's landing page intelligence before writing our own.

Ad creation inside ChatGPT is conversational. You describe the ad to the manager in chat, and it drafts copy and generates images in the same thread. We iterated on headlines this way, testing angles like "access any ad manager, any ad, across platforms," "the best ads in seconds," and "winners in seconds." One of the AI-generated ad images came out genuinely good, better than we expected from a first pass. The building experience is the strongest part of the product right now.

We set a modest daily and total budget and intended a clean two-week run. Because targeting availability is geo-limited, we ran US-focused. That was a constraint, not a choice, and it is one you should plan around rather than fight. If you want to understand how geo restrictions shape reach, our note on geotargeting covers the mechanics, and the platform's own geo filters mirror the same logic on the research side.

The headlines we tested and why they were weak

Honesty first: our creative was thin. We wrote headlines that described the product instead of hooking a pain. "Winners in seconds" is a feature claim, not a reason to click. A stronger angle would have opened on the frustration the reader already feels: the tedium of digging through ad libraries by hand. We know this. We teach it. We just did not do it here, because this was a test, not a launch. The lesson is old and it held: levels of awareness matter, and a cold ChatGPT audience needed a problem-aware hook, not an offer-aware one.

What went wrong: bugs, billing, and a two-day budget

Here is where the field report earns its keep. Three real problems, in order of how much they cost us.

The card auto-charged on verification. When we added a payment method, the manager placed what looked like a verification authorization of about €100 on the card. Not a spend, but not nothing either, and not clearly signposted before it happened. If you are testing with a personal card, expect the hold.

Payments failed, repeatedly. More than once the campaign stalled on a failed payment, and each time we had to re-add the card and restart. That reset cost us momentum and, worse, reset the learning the auction had started to build. Public reporting notes that OpenAI has been lowering billing barriers for the beta, which tracks with a system still being hardened.

The manager pushed higher bids, and we let it. The ad manager kept prompting us to raise our bid and buy pricier clicks. We accepted the nudges. The result was fast: a budget scoped for two weeks was gone in about two days. That is a classic beta-auction failure mode, and the fix is boring discipline. Set a hard total cap, ignore the upsell, and pace it yourself. If you want a sanity check on what a click should cost you before you start, run the numbers through our CPC calculator and an ad budget planner first, not after.

None of this is disqualifying. It is what an early ad system looks like before the guardrails are built. But it is the difference between a $200 experiment and a $200 lesson.

The results, reported straight

No spin. Here is the full two-week window.

Zero conversions is the headline number, and we are not going to dress it up. A ~$5 CPC on a brand-new surface is not outrageous for a beta, and it lines up with the ~$3 to $5 range others have reported. But with weak creative and no conversion tracking maturity, the clicks had nowhere good to land. The conversion funnel leaked at the creative stage before the landing page ever got a fair test. If we ran it again, we would fix the hook first and only then worry about bids.

The honest lesson: SEO and GEO out-earned the ads

The most useful finding was not on the ad dashboard. Over the exact same two weeks, organic search traffic and referral traffic from ChatGPT itself, people arriving because ChatGPT cited or mentioned us in an answer, sent more visitors than the paid ads did. We paid the model $200 for ~40 clicks. The model sent us more than that for free, because our content was worth citing.

That is the whole media arbitrage argument in one data point. Right now, for us, being cited by the model out-earns paying the model for clicks. The economics of earnings per click only justify scaling paid spend once the creative and the funnel clear the bar organic already clears. Ours did not, yet.

This is exactly why optimizing content for AI search and LLM visibility has moved from a nice-to-have to the core play. Generative engine optimization, earning citations inside AI answers, is the referral channel that beat our paid campaign without a media budget. Our SEO strategy for AI search guide walks the mechanics, and if you are weighing which model to write for, our take on Claude versus ChatGPT for marketers is a useful companion. The punchline writes itself: this article is trying to get cited by ChatGPT for "ChatGPT ads," which is a cheaper path to the same audience than the ads we bought.

None of that means paid is dead. It means, for our unit economics in this window, organic won. Your math may differ, especially if your product monetizes clicks harder than ours does.

