Running ChatGPT Ads: hands-on setup for B2B
Step by step to your first ChatGPT Ads campaign: setup, context hint, creatives — and the tier question that decides the channel.
Since 1 September 2026 you can advertise inside ChatGPT
OpenAI opened self-service access for European advertisers on 1 September 2026. The Ads Manager is listed as available for Germany, Austria and Switzerland. Booking runs via ads.openai.com, alternatively through the OpenAI Ads Solutions Team or agency partners. The backstory: sponsored placements were first tested in ChatGPT in February 2026, the self-serve Ads Manager beta opened on 5/6 May 2026 for US advertisers, and from 24 August 2026 the rollout ran across 31 European countries.
Every figure in this guide is as of 09/2026. The platform changes fast, so check limits and availability against the interface before you launch.
This guide is a do-it-yourself walkthrough: work through it top to bottom and you can put a campaign live with OpenAI’s own tooling. We at CegTec have measured no results of our own on this channel and therefore publish no benchmark numbers that do not exist.
Step 0: Answer the tier question first
Before you create an account, settle the one question that decides the value of the channel:
Ads are served only to users on the Free and Go tiers. Anyone subscribed to Plus, Pro or Team sees no advertising.
That is not a footnote, it is the central filter for B2B. A significant share of the decision-makers you want to reach in upper mid-market B2B works on a paid subscription, often a company-provided Team licence. You simply do not reach those people through ChatGPT Ads.
An honest classification follows:
| Good fit | Poor fit |
|---|---|
| High volume, broad purchase intent | Narrow enterprise ABM lists |
| SMB and self-employed demand | A few named target accounts |
| Consumer-adjacent and prosumer offers | Offers whose buyers are almost certainly on a paid tier |
| Self-serve products with a short buying path | Long committee decisions with many stakeholders |
If your target picture reads “25 corporates, by name”, this channel is not your lever. If it reads “every business under 50 employees with a specific problem”, the test is worth running.
The ad appears below the ChatGPT answer, labelled as sponsored and visually separated from the response. You are buying a slot beside the answer, not inside it.
Step 1: Assemble the prerequisites
Have these ready before you create the account:
- Email address for account access
- Business details (legal form, address, contact person)
- Tax ID or VAT identification number (EIN in a US context)
- Billing method
- Landing page that is live and compliant
The landing page is where most approvals fail. It is reviewed together with the creative. Make sure your legal notice, privacy policy and the advertised offer are genuinely findable on the page and that the page delivers on the ad’s promise.
Step 2: Create the account
Register at ads.openai.com. Larger budgets can alternatively be booked through the OpenAI Ads Solutions Team or through agency partners. For a first test you do not need that route: there is no minimum spend.
Step 3: Understand the account structure
Three levels, each with its own job:
- Campaign — you set the objective and the budget.
- Ad group — you set the bid and write the context hint of at most 280 characters.
- Ad — this holds title, copy, image and landing page.
Memorise this mapping; it explains why you cannot find your delivery controls at ad level. The control over where your ad is eligible to appear sits in the ad group.
Step 4: Choose Reach or Clicks
Two objectives with two buying models:
- Reach — billed on CPM, you pay per 1,000 impressions. Sensible when you want awareness inside a topic area.
- Clicks — billed on CPC, you pay per valid click. Sensible when you want visitors on a landing page.
For a first test in a B2B context, Clicks is usually the more honest choice: you pay only when someone actually reacts, and you get a signal you can measure on your own property.
OpenAI recommends starting with a max bid of roughly $3 to $5 per click. Reliable CPC benchmarks for Germany, Austria or Switzerland do not exist as of 09/2026, and we will not invent any. Derive your test budget from your own economics instead: what may a click cost so the channel carries itself at your known close rate?
Step 5: Write the context hint (the actual lever)
Targeting is contextual only. Delivery is based on the topic and intent of the ongoing conversation, plus prior chat history and previous ad interactions. There is no profile targeting, no cookie targeting, no classic behavioural targeting and no keyword buying in the Google Ads sense.
That shifts the work: you do not build a keyword list, you describe a conversational situation. The context hint is free text of at most 280 characters in which you say, in plain language, which conversations your ad should be eligible for. In craft terms that is closer to an ICP description than to a keyword set.
What works in practice:
- Situation, not search term. Describe what the person is doing or asking, not which word they type.
- Include the circumstances. Company size, role, country, trigger.
- List two to four intents. Comparing, checking prices, planning a switch, clarifying requirements.
- Do not overstuff. 280 characters force a choice. A precise hint with less delivery beats a broad hint with expensive spill.
Three worked German examples
Campaigns for DACH markets are written in German, so these are the originals with an English gloss.
Example A — applicant tracking software for SMBs (252 characters):
Gespräche, in denen jemand für ein kleines oder mittleres Unternehmen eine Software für Bewerbermanagement sucht, Anbieter vergleicht, nach Preisen oder DSGVO-Konformität fragt oder wissen will, wie sich Stellenanzeigen automatisiert ausspielen lassen.
Gloss: conversations where someone is looking for applicant tracking software for a small or mid-sized company, comparing vendors, asking about pricing or GDPR compliance, or wanting to know how to publish job ads automatically.
