How should an online store run ChatGPT ads?

The short version

An online store running ChatGPT ads should optimise toward a purchase event carrying real order value, not just a click or a page view, and should expect to run both a product feed campaign covering the catalogue and, where useful, a smaller standard campaign for offers a feed cannot express. Budget for the platform minimum only as a floor to learn on, not a number that will produce a judgable return on spend figure. This guide covers the ecommerce specific decisions; for the mechanics of building the feed itself, see our shopping ads guide and product feed guide.

Ad clickOpenAI adds a click reference called oppref to the landing page URL

Pixel in the browserFires page and conversion events and carries the click reference

Conversions API from your serverSends the same event with the same event ID, including sales that close later by phone or WhatsApp

Ads ManagerMatches the two by event ID so one sale is counted once, then reports and optimises on it

Choosing your objective as a store

The objective and billing model are fixed once a campaign is created, so decide deliberately. A store new to the platform is usually best served starting on Clicks to validate tracking, then moving to a Conversions campaign once real purchase events are confirmed firing correctly and matched to the right conversion configuration.

The purchase event, and why order value matters

A purchase event that only confirms an order happened, with no value attached, gives you a conversion count but not a return figure. OpenAI's own reporting distinguishes this directly: Order Created Sales and Order Created ROAS are calculated only from attributed order_created events, and other standard and custom conversion events do not contribute to those columns. In practice that means a store's purchase event needs to carry the order value through to OpenAI, not just fire as a binary yes or no, or the ROAS figure in Ads Manager will simply be unavailable regardless of how much revenue the campaign actually drove.

Full setup steps for the Pixel and Conversions API pairing are in our conversion tracking guide and ChatGPT ads pixel guide.

Reading Order Created Sales and ROAS correctly

Once value tracking is live, two columns in Ads Manager become the ones a store should actually watch: Order Created Sales, the revenue figure, and Order Created ROAS, spend divided against it. Both are scoped narrowly by OpenAI's own definition, built only from attributed order_created events, so a store running several conversion events, a purchase alongside a lead form for wholesale enquiries, for example, should not expect every event type to feed the same ROAS column. Check which event your campaign is actually configured against before reading a ROAS number as the full picture of what a campaign drove. Allow the same 24 to 48 hour reporting window OpenAI applies to all conversions before treating an early ROAS figure as settled, since totals can still change as recent events finish processing.

Feed campaigns versus standard campaigns, for a store

OpenAI states ads created from product feeds have been among the strongest performing ads in its programme to date, which is a reasonable starting signal for any catalogue business, though it is OpenAI's own claim about its own product rather than an independent benchmark. A feed campaign turns your catalogue into ads automatically and keeps working as stock and prices change, provided the feed itself stays fresh, since items expire after two weeks without a hosted URL or scheduled upload.

A standard campaign still earns its place alongside a feed for anything a structured catalogue field cannot say: a bundle discount, a limited time promotion, a brand story, or a hero product you want hand written copy behind rather than a template pulling straight from your store's product title. Most stores we would set up run both together rather than choosing one.

If your store also has a shopping app, mobile measurement partner integrations are worth checking before you build your own server side purchase event from scratch. OpenAI currently supports AppsFlyer, Adjust and Branch as mobile measurement partners, and Fospha, Hightouch, LiveRamp, Singular, Triple Whale and WorkMagic among wider measurement partners. Connecting an existing tool you already use is usually faster than building a direct integration, and it keeps your app and web purchase events reporting into the same data source rather than two separate ones you have to reconcile by hand.

Budgeting for an ecommerce test, worked from the published minimums

This is arithmetic on the numbers OpenAI publishes, not a promised outcome. Work backwards from how many purchases you need to judge the channel fairly, not from the daily minimum.

The cost per click and conversion rate here are assumptions used only for the arithmetic, not OpenAI figures or a promise of results. At the floor budget, three purchases in a month tells you almost nothing about whether the channel works for your store. A budget sized toward the realistic column gives you a chance at a double digit purchase count, which is closer to a number worth acting on. Put your own store's historical conversion rate into the budget calculator rather than the placeholder above, and see our cost guide for the full billing and pacing mechanics behind these AED figures.

