ChatGPT Ads for B2B: What We Learned Setting Up a Campaign

Since the end of August, businesses in Germany can run ads in ChatGPT. In September we built the channel for a B2B service provider — from the first eligibility check to the enquiry in the CRM. What the project looked like in detail is in the ChatGPT Ads case study. This post is about what you can take from it for your own setup.

One thing up front: you will not find campaign figures here. Why is explained further down — and the reason matters more than it sounds.

Eligibility first, everything else second

The most common mistake with a new ad channel is starting with the exciting part: writing ads, designing a landing page, planning budget. With ChatGPT Ads, that is the wrong order.

In the initial phase, OpenAI focuses on consumer categories — household and lifestyle, local services, travel, digital products and education. Other categories are not generally foreseen at launch. A B2B service fits none of them clearly. Which industries are excluded altogether is summarised in ChatGPT Ads in Germany: who can advertise.

So we named the risk openly before touching anything, and made a deliberate decision: one truthful submission. No workaround variants if it were rejected — the policies explicitly prohibit trying to get around policies or enforcement anyway. The ads were approved. That is a single outcome, not a rule. For you it means: plan the review in before you put time into a setup that may never go live.

Context hints replace keywords

In ChatGPT you do not bid on search terms. Instead, so-called context hints describe the conversational situations in which your offer is useful. The system then decides on delivery.

That calls for a different way of thinking. The question is not "What does my audience search for?" but "In which situation does someone ask ChatGPT for advice where my offer is the answer?". Good context hints describe situations, not product features.

Then there is structure. OpenAI applies a minimum daily budget per campaign, €15 in the euro area. Spread a test budget across many ad groups and every one of them gets too little data. We therefore kept the campaign deliberately lean: few ad groups, clearly separated situations.

The landing page is reviewed too

OpenAI reviews every landing page before approval and uses its content to assess relevance. That requires open access for the OAI-AdsBot crawler — no block in robots.txt, no bot protection, no CAPTCHA in front of it.

We built a dedicated landing page for this instead of using the existing website. It can be tailored exactly to the ad and changed quickly without touching the site. Whether ad and page fit together can be cross-checked with the message match checker — a heuristic that shows word overlap, not a verdict.

The form asks for role, company size and area of interest. Three fields more than necessary? No: in B2B, the quality of an enquiry decides, not the quantity.

Measure only with consent — and still keep the lead

OpenAI's measurement script only loads on our page once visitors agree. Every enquiry is validated on the server and handed straight to HubSpot. Only once HubSpot has accepted the lead — and only with consent — does the page report the conversion to OpenAI. In parallel via pixel and server, with the same event ID, so nothing is counted twice.

Plan for the consequence: enquiries without consent arrive in the CRM but never show up in the platform's figures. Measurement in the ad account is systematically incomplete. So judge the channel by what arrives in the CRM, not only by what the platform reports.

Approvals at fixed points

We delivered the project with an AI agent for development and research. That made it all the more important where a human decides: on the submission risk, on budget and run time, on activation and on every later change. The campaign plan existed as a file, was approved via a checksum and created in paused status. A human activated it.

A system, not a one-off campaign

The real lever is not the first campaign but every one that follows. That is why we built the setup as a system from the start: new landing page variants run on the same tested layout, form and handover path. New campaigns and ads are created as a plan file, automatically checked against the character limits and approved via a checksum. Platform knowledge is kept with source and check date, and automated tests cover every component.

That turns every further campaign, ad or report into a routine step instead of a rebuild. Automatic weekly reports and creating new campaigns straight from approved plan files are already prepared.

Why we do not share figures

OpenAI does not yet publish benchmarks for ChatGPT Ads across advertisers or industries. Figures from a single account on a new platform are still quickly read as a benchmark — and they are not fit for that. On top of that, they belong to the client. Anyone selling you specific click prices or conversion rates for ChatGPT Ads as an expectation today is generalising from a single case.

Frequently asked questions

Is ChatGPT Ads worth it for B2B?

There is no blanket answer yet. Only users on the Free and Go plans see ads — if your only target is decision-makers on company accounts, you will not reach them here. What makes sense is a time-boxed test with clean measurement and an honest read-out afterwards.

Can I use my existing website as the landing page?

In principle yes, if the crawler has access and the page fits the ad. A dedicated page is usually the better choice: it can be tailored and changed without touching the website.

Does it have to be HubSpot?

No. What matters is the principle: the enquiry is validated on the server and handed to your CRM, and the conversion only counts once the CRM has accepted it.

Where does most of the setup time go?

Not into the ads. Eligibility, landing page, consent and the CRM handover make up most of the work — and decide whether you can evaluate the channel at all afterwards.

What you can do today

1. Check your offer against the initial-phase categories before building anything.
2. Write down five situations in which your audience would ask ChatGPT for advice — that is the raw material for context hints.
3. Put your draft ad next to your landing page hero and check the fit in the message match checker.
4. Work out how an enquiry without consent gets into your CRM — and how you evaluate it there.
5. Decide up front who approves budget, activation and changes.

What this looked like in the project is described in the case study. If you want to know whether your offer passes review: how we set this up is described under ChatGPT Ads — or talk it through in a first call.