ChatGPT Ads in the UK:
The Honest Guide

ChatGPT ads are sponsored boxes that appear underneath ChatGPT's answer, clearly labelled and separate from the answer itself. Someone asks for help planning a weekend in Snowdonia, ChatGPT answers, and below that there might be an ad for a holiday cottage. They've been live in the UK since mid-2026, OpenAI's first market in Europe, and UK businesses can buy them directly through OpenAI's self-serve Ads Manager.

It's still officially a test. Features change every few weeks, and the flashier bits you might have read about, like chatting with a sponsored agent, are US-only for now. By October 2026, OpenAI said more than 100,000 advertisers had run campaigns, and panel data reported by Digiday suggests 45.7% of ad-eligible UK users had seen ads.

Here's the bit that matters most. OpenAI says ads don't influence what ChatGPT says, and its shopping results are picked independently and aren't ads. So you can't buy your way into the answer, only the slot underneath it. Getting mentioned in the answer is GEO, a separate, organic job.

The two do interact, though. Your ad sits under an answer that may mention you, a competitor or nobody. It's been reported that brands already cited in answers get better results from their ads. And OpenAI reads your landing page to decide where your ad fits, so the clear, plain pages that help GEO help your ads too.

You buy ChatGPT ads through OpenAI's Ads Manager at ads.openai.com. The UK is listed as available, so there's no waitlist, but the business being billed has to be based here. You sign up, verify the business and add billing (you pay after the ads run), and OpenAI reviews every ad before it goes live. Approval isn't automatic. Pick pounds and UK time when you create the account, because country, currency and time zone can't be changed later.

There are no keywords. Instead you write context hints, plain-English descriptions of the conversations your ad belongs in, like "planning a three-bed house move in North Wales". They guide the matching, they don't guarantee it, and OpenAI also reads your ad copy and landing page to decide where you fit. You can target by location, down to postcode where it's available, and by device. Custom audiences need at least 25,000 matched people, which rules out most small businesses' customer lists.

There are three objectives. Views, paid per thousand impressions (CPM). Clicks, paid per click (CPC). And Conversions, where OpenAI steers your bids towards sales or leads, but you still pay for the views or clicks, not per conversion. The auction is reported as relevance-weighted and second-price, so how well your ad matches counts, not just the size of your bid.

The minimum budget is £15 a day per campaign for accounts in pounds, with no overall commitment. That's a long way from launch, when big advertisers were reportedly asked to commit $200,000 or more. Bidding is changing too. From November 2026, new conversion campaigns use Target CPA and new click campaigns use Target CPC, and the old Max Bid and Max CPC options are on their way out (reported for January 2027). OpenAI's suggested starting bid is $3 to $5 a click, but that's US guidance, not what a click costs in Llandudno.

I've spoken to OpenAI's ads team and worked through their own launch guide. Here's their advice for a first campaign, and where I'd add a caveat.

Budget. £15 a day is the minimum, but they suggest testing at around £50 a day for two or three weeks, then moving to £100 to £200 a day only if it's working. That's sensible. A test too small to learn anything is money wasted.

Volume. Their guide says launch with at least 100 distinct, checked ads, spread across focused ad groups, each built around a single theme. Not 100 tiny rewrites of the same line, 100 different ways people ask about the same need.

Context hints. Write them as natural phrases of about four or five words describing a real situation, not keywords. "plumber, boiler, heating" is weak. "boiler stopped working in winter" or "choosing a reliable local plumber" is what they're after. Hints guide the matching, they don't guarantee it.

Everything has to match. The context hint, the ad and the landing page need to tell the same story. If they don't, the ad gets rejected or quietly underdelivers. Point ads at the most relevant product or service page, not your homepage. Ads and brands are reviewed first, and that can take a while.

The fiddly bits. Don't let your website block OpenAI's crawler, or your pages can't be reviewed. Images need to be public, square PNG or JPG links. And check your country, currency and time zone when you open the account, because you can't change them afterwards.

Delivery comes down to your objective, your bid, how relevant you are, the quality of the ad and the action you're asking for. Good pages and good copy make every pound go further, same as Google. Where I'd disagree is tracking: their guide treats it as a later job. I'd sort it before spending properly.

Measurement is where I'd take the most care. OpenAI gives you a pixel for your website and a Conversions API that sends the same events from your server. Use both, with matching event IDs so a sale isn't counted twice, and tag every ad link so GA4 knows where visits came from. Since 30 September 2026 the default attribution is 30 days after a click and 1 day after a view. Reports cover the basics: impressions, clicks, spend, costs and conversions. You never get the conversations themselves, competitor data or benchmarks.

Then expect the numbers not to match. One advertiser reported 57 clicks in Ads Manager against fewer than 20 sessions in Google Analytics, and agencies are still reporting gaps. A small gap between any ad platform and GA4 is normal. A gap that size isn't, so check Ads Manager clicks against GA4 sessions from day one. If your tracking's already wobbly, a tracking audit is the cheaper first step, and server-side tracking is the sturdier way to send conversions.

Watch for junk clicks too. TrafficGuard, which sells fraud protection so has a horse in the race, looked at 47 advertisers and found invalid click rates anywhere from 0.1% to 34%, with clicks from data centres and proxies at 3.6 times the Google Ads rate for the same advertisers. GA4 sessions with no engagement at all are the tell.

The rest are the usual risks of a new platform. Controls and reporting are thin, the rules keep changing, and it's still officially a test. Some early testers have spent real money for nothing: one reported $415 for 60 clicks and no sales, another $2,319 for no leads against two from Google. And there's trust. OpenAI says ads won't shape answers and advertisers never see your conversations, but those are its own promises, not independently checked.