Advertising inside the answer.
ChatGPT now carries advertising, and the self-serve Ads Manager is open. The auction is young, the competition is thin, and that combination does not last long.
- CPM and CPC
- Geographic and custom audiences
- OpenAI pixel and Conversions API
- Ads Manager in beta
- October 2026
Early auctions are cheap auctions.
Every ad platform has the same life cycle. It opens, almost nobody uses it, the few who do buy attention far below what it is worth, and then the rest of the market arrives and the price settles. Google Ads had that window. Meta had it. Most businesses heard about both several years too late.
ChatGPT Ads is in that window now. The self-serve Ads Manager is in beta and rolling out gradually, the European rollout is recent, and the overwhelming majority of UK businesses have not placed a single pound into it.
That is not a reason to move your whole budget. It is a reason to take a measured position early, learn the platform properly while mistakes are cheap, and be competent in it before your competitors have started.
We will say plainly that this is a young platform. Formats are limited, reporting is less mature than Google’s, and OpenAI controls delivery decisions.
Anyone telling you it will replace paid search this year is selling you something.
The audience detail most people miss.
Ads in ChatGPT are shown to people on the free and lower tier plans. Plus, Pro and Enterprise subscribers do not see them at all.
That single fact should shape your entire approach. If your buyer is a senior decision maker at a company that pays for AI tooling, a meaningful share of them are outside this inventory. If you sell to consumers, small businesses or early career professionals, you are looking at exactly the right audience.
We work that out before spending anything, because it is the difference between a channel that fits your business and one that quietly burns budget on people who were never going to buy.
Deliverables.
Platform fit assessment
An honest read on whether your buyer sits inside the ad-supported tiers, before any money is committed. Sometimes the answer is no.
Account and pixel setup
Ads Manager configured, the OpenAI pixel installed, and the Conversions API wired so you are measuring real commercial events rather than clicks.
Campaign build
Geographic and custom audience targeting, CPM or CPC chosen against your objective, budgets and pacing set deliberately rather than by default.
Creative for an answer surface
Ads here sit beneath a generated response, in a reading context, not a feed. That demands a different tone from a search ad or a scroll stopper.
Cross-channel reconciliation
Results read against your Google Ads and Meta performance, so you can see what this channel genuinely adds rather than what it claims.
A kill criterion
Agreed before launch. The spend level and the time period at which we stop if it is not working. Written down, so nobody has to argue about it later.
Who this suits, and who it does not.
It suits businesses whose buyers ask questions before they buy. Anything considered, comparative or advisory, where somebody is likely to describe a problem to an assistant rather than type two words into a search box. Professional services, software, health and finance, specialist suppliers, anything with a research phase.
It suits businesses that can tolerate learning. This is an early platform, the reporting is thinner than Google or Meta, and some of what you spend in the first month buys information rather than customers. If that is uncomfortable, wait six months and pay more for the same placement.
It does not suit impulse purchases, pure price competition, or anyone who needs every pound attributed before they will spend it. It also does not suit a business with no working measurement, because you will not be able to tell whether it worked, and we would rather fix that first.
Right for considered purchases with a research phase.
Wrong for anyone who needs full attribution before they will spend.
What a real test costs.
A test has to be big enough to escape noise. Too small and you learn nothing, then conclude the channel does not work when what actually happened is that you bought a sample size of eleven.
We size the test from your own numbers rather than a standard figure: what a customer is worth, roughly what you pay per acquisition on your best existing channel, and how many conversions you need before the answer means anything. That calculation usually lands on a figure you can commit to for a month without it mattering if it fails, which is the right size for an experiment.
Budget is not the only cost. Creative and landing pages take time, measurement has to be set up before anything runs, and somebody has to read the data weekly. A test with a budget and no attention attached to it is a donation.
The test is sized from your economics, not from a price list.
Big enough to be readable. Small enough that failing is fine.
How a 30 day test runs.
Before anything spends
Measurement first. The OpenAI pixel and the Conversions API are installed and verified against a real conversion, because an unmeasured test produces an opinion rather than a result.
Days 1 to 7
A deliberately broad launch. The point of week one is to find out who responds, not to be efficient. We change as little as possible while the platform collects.
Days 8 to 14
First read. Which audiences and placements produced anything, which produced nothing, and whether the cost per action is within sight of your other channels. Usually one or two clear cuts.
Days 15 to 28
Concentrate the budget on what worked and test new creative against it. This is where most of the learning happens, and where the decision for month two gets made.
Day 30
A straight answer. Continue, continue with changes, or stop. Written down with the numbers behind it, including the case where the honest recommendation is to stop.
What you can and cannot know.
You can measure clicks, conversions tracked by the pixel and the Conversions API, and cost per action within the platform. That is enough to run the channel and enough to decide whether it is paying.
What you cannot do yet is see the full picture you are used to from Google. The reporting is younger and less granular, view through effects are hard to isolate, and attribution across an assistant conversation is a genuinely unsolved problem rather than something a better setup would fix.
Our approach is to hold a view of total enquiries alongside platform reported conversions. If platform conversions rise and total enquiries do not, something is being claimed rather than created. That check catches most of what attribution hides, and it works on every channel, not just this one.
Platform numbers alongside total enquiries, always.
If one rises and the other does not, believe the second one.
Against Google Ads and Meta.
Google Ads captures demand that already exists and is the best channel in the world at it. It is also fully priced. Everyone knows what a click is worth and the auction reflects that, so the opportunity is in execution rather than in arbitrage.
Meta creates demand rather than capturing it, which makes it right for some businesses and expensive for others. Its strength is reach and creative, its weakness is that intent is borrowed rather than real.
Advertising inside an assistant sits between the two. The intent is closer to Google, because somebody has described a problem in their own words. The pricing is closer to early Meta, because not enough advertisers have arrived yet. That gap is the entire reason to be here now, and it is temporary by definition.
Google: real intent, full price.
Meta: borrowed intent, cheap reach.
Assistants: real intent, early price. For now.
Straight answers.
Is this actually available to UK businesses?
Yes. ChatGPT Ads rolled out across European markets during 2026 and UK advertisers can buy. Access to the self-serve Ads Manager is still being rolled out gradually, so availability can vary by account. We check your specific access before promising a start date.
How much should I put into it?
Less than you think, to begin with. This is a learning budget, not a scaling budget. We would rather establish whether the channel produces qualified enquiries for your business on a small, controlled spend than discover the answer expensively.
Can I target specific keywords like Google Ads?
Not in the same way. OpenAI’s system controls delivery decisions, and the levers advertisers get are geographic targeting, custom audiences, bidding and creative. If you are expecting keyword-level control, this platform will frustrate you, and it is better you know that now.
How do I know whether it worked?
Through the OpenAI pixel and Conversions API, tracking real actions such as enquiries and purchases rather than impressions. Reporting is aggregated, so you will not see individual conversations. We reconcile what the platform reports against what actually lands in your CRM, the same as we do for Meta.
Should I do this instead of Google Ads?
No. Google Ads captures people actively searching with commercial intent and remains the stronger channel for most businesses. ChatGPT Ads is an addition while it is cheap, not a replacement. Any agency suggesting otherwise is chasing a headline.