ChatGPT Ads Profit Tracking: Why Your Best Campaign May Lose Money
Table of Contents
Key Takeaways
- The first large study of ChatGPT Ads (1,898 shops, nearly 10 billion visits) found paid ChatGPT visits earned $0.89 each, against $2.45 for Google Search ads — 64% less.
- ChatGPT Ads reports revenue and Sales ROAS. Revenue ignores margin, discounts, returns and returning customers, so the campaign with the best ROAS can be the one losing money.
- Work out your break-even CPC before launch: conversion rate × average order value × margin. At the study's averages it is about $0.36, while advertisers reported paying $1.72 per click.
- Track purchases server-side through the Conversions API, deduplicate with the order ID, and measure POAS and new-customer share in your own reporting.
- Run a 90-day test with kill rules agreed in advance, and watch organic ChatGPT referrals: in the study they fell 6–8% after ads launched.
What the First Large Study Found
ChatGPT Ads moved fast in 2026. It started in February as a US pilot sold on impressions, added a self-serve Ads Manager with CPC bidding, a Pixel and a Conversions API in May, conversion campaigns in early summer, Sales ROAS reporting and a European rollout in July–August, and in November Target CPA and Target CPC began replacing Max Bid in Ads Manager.
Now there is data on what that traffic is worth. Researchers at Frankfurt School of Finance & Management and WU Vienna analysed nearly 10 billion visits to 1,898 online shops between June 2025 and August 2026. For US paid traffic, the gap with Google Search is large:
Paid ChatGPT traffic vs Google Search ads
US paid traffic, study of 1,898 online shops, June 2025 – August 2026.
- Most advertisers sold nothing. 82% of the 102 advertisers with complete revenue data recorded no purchase from ChatGPT ads.
- The price is the problem. To match Google's return, a ChatGPT click would have to cost about $0.26. Advertisers reported an average of $1.72, and early recommended starting bids were higher still.
- Organic beats paid. Unpaid ChatGPT referrals earned $3.27 per visit with a 1.99% conversion rate — and they fell 6–8% after sponsored units launched.
Read these as averages of a very young market, not a verdict on your store. One large brand produced most of the US paid revenue in the sample, and the study counts tracked, attributed revenue only. But the direction is clear: you cannot judge this channel by the number Ads Manager shows you.
Why ROAS Is the Wrong Scoreboard for ChatGPT Ads
Everything ChatGPT Ads reports is built on revenue: attributed sales value and Sales ROAS. Revenue says nothing about what you kept. Two campaigns from the same shop show the problem:
| Campaign | Spend | Revenue | ROAS | Margin | Gross profit | POAS | Profit after ads |
|---|---|---|---|---|---|---|---|
| A · Bestsellers, 20% off | €2,000 | €12,400 | 6.2 | 13% | €1,610 | 0.8 | −€390 |
| B · Full-price range | €1,500 | €5,100 | 3.4 | 47% | €2,400 | 1.6 | +€900 |
On ROAS, campaign A wins and gets the budget. On profit, A loses €390 and B makes €900. The two formulas that fix this:
- POAS = gross profit ÷ ad spend. Break-even is always 1.0. Above it the campaign adds profit; below it, it loses money.
- Break-even ROAS = 1 ÷ margin. At a 30% margin you need ROAS 3.3 just to break even. At 13%, you need 7.7 — so A's "great" 6.2 is a loss.
Revenue hides three more things. Many ChatGPT users research brands they already know, so a share of "new" sales comes from existing customers you didn't need to pay for again. Revenue is counted at checkout, before returns and cancellations. And if one platform reports revenue with VAT and another without, your cross-channel comparison is wrong before you start. If this sounds familiar from Google Ads, it is the same trap we cover in how tracking errors inflate ROAS.