Bidding for a store

Maximize Results is the default bid strategy for eligible new ad groups, including on a product feed campaign, and adjusts bids automatically toward your selected goal within budget. OpenAI is explicit that it prioritises result volume, not a guaranteed cost efficiency target, so a store that needs to protect margin on a specific bid ceiling should use manual bidding or Bid Cap instead. Our approach: run Maximize Results on a feed campaign covering the broad catalogue, where result volume is the point, and consider manual bidding on a smaller standard campaign promoting higher margin or hero products, where the cost per result matters more than raw volume. The full comparison is in our bidding strategies guide.

A four week test plan for ecommerce

This is our own recommended plan, not an OpenAI benchmark.

1. Week 1: wire purchase tracking with value, and validate the feed

   Install the Pixel and Conversions API sending order value, confirm a real test purchase appears correctly attributed, and get your product feed uploaded and passing validation.

2. Week 2: launch on Clicks or a modest Conversions budget

   Run the feed campaign alongside a small standard campaign for one or two hero offers, watching for tracking issues before committing more spend.

3. Week 3: scale the budget toward the realistic test size

   Once purchase events are confirmed accurate, move budget toward a level that can produce enough orders to be meaningful, following the sizing arithmetic above.

4. Week 4: read Order Created ROAS alongside your own store data

   Compare Ads Manager's reported figure against your store's own order records for the same campaigns and date range, since attribution methods and windows can differ between the two.

Common ecommerce mistakes we see

UAE and Gulf specifics

Price your feed and your landing pages in the same currency and format, since a mismatch between an ad's price and the page a UAE shopper lands on is one of the fastest ways to lose a sale that was otherwise close. If you run a bilingual store, decide whether Arabic and English ranges sit in one feed with language variants or two separate feeds before your first upload, since the ad template pulls titles and descriptions straight from whichever feed the ad group selects. Confirm your Shopify, Salla or Zid feed export refreshes on a schedule rather than relying on someone remembering to reupload it every two weeks, and check that your hosting or CDN allows OpenAI's crawlers to reach both the feed URL and the product image URLs, since a blocked image host can fail feed processing the same way a blocked landing page fails ad review. For the account setup steps before any of this can run, see ChatGPT ads in the UAE, and for the technically detailed feed specification, see our product feed guide.

Quick answers

Which objective should an online store use on ChatGPT ads?

A Conversions campaign optimised toward a purchase event, once tracking is proven. Start on Clicks first if your Pixel and Conversions API are not yet validated, since a Conversions campaign requires at least one supported standard event before it can be created.

Does the purchase event need to include order value?

OpenAI's reporting shows Order Created Sales and Order Created ROAS calculated only from attributed order_created events, so sending value with your purchase event is what makes a return on spend figure possible rather than only a conversion count.

Should a store use a product feed campaign or standard ads?

Most stores benefit from both: a feed campaign covering the catalogue broadly, since OpenAI states feed based ads have been among its strongest performers, alongside a small standard campaign for a hero offer or seasonal push a feed cannot express.

What is the minimum budget for an ecommerce test on ChatGPT ads?

It depends on billing currency, AED 65 a day for a dirham account. That floor is enough to see the platform work, not enough to judge return on spend, which needs enough purchases to be statistically meaningful.

How is oCPM different from oCPC for a store with a longer browse to purchase gap?

oCPM bills per 1,000 impressions while weighing a wider set of exposures, including a person who saw an ad, left, and purchased later without clicking it again. oCPC bills per valid click and optimises toward conversions that follow that click specifically.

Can ChatGPT ads track cart abandonment?

OpenAI has not published a specific cart abandonment feature. A store can track an add to cart or checkout started event as an intermediate action alongside the purchase event, within OpenAI's general conversion measurement setup.

How long should an ecommerce test run before judging ROAS?

Our view is four weeks minimum, sized to reach enough purchases to be meaningful, since a handful of orders either way can swing a reported return on spend figure without reflecting a real trend.

Related