Should Your Brand Run ChatGPT Ads? The Profit Maths Before You Test

ChatGPT ads went from a $200,000-minimum pilot to a $25-a-day self-serve channel in less than a year. Whether your brand should run them comes down to one number you already have: how many dollars of contribution margin your first order leaves before ad spend.

Here's the short answer first, then the maths and a 30-day test plan.

ChatGPT ads in 30 seconds (as of October 2026)

• What they are: sponsored units shown below ChatGPT's answers, matched to the conversation rather than a typed keyword. Each one carries your name, logo, headline, copy, an image and a landing page, according to OpenAI's Ads basics page. Retailers can also run product-feed ads with prices and star ratings.
• Who sees them: people on the Free and Go plans only. Plus, Pro, Business, Enterprise and Edu users, and anyone OpenAI identifies as under 18, don't see ads.
• Where: self-serve Ads Manager is open in about 60 countries, including the US, UK, Ireland, Canada, Australia and the EU, per OpenAI's availability list.
• Minimum spend: a minimum daily budget of $25, £15 or €15. No upfront commitment.
• How you buy: Views (CPM), Clicks (CPC) or a Conversions objective that still bills per click or per 1,000 impressions, never per order. OpenAI suggests a $3–$5 maximum CPC to start.
• Who should test now: brands in allowed categories selling considered, higher-ticket products, whose first order leaves roughly $50 or more of contribution margin, and who can fund a few thousand dollars of testing.
• Who should wait: low-AOV, thin-margin brands, and anyone whose Meta and Google spend isn't profitable yet. OpenAI's ad policies also ban alcohol, tobacco and vapes, sexual health products, THC and gambling, approve US dietary supplements case-by-case, and say health services ads outside the US are "generally prohibited".

What's confirmed, and what isn't

The timeline, from OpenAI's own posts: ads started testing in the US on 9 February 2026, a week after Adweek reported a $200,000 minimum commitment for pilot advertisers. Self-serve Ads Manager and CPC bidding arrived on 5 May, 31 European markets went self-serve on 31 August, and by 23 September ads were in more than 60 countries, with OpenAI reporting a $1 billion annualised revenue run rate. On 5 October it announced image ads during image generation, testing in the US this month, plus new attribution and incrementality partners.

What isn't confirmed is price. Launch-week CPMs of $60 and later reports of $25 come from press and single accounts, and OpenAI's own FAQ says it "does not yet have performance benchmarks across advertisers, industries, or campaign types". Two other limits worth knowing: some new self-serve accounts can only advertise in their home country until they pass verification and an unpublished spend threshold, and personalised ads and custom audiences aren't available in the EEA or Switzerland, per OpenAI's campaign guide.

The profit maths: break-even CPA in dollars

Break-even CPA = contribution margin in dollars per first order, before ad spend. That's net sales on the order minus product cost, shipping, fulfilment, payment fees and expected returns. If a $120 first order leaves you $54, you can pay up to $54 to win that customer and break even on the first purchase. Our contribution margin formula guide shows what belongs in that line.

The percentage is the guardrail, not the headline. Break-even ROAS = 1 ÷ your contribution margin ratio, so $54 on a $120 order (45%) means $2.22 of revenue per $1 of spend just to stand still. Plug in your own numbers with the break-even ROAS calculator.

Same channel, same auction, very different decisions: a brand with $36 of margin per order needs ChatGPT to return $3.33 per $1 on the first purchase, while one with $72 can break even at $1.67.

Why we measure contribution margin, not ROAS

ROAS divides revenue by ad spend, so it ignores everything else an order costs you: product, shipping, packaging, payment fees and returns. A 3x ROAS on a $36-margin order loses money; on a $72-margin order it's comfortably profitable. ROAS also comes from each platform's own attribution, and those overlap: ChatGPT, Meta and Google can all claim the same order, so adding up platform ROAS overstates what you earned. Contribution margin in dollars, measured on new customers and checked against blended numbers, is the only figure that can't double-count. We've made the longer case against attribution before.