Work Out Your Break-Even CPC Before You Launch
ChatGPT Ads charges per click. Before you bid, calculate what a click is worth to you — your profit per visit:
| Scenario | CVR | AOV | Margin | Break-even CPC | vs $1.72 reported average |
|---|---|---|---|---|---|
| Study average (US paid traffic) | 0.37% | ≈ $240 | 40% | $0.36 | Loses ≈ $1.36 per click |
| Mid-ticket store, strong landing page | 1.0% | $180 | 45% | $0.81 | Still loses ≈ $0.91 |
| High-AOV niche (furniture, B2B kit) | 0.8% | $900 | 35% | $2.52 | Profitable headroom |
| High-AOV with repeat buyers (×1.5 LTV) | 0.8% | $900 | 35% | $3.78 | Scale candidate |
The study's average order value is implied: $0.89 revenue per visit ÷ 0.37% conversion rate ≈ $240. The margins and the other rows are illustrative — plug in your own numbers. Then use the result:
- Start with Target CPC at 60–80% of break-even while you collect data.
- Switch to Target CPA only once conversions flow reliably, and keep it below your profit per order (AOV × margin) minus a safety buffer.
- If your break-even CPC is far below what the auction asks, ChatGPT Ads is a brand or research play for you today. Budget it that way, or wait.
Who has a realistic chance right now: stores with high order values or high margins, considered purchases people research in chat (furniture, travel, software, electronics, B2B equipment), strong repeat purchase, and lead-gen businesses that know what a qualified lead is worth.
The ChatGPT Ads Profit Playbook (PDF)
Everything in this article on 12 printable pages — to hand to your team, your client or your finance lead before you spend a dollar on ChatGPT Ads.
Your download is starting. We've also sent a confirmation link to your inbox — click it to keep getting new guides. Didn't start? Click here.
🔒 No spam. Guides and marketing tips only. Unsubscribe anytime.
Set Up Tracking ChatGPT Ads Can Learn From
Ads Manager offers two ways to send conversions: a browser Pixel and a server-side Conversions API (CAPI). OpenAI recommends the Conversions API as the more reliable source and the Pixel as a complement. Conversion bidding needs events already flowing, so tracking comes first:
- Create the Pixel ID and a Conversions API keyBoth are in the Conversions tab of Ads Manager. The Ads API key you'd give a reporting tool is a different key — keep them apart.
- Send purchases from your serverUse your platform's OpenAI app or a partner connector, a server-side GTM container, or your own backend. Ad blockers and browser cookie limits can't drop server events.
- Deduplicate with one event IDWhen the Pixel and CAPI both report a purchase, give them the same event ID — the order ID works. Without it, each order counts twice and ROAS doubles on paper.
- Send values in minor units, with a currency$129.99 goes out as
12999plusUSD. Sending dollars instead of cents is the classic 100× error. - Match product IDs to your feedUse the IDs your ChatGPT product feed uses — often the same as your Google Merchant Center feed — so product ads are credited with what actually sold.
- Pass first-party identifiers and check Event QualitySend hashed email and phone where consent allows, then review Event Quality and Diagnostics before you switch on conversion bidding.
Map the standard events you need: page view, content view, add to cart, checkout started and order created for shops; lead created, registration and appointment scheduled for lead-gen; trial started and subscription created for subscriptions. Fire them only after your cookie banner's consent signal. In the EEA and Switzerland, ChatGPT Ads launched without ad personalisation — that limits targeting, not measurement.
Add the Profit and New-Customer Layer Yourself
Conversion campaigns optimise towards the number of one standard event, not its value. So keep bidding on purchases, and build the profit view on your side. You can build it with your shop data, GA4 and Looker Studio, or with a dedicated profit-tracking tool. Either way, the steps are the same:
- Calculate profit per order: product cost, minus discounts, shipping, payment fees and a returns allowance. Store it on the order.
- Flag new vs returning customers against your full order history — email or customer ID, not a cookie.
- Import ChatGPT ad spend next to Google and Meta, via the Ads API or a CSV export.
- Tag every ad with UTMs (
utm_source=chatgpt,utm_medium=cpc). Sessions from chatgpt.com without them are organic referrals — report them separately.
- POAS — gross profit ÷ spend. Healthy: above 1.0 and rising.
- New-customer share — new-customer orders ÷ all orders. Healthy: higher than your Google brand search.
- Cost per new customer — spend ÷ new customers. Healthy: below first-order profit or 90-day value.
- Profit per click — gross profit ÷ clicks. Healthy: above what you actually pay per click.