There is no ChatGPT ad library yet

Here is the gap that matters for anyone doing competitive research. There is no ChatGPT ad library. You cannot browse the sponsored placements other advertisers are running the way you can browse Meta's Ad Library or Google's Ads Transparency Center. OpenAI labels ads as sponsored and gives users controls to dismiss or learn why they saw an ad, but there is no public, searchable transparency surface for ChatGPT ads as of mid-2026, and none has been committed to.

That is a real blind spot. Meta's Ad Library set the expectation that ad transparency is browsable, and it remains the originator of that model. Until ChatGPT has an equivalent, you cannot spy on ChatGPT ad creative, you cannot see competitor headlines, and you cannot build a swipe file. When a ChatGPT ad library does exist, adlibrary will cover it, the same way we already unify Meta, Instagram, TikTok, YouTube, Snapchat, Pinterest, LinkedIn, and Google into one unified ad search. Multi-platform coverage is the entire point of the product, and a new ad surface is a new column, not a new tool.

Until then, the research workflow for AI-era advertising runs through the platforms that do publish their ads. Our competitor ad research workflow and the saved ads board are where you build the creative library ChatGPT does not give you, and AI ad enrichment tags the angles so you can see patterns across thousands of ads at once. For teams that automate this, the API access tier is a paid power-user upgrade over Meta's free Ad Library API: more data per ad, coverage across platforms instead of Meta only, and no app-review or verification gauntlet to implement it. Meta's free API is the origin of the category; ours is the multi-platform layer on top.

Who should try ChatGPT ads right now

Try it now if you sell into the US, Canada, Australia, New Zealand, or the UK, you have a genuine conversion-tracked funnel, and you can commit real creative, not a test ad. The intent quality is high, and being early on a channel has compounding advantages, the same reason our media buyer workflow treats new surfaces as land-grabs.

Wait if you are hoping ads will out-earn a mature SEO or GEO program, or if you cannot yet track conversions properly. Without a working pixel and clean attribution window, you are buying clicks blind, and view-through conversions on a new channel are even harder to trust. Our note on tracking Meta ads attribution accurately applies here too: measure before you scale.

For the strategic view of where this all heads, how ChatGPT ads will reshape digital marketing is the prediction piece, and how agencies can prepare for ChatGPT ads is the prep guide. This post is the results. Read all three and you have the arc: forecast, preparation, and the first honest field data. If you want to see how AI agents are already being pointed at ad research, LLM agent ad research tools covers the automation side.

Frequently asked questions

Do ChatGPT ads work? Mechanically, yes, ChatGPT ads deliver impressions and clicks with clean reporting. In our two-week test they did not convert, but that was weak creative on our side, not a broken platform. The intent quality is high; the surface is early. Whether they work for you depends on your creative, your tracking, and whether you are in an available country.

How much do ChatGPT ads cost in 2026? ChatGPT ads run on CPM and CPC bidding with reported click costs around $3 to $5, and the self-serve ads manager dropped the earlier six-figure minimum. We spent $200 total across two weeks. Note that a two-week budget can burn in days if you accept the manager's prompts to raise bids, so set a hard cap.

Is there a ChatGPT ad library to spy on competitors? No, there is no ChatGPT ad library as of mid-2026. Unlike Meta's Ad Library or Google's Ads Transparency Center, ChatGPT has no public, searchable surface for browsing sponsored placements. When one launches, adlibrary will add it to its multi-platform coverage.

Which countries can run ChatGPT ads? As of mid-2026, advertiser access covers the US, Canada, Australia, New Zealand, and the UK, with Japan, South Korea, Brazil, and Mexico named as next markets. It is a limited beta, so confirm current availability before planning a campaign.

Should I use ChatGPT ads or SEO? For us, SEO and GEO out-earned ChatGPT ads over the same window, because getting cited by the model sent more traffic than paying it for clicks. If your funnel monetizes paid clicks efficiently and you can track conversions, ChatGPT ads are worth an early test. Otherwise, invest in AI-search visibility first.

We do not think ChatGPT ads are a mistake. We think we ran a sloppy first test on a promising channel, and the honest numbers earned us a clearer plan for the next one. Revisit the surface as it matures, keep your creative sharp, and never let an ad manager set your budget for you.

Want to build the creative library ChatGPT does not give you? Start free or see pricing.