Your customers are already asking ChatGPT what to buy. As of February 2026, OpenAI reported 900 million weekly active users, and roughly one in five conversations on the platform now carries some kind of shopping intent. Until May 2026, the only businesses that could pay to show up in those conversations were ones willing to commit a $200,000 minimum spend. That changed on May 5, 2026, when OpenAI opened a self-serve Ads Manager to every business in the US with no minimum at all.
That is a real shift, not a hype cycle. But "you can now advertise here" and "you should advertise here today" are different questions, and the honest answer depends on what you sell and how much risk you can absorb on a new, still-maturing channel.
What actually launched on May 5
The self-serve platform at ads.openai.com lets any US business set up a campaign the same day, starting around $20 to $50 per day. You choose between two bidding objectives: Clicks, which buys on a cost-per-click basis, or Reach, which buys on cost-per-thousand-impressions for awareness goals. A cost-per-action option, where you only pay when someone clicks through and completes a signup or purchase, began rolling out to select advertisers on May 28, 2026.
Pricing sits in a fairly narrow band right now: $3 to $5 per click, or $25 to $60 per thousand impressions. Bids under $3 frequently fail to win any impressions at all, so there is a real floor on what "testing cheap" actually looks like.
What the ChatGPT Ads numbers actually say
Pricing and performance data from OpenAI's self-serve launch and early advertiser reporting, mid-2026.
That $200,000 starting minimum in February makes more sense once you see who OpenAI signed first: retail media platform Criteo joined as a launch partner, bringing roughly 2,000 brands onto the platform through its existing product-feed infrastructure. Self-serve access for everyone else came three months later, once the ad-serving and measurement systems had run at scale with those larger advertisers.
The numbers that matter before you spend anything
Criteo's most recent public update, from May 2026, reported that traffic referred from ChatGPT Ads is converting at close to double the rate of standard referral traffic in specific categories: consumer electronics, lifestyle and wellbeing, and home and garden. That is a meaningful signal, but it is an average across a favorable sample of established retail brands already running product feeds, not a guarantee for a five-person local business testing its first campaign.
Category-level conversion benchmarks from early 2026 advertiser data show a wide spread:
- Food and beverage: 6.0% to 11.0% engagement-to-conversion, the strongest of any category tracked
- Beauty and personal care: 5.5% to 9.0%
- Home goods and furniture: 4.0% to 8.0%
- Apparel and fashion: 3.5% to 6.5%
- Electronics and accessories: 3.0% to 6.0%
Across all industries and categories combined, reported ChatGPT ad conversion rates span 0.2% to 5.8%, with the gap driven almost entirely by purchase intent and where the shopper is in their decision when the ad appears.
The pattern across all of this data points the same direction: ChatGPT Ads work best for products people already research conversationally, where someone types something close to "what's a good [product] for [use case]" and gets an answer with your brand embedded in it. That is a different intent signal than a Google search ad, which catches someone actively comparing options, or a Meta ad, which interrupts browsing with something they were not necessarily looking for.
How to test it without wasting your budget
1. Set a learning budget, not a performance budget. $500 to $2,000 for your first 60 days is the range most early advertisers land on. The platform's attribution tooling is genuinely early-stage, so treat the first cycle as data collection: which categories respond, which creative angle performs, what your real cost per click ends up being for your specific audience.
2. Bid at or above $3 per click. Anything lower routinely fails to deliver impressions on this auction right now, which means an underbid campaign does not just underperform, it often produces no usable data at all.
3. Write the ad like a recommendation, not a search listing. Dense keyword strings and generic calls to action read poorly inside a conversational answer format. Copy that sounds like a genuinely useful suggestion, specific, concrete, with real product detail, is what's converting in early advertiser reports.
4. Do not pull budget from Google or Meta to fund this. Those channels have over a decade of attribution and retargeting maturity behind their auctions. Fund a ChatGPT Ads test from a separate testing line, and only reconsider the split once you have a full 60-day cycle of your own results.
5. Pair it with GEO, not instead of it. Paid placement inside ChatGPT and organic citation inside ChatGPT are two different systems that reinforce each other; a brand that shows up organically in generative answers tends to build more trust with the same audience an ad reaches. If you have not started that groundwork, our GEO and AEO guide covers how citations get earned in AI search separately from anything paid.
How targeting actually works, and how to track results
ChatGPT Ads does not target the way Google or Meta does. There is no age, income, or interest-history targeting available to advertisers. Instead, you write "context hints" at the ad group level, plain-language descriptions of the kind of conversation or situation where your ad should appear (something like "someone asking for a reliable local plumber" or "someone comparing standing desks for a home office"). OpenAI's system matches those hints against the topic and intent signals of the live conversation, not a stored profile of the user. That is a meaningfully different skill than writing a Google Ads keyword list; it rewards specific, situational phrasing over broad category terms.