The conversion rate you need at today's bids

If you buy clicks, break-even conversion rate = CPC ÷ contribution margin in dollars per first order. At $3 a click and $54 of margin, 5.6% of the people who click must buy on their first visit for the first order to pay for itself.

Compare that with your own conversion rate from cold paid traffic in Shopify or GA4. If yours is half the number in your cell, a test only works if real CPCs clear well below the $3 starting bid (the auction is second-price, so you usually pay less than your maximum), your margin per order is higher, or repeat purchases bring the customer into profit within your CAC payback period. OpenAI says product-feed ads have been "among the strongest-performing ads in our program to date" in its product feed guide, so retailers should start there, while treating that as OpenAI's claim rather than a benchmark.

What a readable test costs

Test budget = target orders ÷ conversion rate × CPC. We use 30 new-customer orders as a working minimum before reading a channel's CPA. It's not a statistical guarantee, but below it a few lucky or unlucky days decide the result.

At a 2.5% conversion rate and $2.50 a click, 30 orders needs 1,200 clicks and $3,000, or $100 a day for 30 days. The $25 daily minimum only buys $750 a month, which at $3 a click and 2.5% is about 6 orders: a look, not a read. If you buy on CPM, convert with CPC = CPM ÷ (1,000 × CTR). All of these inputs are illustrative, because no ChatGPT benchmarks exist yet. At the time of writing, ads.openai.com also offers new advertisers $500 of credit when they spend $500.

A 30-day ChatGPT ads test plan

These are StoreHero's rules of thumb, not OpenAI guidance. Write them down before day 1, because founders who set the rules afterwards usually find a reason to keep spending.

1. Set the bar. Work out break-even CPA in dollars from your first-order contribution margin, and size the budget for about 30 new-customer orders.
2. Install measurement first. Add the OpenAI pixel and, ideally, the Conversions API (the same server-side principle as our Shopify tracking guide), and tag every ad with UTMs such as utm_source=chatgpt and utm_medium=paid.
3. Pick one campaign. A product-feed campaign if you have a catalogue; otherwise a Clicks campaign with a few ad groups and context hints describing the conversations you fit. Use a campaign-total budget so spend can't drift.
4. Hold out a region. For a standard US campaign, target half your states (matched on past new-customer orders) and compare new-customer orders against the other half. Feed campaigns only target countries, so split by country instead. At a few thousand dollars this is a sanity check, not proof.
5. Days 1–7: check delivery, not results. Confirm ads are serving and pixel purchases match UTM-tagged Shopify orders. Ads Manager spend can lag 7–8 hours. If you've spent 2× break-even CPA with no purchases recorded, check tracking and the landing page before blaming the channel.
6. Measure the right things. New-customer CAC in dollars (spend ÷ first-time customers from tagged orders), first-order contribution margin in dollars after ad spend, and blended new-customer CAC and MER before and during the test. Not platform ROAS: ChatGPT's own reporting is still in beta, and ROAS ignores your costs and the attribution overlap with Meta and Google.
7. Day 30, kill: new-customer CAC above 1.5× break-even CPA, nothing in the holdout, and blended new-customer CAC got worse.
8. Day 30, extend 30 days: CAC between 1× and 1.5× break-even and your repeat data says those customers pay back inside your target window. Use the time to test a better landing page or offer.
9. Day 30, scale: CAC at or below break-even, blended MER flat or up, and the holdout agrees. Raise budget 20–30% a week and re-check CAC at each step, because cheap early inventory rarely stays cheap.

A worked example in dollars

Say your first order averages $120 and leaves $54 of contribution margin before ads. You spend $3,000 over 30 days and get 40 UTM-tagged orders, of which 32 are first-time customers. New-customer CAC is $3,000 ÷ 32 = $94, and those 32 first orders earned 32 × $54 = $1,728 of contribution margin, so the new customers cost you $1,272 more than their first orders earned (the 8 existing customers would probably have bought anyway). Platform reporting would credit ChatGPT with all 40 orders, a 1.6x ROAS, which says nothing about whether those orders were new or profitable. At a $54 break-even CPA, a $94 CAC is 1.7× break-even, which is heading for a kill under the rules above unless the holdout shows a real lift and your repeat data says those customers earn the gap back. The same 32 customers at a $72 margin would have cut the first-order loss to $696, which is why margin per order decides this channel before the bid does.

