ChatGPT Ads has changed more in a month than most platforms change in a year

Back in February, ChatGPT Ads was a closed room for enterprise budgets: a small group of advertisers, six-figure spend commitments and CPMs of around $60. When we got into the beta in June, we'll be honest: it still felt like a testbed. Targeting was thematic rather than keyword based, there was no query volume data, and it seemed like OpenAI was studying advertisers as closely as users. We were happy to be the guinea pig, and nobody was expecting it to grow up this fast.

The last four weeks

On 31 August, OpenAI said ChatGPT Ads had passed a $1 billion annualised revenue run rate in under 200 days, with tens of thousands of advertisers on board. The same day, self-serve buying opened across India, Europe, the Middle East and North Africa. It also said CPC and outcome-optimised bidding now make up the majority of campaigns, and that more than 50 technology and measurement partners are plugged in.

On 16 September the product itself moved on. OpenAI started testing "Sponsored Agents", where a user can click an ad and then chat with a brand's own agent, and added an opt-in feature that adapts ad copy to the conversation. HubSpot and Shopify came on as the first CRM and ecommerce partners, with the Shopify app pulling products straight from Shopify Catalog and opening internationally from 23 September.

Then on 23 September it rolled out across seven more Asian markets, taking it to more than 60 countries.

That's a closed pilot to a global, self-serve, commerce-connected channel in the space of a month! We have been in paid media long enough to know that doesn't happen often.

Contextual targeting is getting very sharp, very fast

This is the genuinely fascinating bit for e-commerce.

Traditional contextual reads a webpage. ChatGPT reads a live conversation. OpenAI says its system looks at the context of the current chat and, where personalisation is switched on, selected signals from the user's wider ChatGPT activity. Ads are chosen on expected relevance and outcomes through a relevance-weighted second-price auction. There are no keywords to bid on. Advertisers write "context hints" that guide the matching rather than acting as hard targeting rules.

Compare that to June. "Vaguely thematic" has become a system that gets fed more outcome data every week. Conversion-optimised campaigns now steer delivery towards a specific standard event, like a purchase or a sign-up, and OpenAI describes Pixel and the Conversions API as core foundations for measurement and optimisation. With copy now able to flex to the conversation too, the creative is learning alongside the targeting. Put simply, the platform is learning with Terminator-like efficiency: every conversion that flows back sharpens which ad it serves in which conversation.

For UK and European brands there's an important nuance. Personalised ads aren't initially available in the EEA or Switzerland, and users who turn personalisation off still see ads based on their current chat. OpenAI also advises against custom audiences for campaigns aimed at those markets for now. So across much of Europe, contextual relevance isn't one lever among many. It's the whole engine, and the brands that get good at writing context hints and conversation-led creative now will be a long way ahead when everyone else turns up.

Why retail brands would be leaving money on the table not starting now

Start with the shopper. Adobe reports that traffic from AI sources to US retail sites grew 393% year on year in Q1 2026. In March it converted 42% better than non-AI traffic, a record high, with 12% higher engagement. By May, AI-referred retail traffic was up 138% year on year and converting 54% better than non-AI sources. To be clear, that's all AI referral traffic, mostly organic, and it's US data. But it tells you where shoppers are going and how ready they are to buy when they get there.

Then we must look at the toolkit. OpenAI points to product feeds, platform and geographic targeting, and custom audiences as the tools helping advertisers reach the right people. The early results it's sharing are encouraging: one ecommerce advertiser hit a 3x return on ad spend over 28 days, and a technology partner found more than 80% of ad-driven ChatGPT traffic came from new customers.

And then there's timing. Peak is weeks away. A young auction, an optimisation engine that improves with every conversion, and more than 1 billion weekly active users all point the same way. What you learn in October shapes what you can scale in November and December. Wait for mature benchmarks and you'll be buying into a busier, pricier auction, having skipped the learning phase everyone else has already paid for.

Let's not get carried away

This isn't a magic channel, and we'd rather say that upfront. Search Engine Journal's reporting, picked up by MediaPost, found some advertisers paying under $3 a click and others $10 to $13, with results all over the place and still no benchmarks across advertisers, industries or campaign types. First Page Sage found ChatGPT ad conversion rates were generally lower than organic search in most of the industries it looked at.

That's exactly why this is a test-and-learn channel, not a go-big-blind one. The retailers getting it right are:

• Setting clear success metrics upfront
• Getting Pixel and Conversions API tracking live before spending a pound
• Plugging in clean product feeds
• Writing context hints around real customer questions, not keyword lists
• Treating the first six to eight weeks as paid learning

Back in June we said we'd rather absorb the uncertainty on our own site than learn on a client's budget later. That's still the right instinct. The retailers who start testing now, while the stakes are low, will understand this platform by the time everyone else decides it's worth a look.