Measurement caught up to the ad platform in May 2026, when OpenAI shipped two tracking tools alongside the self-serve launch. A JavaScript pixel, distributed as a small SDK called OAIQ, drops a first-party cookie on your domain when someone clicks through from an ad, good for a 30-day attribution window. A server-side Conversions API sends purchase and signup events directly from your website to OpenAI, which holds up better than pixel-only tracking as browsers keep tightening third-party cookie rules. Attribution defaults to a 7-day click and 1-day view window, configurable per campaign. Skipping both tools is the fastest way to burn a test budget with no usable read on what happened; install at least the pixel before your first campaign goes live.
Who can actually advertise here right now
OpenAI is running a restricted category list during this test period, and it matters more than the pricing for deciding whether to bother. Every ad clears a two-stage review, automated classifiers plus human review for anything borderline, before it runs. Lifestyle and household goods, local services, travel and experiences, and digital products or education are the categories OpenAI is actively approving. Health, mental health, dating, politics, crypto, alcohol, tobacco, gambling, and credit repair are excluded outright, and financial, healthcare, and legal advertisers currently need pre-approval even when the category is technically allowed.
That local services inclusion is worth pausing on, because it cuts against the assumption that this channel only works for e-commerce. A home services or local professional business, the kind of client who might otherwise run Google Ads for lead generation, sits inside an explicitly approved category here, even though the conversion benchmarks published so far lean toward retail categories simply because retail brands were the first advertisers on the platform. Check current policy at OpenAI's own ad policy page before building a campaign; the approved list has moved twice already since February and will likely keep moving as the platform matures.
Where this fits next to what you already run
If you are already running Google Ads or Meta Ads, ChatGPT Ads is not a replacement conversation yet. It is a small, bounded experiment worth running if your category is on the approved list and your product fits a conversational research pattern, whether that is a considered retail purchase in home goods, beauty, electronics, or food and beverage, or a local service someone is actively asking ChatGPT to recommend. Businesses in the excluded categories, health, finance without pre-approval, politics, should not plan around this channel yet regardless of demand.
The honest verdict: test it if $500 to $2,000 is money you can spend on genuine learning rather than money you need back this quarter, your category is approved, and you install the pixel or Conversions API before you launch. Skip it for now if your category is excluded, and revisit in another two quarters once the approved list widens and attribution tooling matures further. Either way, this is a channel worth having an opinion on well before your competitors do, the kind of shift covered in more depth in our broader look at what's actually working in SEO after the AI search shift.
If you want help deciding whether a ChatGPT Ads test fits your specific product and budget, talk to our team about where it sits in your broader marketing mix.
FAQ
Questions, answered.
What small business owners ask before testing ChatGPT Ads.
No. The self-serve platform that opened May 5, 2026 has no minimum spend, and you can start a campaign for $20 to $50 a day. That said, treat your first month as a $500 to $2,000 learning budget, not a performance budget. Attribution tools on the platform are still early, so the goal of a first test is data on your specific product and audience, not an immediate return.
GEO (generative engine optimization) is about earning organic citations when someone asks ChatGPT a question in your category, the same logic as SEO for search engines. ChatGPT Ads is a paid placement inside the conversation, bought through an auction, separate from anything you rank for organically. The two are complementary, not substitutes: GEO earns visibility for free but takes months to build, while ads buy visibility for a specific product or offer starting today. See our guide on GEO and AEO for the organic side of this.
Early category data favors businesses with a clear, considered-purchase product where someone is already asking ChatGPT for a recommendation: home goods and furniture, beauty and personal care, electronics, and food and beverage all show conversion rates from roughly 3.5% to 11% depending on category. Businesses selling low-consideration, low-margin items or services with long, relationship-driven sales cycles are seeing weaker early results, since the ad appears inside a single conversational answer, not a full-funnel experience.
Not yet. Google and Meta have over a decade of attribution data, retargeting infrastructure, and predictable auction dynamics that ChatGPT Ads does not have. Treat a ChatGPT Ads test as incremental spend from a testing or innovation budget, not a reallocation from channels with a proven CPA. Revisit the split after you have at least one full learning cycle of your own data, roughly 60 days.
The most common mistake is treating a ChatGPT ad like a search ad: dense keywords, no personality, a generic call to action. The format sits inside a conversational answer, so copy that reads like a helpful recommendation outperforms copy that reads like an ad. The second mistake is bidding under $3 per click to save money; those bids frequently fail to deliver impressions at all, which wastes the test window without producing any usable data.