Example B — bookkeeping tool for the self-employed (246 characters):
Gespräche von Selbstständigen und kleinen Betrieben in Deutschland, die ihre Buchhaltung selbst machen, ein Tool für Rechnungen, EÜR und Umsatzsteuervoranmeldung suchen oder fragen, wie sie vom Steuerberater-Excel auf eine eigene Lösung wechseln.
Gloss: conversations from freelancers and small businesses in Germany who do their own bookkeeping, are looking for a tool for invoices, VAT returns and annual accounts, or asking how to move off their accountant’s spreadsheet onto their own system.
Example C — field service scheduling in the trades (224 characters):
Gespräche über die Einsatzplanung von Monteuren und Servicetechnikern im Handwerk: Touren planen, Termine koordinieren, Arbeitszeiten erfassen — inklusive Fragen nach Apps, die auch ohne Netz auf der Baustelle funktionieren.
Gloss: conversations about scheduling field technicians and service engineers in the trades — planning routes, coordinating appointments, tracking hours, including questions about apps that work offline on site.
All three deliberately target SMB and self-employed demand. That is not a coincidence but the direct consequence of Step 0: that is exactly where the audience sits that predominantly works on Free and Go access.
Create one context hint per ad group and test variants as separate ad groups. Only then will you know afterwards which description delivered.
Step 6: Build creatives inside the character limits
ChatGPT Ads are text-led, not visual. The image is supporting material; the text carries the ad.
OpenAI’s own figures as of 09/2026:
| Element | Recommended | Accepted |
|---|---|---|
| Title | 16–24 characters | up to 50 characters |
| Copy | 32–48 characters | up to 100 characters |
| Image | one square asset, up to 1200 × 1200 px | — |
A note on conflicting figures: third-party blogs circulate different limits, for example 50 to 80 characters for the headline and 150 to 200 characters for the description. We stick to the recommended and accepted values OpenAI itself publishes, shown above. Which number applies in case of doubt is settled by the input mask in the Ads Manager, not by a blog post.
Each ad renders: advertiser name, favicon or logo, title, copy, image and landing page. Your company name is already beside the ad, so do not waste the title repeating it.
Examples that fit the recommended limits
| Title (characters) | Copy (characters) |
|---|---|
| Quotes in 2 minutes (19) | Invoices, VAT and books in one tool. (36) |
| Plan your field techs (21) | Route planning for trades. 14 days free. (40) |
| Bookkeeping, no Excel (21) | Track applicants, GDPR-compliant. (33) |
At 16 to 24 characters exactly one thought fits. Pick the benefit, not the product category.
Step 7: Plan for review
Campaigns are reviewed manually. Most clear within 24 hours, provided creative and landing page comply with OpenAI’s ad policies. Budget that day in rather than scheduling a launch to the hour. If you have to be live for a date, submit with a buffer.
Step 8: Measure correctly despite the attribution delay
Attribution reports run on a 24 to 48 hour delay. A same-day glance at the dashboard shows you an incomplete picture and invites bad decisions. Two consequences:
- Never judge a single day. Evaluate in windows of at least a week and only compare once the delay has cleared.
- Measure additionally on your own side. Your own UTM parameters, a clean landing page goal and handover into your CRM give you a signal that does not depend on the reporting delay. For the question “did the channel produce pipeline”, your own system is the more defensible source anyway.
For a sense of channel scale: ChatGPT Ads reportedly reached a $1B annualised revenue run rate within around 200 days of launch. That says something about demand for ad inventory, but nothing about your results.
Decision checklist: is this channel worth it for us?
Answer these six questions before you plan budget. Three or more noes mean: not for now.
- Do our buyers plausibly sit on Free or Go? Smaller businesses, the self-employed, prosumers and near-personal usage argue yes. Corporate IT on a Team licence argues no.
- Is our demand broad enough? A 280-character context hint describes a class of conversations, not a name list. With twenty target accounts the channel runs empty.
- Do we have a landing page that carries the sale without a call? You are buying clicks, not meetings.
- Can we measure independently of the ads report? If not, you are buying a number you see only two days later and only partially.
- Does our economics tolerate a click price in the order of the recommended starting bid? Calculate with your own close rate, not with someone else’s benchmarks.
- Do we have two or three serious context hint variants? A single phrasing is not a test.
A cheap first test
If the checklist argues in favour, keep the first run small and cleanly separated:
- One campaign, objective Clicks, budget in an order of magnitude you could write off without discussion.
- Two to three ad groups, each with a different context hint. That is your actual test variable.
- Two ads per ad group inside the recommended limits, one benefit-led, one problem-led.
- Run at least two weeks so the attribution delay does not dictate your read.
- Evaluate on your side: clicks, landing page behaviour, qualified enquiries in the CRM.
After that you know the decisive thing: whether conversations that match your offer exist in this channel at all. That question precedes any optimisation.
What to realistically expect
ChatGPT Ads are bookable in DACH from now on, the setup takes an afternoon, and the channel is new enough that competition for the same conversations is still thin. At the same time the tier filter cuts away a relevant part of the B2B audience, and nobody quoting you defensible DACH benchmarks in September 2026 can substantiate them. Treat the channel for what it is: a testable hypothesis with a low entry cost, not a proven revenue channel.
If you need the decision logic behind the channel in a shorter form, read ChatGPT Ads in DACH: is the channel worth it?. How to appear in AI answers independently of paid placements is covered in improving ChatGPT visibility and in the AI search visibility guide.
Start a free trial · 4 weeks free, no credit card. Prefer to see it running first? Book a demo.