OpenAI advertising · beta
OpenAI advertising, still in beta. I handle access, campaigns and measurement.
Two objectives so far, Reach and Clicks, a second-price auction weighted by relevance, and an announced expansion across 31 European countries, France and Spain included.
Google Partner · 5/5 on Malt (46 reviews) · 100+ clients supported
Book your consultationThe shift
On Google, your client types three words and you bid on them. In ChatGPT, they lay out the whole situation: team size, budget, what they already tried. Then they read an answer. Your ad shows up at that exact moment, below that answer, if it matches the thread. The trigger is no longer a keyword, it is a need already spelled out.
See the Google Ads side of this →which CRM for a 10-person sales team on a tight budget?
For ten sales reps on a contained budget, three families of tools hold up, depending on what you want to automate first.
The platform today
Four facts published by OpenAI, worth knowing before you open an account.
Facts published by OpenAI (Ads Manager documentation, developer documentation and the European expansion announcement), read on 1 September 2026. These are not campaign results: the platform is too new for a performance average to mean anything, and I will not publish one before I have the hindsight that makes it honest.
Below the answer, never instead of it, and only when the subject of the conversation matches. So the work is deciding which conversations your brand adds something to, then writing a message that holds up next to an answer the person has just read.
Tracking wired before the first euro is spent: pixel, server-side conversions, event setup, then reports broken down by product, country and device. On a new platform, measurement is not a final step, it is what makes the test readable at all.
Total or daily budget, a maximum bid set at ad group level, and a second-price auction weighted by relevance. Put plainly, the tallest bar does not always win: raising your bid does not fix a message that misses the point.
OpenAI reviews both before delivery is allowed, with human escalation on sensitive cases. A page that is vague about who is selling what does not just cost conversions, it stops the ad from going live.
Who I am
Me, and no one else. No intern, no outsourcing, no account split between three juniors taking turns. You talk to the person who touches the bids.
Five years of Paid Media, including two years at Mediabrands, a media group in the global top 6, where I supported more than 40 companies. Now freelance, working directly with my clients.
Choosing well
On a platform that opened a few months ago, seniority proves nothing. Here is what does.
"ChatGPT Ads expert", "ChatGPT Ads consultant", "OpenAI advertising specialist": these titles are appearing on hundreds of pages right now, for a platform that opened in beta a few months ago. Nobody has three years of experience on it, myself included, and anyone who implies otherwise has already told you something useful about themselves. Here are the four checks that still work when seniority cannot be the proof.
OpenAI documentation is public and dated. Ask for the source of any claim about objectives, bidding or targeting, and the date it was read. On a platform that changes every month, an answer from memory is worth less than a sourced one, and someone who names their source hands you the means to prove them wrong.
A new platform has no settled best practice yet. What transfers are the habits that hold everywhere: structure so you can learn, wire measurement before you spend, isolate what you are testing, judge on cost per result rather than volume. Ask how they plan to learn, not what they already know.
In many setups, the person who sells the work is not the one who launches the campaigns. Ask who touches the account, how often, and with what seniority. Here the answer is simple: me, and no one else.
With OpenAI, the advertiser account is created by the business itself, then people are invited into a role. Your consultant works inside your account, not the other way round. OpenAI advertising terms also require any intermediary to hold a written agreement with the advertiser: that clause protects you, it is not paperwork.
Applied to my own profile: Google Partner certification you can check in the official directory, five years of Paid Media including two at Mediabrands, more than 100 businesses supported, and a single person operating your account. On ChatGPT Ads specifically, I will tell you exactly where I stand during the call, rather than display a track record the platform has not let anyone build yet.
In practice
Checking that your business can run at all before building anything: some categories need enhanced advertiser verification, others stay ineligible. In Europe, access currently runs through the OpenAI teams and their partners, with self-serve announced for later.
Campaigns, ad groups, ads. A campaign carries one objective and a budget, ad groups carry the bids and the message variants. Structure is decided upfront because it is what you will be able to read in the reports three weeks later.
The Reach objective, paid per 1,000 impressions, or the Clicks objective, paid per valid click, with a maximum bid at ad group level. The auction is second-price and weighted by relevance: the message counts as much as the amount.
Building and maintaining custom audiences, location targeting, and product feeds for catalogues. All of it kept readable once several campaigns run in parallel.
Pixel, server-side conversions with their own key, event setup, then reading reports by product, country and device. This is the part I handle first: a test launched without measurement does not give you partial data, it gives you nothing you can use.
Preparing ads and landing pages for the automated review, respecting the rules on advertiser identity and trademark use, and following up on rejections to fix the cause instead of resubmitting the same ad.
To avoid
What the structure of the platform makes possible, and what it punishes immediately.
On Google, someone types a query. Here, the ad appears under a response when it matches the conversation. Copying a Search account structure word for word means optimising a mechanism the platform does not use, and concluding far too early that the channel does not work.
The platform exposes a pixel and server-side conversion sending. Launching before both are in place funds a learning phase nobody will be able to read afterwards. On a platform with no public benchmarks, this is the expensive one, because it makes every other decision impossible to defend.