Then stop paying for customers you already have: upload your customer list and exclude existing buyers from acquisition campaigns. If a campaign's sales come mostly from returning customers, cut it or move it to a separate, lower-bid retention budget.
Run a 90-Day Test With Kill Rules
Most early ChatGPT campaigns fail quietly: small spend, no server-side tracking, a ROAS screenshot in a slide. Agree in advance what result means scale, fix or stop.
A 90-day ChatGPT Ads test
Foundations first, bidding second, decisions on profit.
- Days 1–14: tracking + feed
- Days 15–60: Target CPC, learn
- Days 61–90: Target CPA where it works
- Day 90: scale, fix or stop
- Budget: fund the test with new money or a small slice of search budget — we suggest no more than 5–10% of your Google Search spend. At a 0.5% conversion rate, 1,000 clicks buy about five orders, so plan for enough clicks to see a result.
- Structure: two or three campaigns by product theme or margin band, reusing your Merchant Center feed, with context hints written as real buyer questions and landing pages that answer the question asked.
- Comparison: at day 90, compare POAS and cost per new customer with Google Shopping at the same spend.
- 0 orders after 1,500 clicks → stop and review landing pages and offer.
- POAS below 0.5 after 60 days → stop.
- POAS 0.5–1.0 but new-customer share above 60% → keep, and judge on 90-day customer value.
- POAS above 1.0 for four weeks in a row → raise budget 20% per week.
Don't Buy Traffic You Used to Get Free
The study's most uncomfortable finding for advertisers: organic ChatGPT referrals to shops fell about 6–8% after sponsored units launched, while organic visits were worth $3.27 each against $0.89 for paid. If your paid ChatGPT sessions rise while organic chatgpt.com sessions fall by a similar amount, you are paying for customers you used to get for free.
- Record a baseline of organic ChatGPT sessions and revenue in GA4 before the test, and track both lines weekly.
- Keep investing in AI visibility: clean structured product data, comparison and FAQ content, reviews. Our guide to answer engine optimization covers how to get recommended without paying for it.
- Expect the format to change. The researchers note that OpenAI faces a trade-off between more commercial ad placements and users' trust in its answers — today's CPCs and click quality won't last unchanged.
ChatGPT Ads is early, and so is the competition for it. The advertisers who measure profit and new customers from day one will know where to scale — while everyone else is still comparing ROAS screenshots.
Frequently Asked Questions
Are ChatGPT Ads profitable for e-commerce?
On average, not yet. The Frankfurt School and WU Vienna study found paid ChatGPT visits earned $0.89 each versus $2.45 for Google Search ads, and 82% of advertisers with full revenue data had no purchases. High-AOV, high-margin and repeat-purchase businesses can still be profitable if they bid below their break-even CPC and track profit, not just revenue.
What is POAS and why use it instead of ROAS?
POAS (profit on ad spend) is gross profit divided by ad spend. Break-even is always 1.0, regardless of margin. ROAS uses revenue, so a campaign selling discounted, low-margin products can show a high ROAS while losing money.
How do I calculate my maximum CPC for ChatGPT Ads?
Multiply your conversion rate by your average order value and your margin. With a 1% conversion rate, a $180 average order and a 45% margin, break-even CPC is $0.81. Start with Target CPC at 60–80% of that figure.
Do I need the Conversions API or is the Pixel enough?
Use both, with the Conversions API as the main source. OpenAI recommends CAPI as the more reliable option because ad blockers and browser restrictions can't drop server events. Send the same event ID from both so purchases aren't counted twice.
Can ChatGPT Ads bid on profit?
Not at the time of writing. Conversion campaigns optimise towards the number of one standard conversion event, not its value. Keep campaigns on a standard purchase event and use POAS and new-customer share in your own reporting to decide where budget goes.
Do ChatGPT Ads take away organic ChatGPT traffic?
The study found organic ChatGPT referrals to shops fell about 6–8% after ads launched, while conversion rates for the remaining visitors were unchanged. Track organic chatgpt.com sessions in GA4 before and during any paid test to see whether you are buying traffic you already had.
Testing ChatGPT Ads?
Maksym sets up server-side tracking and profit reporting that ad platforms can optimise on. Send him your site — he’ll tell you whether ChatGPT Ads can pay for your store and what your tracking is missing.