This is where StoreHero earns its place. It puts blended MER, new-customer CAC and contribution margin in dollars by channel on one daily P&L, so if you add ChatGPT spend as an "other marketing" cost it lands in the same blended numbers as everything else, and the Spend Advisor shows the marginal profit of the next $1 on your connected channels, which is the bar a ChatGPT test has to clear.

ChatGPT ads vs Meta and Google

ChatGPT's edge is intent: people describe what they want, their constraints and their budget in full sentences. OpenAI's partner-reported results are encouraging (Triple Whale found 93% of Portland Leather's visitors from ChatGPT ads were new), but they're vendor-selected, and buyers quoted by Digiday still complain about thin measurement, few formats and too little inventory for big budgets. Treat ChatGPT as a test funded from the same new-customer CAC target as everything else, not a replacement for channels you can already measure.

Paid ads are also separate from ChatGPT's organic shopping. OpenAI's merchant page now says shoppers are sent to your own site "with no fees for purchases made on your site or app", a change from the fee merchants paid on in-chat Instant Checkout purchases at launch, so check how many orders ChatGPT already refers for free (our Claude and ChatGPT for Shopify guide covers the setup) and tag paid traffic so a test can't claim them.

The verdict

If your first order leaves $50 or more of contribution margin, you sell a considered product in an allowed category, and you can fund about $3,000 of testing without starving what works, run a 30-day test now and judge it on new-customer CAC and contribution margin in dollars. If your margin per order is thin, your AOV is low or your current channels aren't profitable, wait: the channel isn't going anywhere, and every month brings better measurement.

Key takeaways

• As of October 2026, ChatGPT ads are self-serve in about 60 countries, shown only to Free and Go users, with a $25 (£15, €15) minimum daily budget.
• Break-even CPA is your contribution margin in dollars per first order before ads; the ratio is the guardrail, not the headline.
• At OpenAI's suggested $3 starting bid, a $120 order with $54 of margin needs a 5.6% conversion rate to break even on the first purchase.
• Size the test for about 30 new-customer orders: orders ÷ conversion rate × CPC, usually around $3,000, not $750.
• Judge it on new-customer CAC and contribution margin in dollars, not platform ROAS, which ignores your costs and double-counts across channels.
• Organic ChatGPT shopping is separate from paid ads: tag paid traffic so a test can't claim orders you already get for free.

Frequently asked questions

Sources

• OpenAI: Testing ads in ChatGPT (updated 11 August 2026)
• OpenAI: New ways to buy ChatGPT ads (5 May 2026)
• OpenAI: ChatGPT Ads expands across Europe (18 and 31 August 2026)
• OpenAI: ChatGPT Ads expands to Southeast Asia and Taiwan (23 September 2026)
• OpenAI: Building advertising for the way people use AI (5 October 2026)
• OpenAI Help Center: Ads in ChatGPT, the basics
• OpenAI Help Center: Create campaigns for ChatGPT Ads (minimum daily budgets)
• OpenAI Help Center: Ads Manager availability
• OpenAI Help Center: Create campaigns from product feeds
• OpenAI Help Center: ChatGPT Ads frequently asked questions
• OpenAI: Ad policies (updated 10 September 2026)
• Adweek: OpenAI confirms $200,000 minimum commitment for ChatGPT ads (31 January 2026)
• Digiday: OpenAI moves to make ChatGPT ads more measurable and more visual (5 October 2026)
• OpenAI: Buy it in ChatGPT, Instant Checkout and the Agentic Commerce Protocol (29 September 2025)
• ChatGPT: Power product discovery in ChatGPT (merchants)