A minimum daily spend applies per campaign, and creating many campaigns immediately spreads an already small budget across lines that will never reach a readable volume. A few campaigns that learn beat many that hesitate.
OpenAI reviews the ad and the page it sends people to. A landing page that does not clearly say who sells what can hold up the whole rollout, and the lost time is counted in days where nothing runs.
How it works
A simple, smooth process is what produces results.
By your side at every step, for an experience without friction.
Book your consultationChecking whether your business is eligible, opening the advertiser account in your name, assigning roles, and framing what the platform actually allows for your market today. You leave with a clear decision: we start now, or we wait for one specific milestone.
Building campaigns and ad groups around what you want to learn, then wiring conversion tracking on the site and server side. We secure the reading before we raise the spend.
Testing messages and audiences on a deliberately narrow scope, reading reports by product, country and device, then scaling at the pace your margins allow. What you cannot read, you cannot scale.
Trajectories
Representative trajectories from the accounts I support, with no named data.
Catalogue connected through a product feed, a Clicks campaign on a narrow scope, server-side measurement to reconcile with real orders.
Clicks objective to a dedicated page, conversion events set on the qualified lead rather than on the form submission.
Reach objective on selected markets, with an impact measure designed before launch instead of reconstructed afterwards.
Client reviews
Reviews published on Malt and LinkedIn, with the company logo: you can trace them back.
He's incredibly skilled, proactive, and always goes the extra mile to deliver top-quality work. His deep knowledge of Google Ads and his strategic mindset make him a key asset in any project. Théo is not only highly professional and detail-oriented, but also genuinely attentive and pleasant to work with.
Theo and I worked very closely to deliver ads management services for our clients. Theo is very collaborative and produces very high quality work. Theo is a great partner and is easy to work with. I highly recommend Theo.
Read the client reviews (5/5) →
These engagements cover Google Ads and Meta Ads: see my freelance Google Ads work.
FAQ
Everything businesses ask me before opening a ChatGPT Ads account.
Yes. OpenAI announced an expansion covering 31 European countries, France and Spain included. One nuance matters: the self-serve Ads Manager does not open to everyone at the same time. Access runs first through the OpenAI Ads Solutions team and their agency and technology partners, with self-serve announced for later.
This is the cheapest moment to position yourself in terms of attention: few advertisers, so little competition on the inventory, and a lot to learn before the best practices become public.
Below ChatGPT responses. OpenAI describes one or more ad units shown under a response when there is a relevant match with the chat, and a unit may feature one or more items, from one or several advertisers.
The practical consequence matters more than the format: it is not an isolated query that triggers the ad, it is the subject of the conversation. So the message has to add something at that point of the discussion, not repeat a Search headline.
Two objectives today: Reach, which optimises impressions and is paid per 1,000 impressions, and Clicks, which optimises clicks and is paid per valid click. The available models are CPM, CPC and oCPC.
The maximum bid is set at ad group level, and delivery is decided by a relevance-weighted second-price auction. In other words, a more relevant ad can win against a higher bid: working on the message has a direct effect on what you pay.
On the platform side, a campaign runs on a total or a daily budget, with a minimum daily spend per campaign, set at $25 as of 1 September 2026. That is a technical floor, not a recommendation: the useful budget depends on what you want to learn and how fast.
On my side, I do not publish a rate card. Scope changes everything, so I would rather understand your situation on a call and then give you a number.
The platform exposes a pixel, server-side conversion sending with its own key, and event settings. Reporting returns performance at ad account, campaign, ad group and ad level, with breakdowns by product, country and device.
On a platform with no public benchmarks, measurement comes first. Until a spend can be tied to a result, no performance number deserves comment, and certainly not a comparison with Google or Meta.
That gets checked before anything is built. Some categories, financial services for example, may require extra safeguards such as enhanced advertiser verification or manual review. Other activities, including healthcare and weapons, may be ineligible.
Ads and landing pages go through an automated review before they can run, with human escalation depending on how sensitive the case is. It is the first thing I look at: there is no point building a setup that cannot be served.
Not in the European Economic Area or Switzerland: OpenAI states that personalised ads are not available there at launch. The match is made with the subject of the current conversation, not with a profile built from history.
For a European advertiser that changes how you think about targeting: you are not betting on a reconstructed audience, you are betting on how relevant your message is to a need expressed right now.
No, and I would be careful with anyone telling you otherwise today. Google Ads captures an intent typed as a query, Meta captures an audience while scrolling, ChatGPT captures a moment in a conversation. Three different surfaces, with very different measurement maturity.
My position: the budget for a ChatGPT Ads test should not come out of what already performs. Treat it as a learning investment, with an amount you accept you may not recover and one precise question it has to answer.
It depends on three things: the scope (opening and framing, ongoing management, or coaching your team), the number of markets, and the state of your measurement at the start, since everything else rests on it. Rather than a fixed rate card, I would rather understand your situation on a call and propose a precise scope with a matching price. You know exactly what is included before you commit.
Answer 7 quick questions and let’s plan a 25-minute call, no strings attached.
