OpenAI is testing a new ad format on ChatGPT: when a user clicks an ad, they don't land on the brand's website. They open a conversation with an AI agent instead. No landing page, no traffic to track in Google Analytics.
For anyone running paid campaigns, this changes the underlying logic: the point of contact moves inside the platform. Classic metrics like sessions, bounce rate and time on site become irrelevant. What matters is what happens in the chat.
It isn't available to everyone yet. But it's worth understanding now, before it arrives in media plans as a standard option. If you manage ad budgets in industries with long sales cycles or complex products, you have more to gain from this format than those selling simple, low-priced products.
A click that opens a chat, not a page
The new format announced by OpenAI, as described by Digiday, works like this: the brand creates an ad inside ChatGPT, the user clicks it, and instead of landing on a website they end up talking to an AI agent configured by the brand itself.
In this context, an AI agent is a conversational assistant trained to answer questions about a specific product or service. It can answer questions, qualify interest, collect contact details and guide the user toward a purchase decision.
The difference from any other form of digital advertising is structural: the funnel, meaning the path from ad to conversion, never leaves the ChatGPT platform.
What it breaks in today's campaign logic
If you run Google or Meta campaigns, you know the click is only the beginning: it drives traffic to a landing page, and that's where everything gets measured. Bounce rate, time on page, form completion, purchase. Every step can be tracked and optimized.
With ChatGPT's click-to-chat format, that chain breaks. Traffic never reaches your site. Sessions don't show up in Analytics. The Meta pixel sees nothing.
This creates two concrete problems:
- Attribution: how do you measure whether a conversation generated a customer? Ad attribution is already complex today, and when attribution models mislead your budget is a real risk even without adding new opaque channels.
- Creative control: the brand defines the agent's instructions, but the conversation is generative. The AI answers every user differently. There's no fixed copy to approve.
These problems aren't insurmountable, but you need to factor them in before allocating budget.
Who this format is really for
Click-to-chat makes sense when a product or service needs qualification before the sale. Typical sectors:
- B2B software with on-request demos
- Professional services (consulting, training, legal)
- Products with custom configurations
- Finance or insurance, where the buyer's needs are complex
In these cases, the AI agent can do the job that a contact form plus a pre-qualification call does today: understand the need, handle initial objections and collect useful data.
It makes less sense for those selling simple, fixed-price products, where the ideal path is still click → product page → cart → purchase. Adding a conversation in the middle slows things down instead of speeding them up.
It's also worth looking at what's already happening with integrations between AI platforms and advertising: Amazon Ads on ChatGPT has already opened up this territory for marketplace sellers, and Amazon DSP on ChatGPT Ads shows how Italian B2B companies can reach users who are actively searching.
The budget problem: where control ends
One of the least discussed risks of new AI formats is the loss of budget governance. With Google and Meta, automation is advancing, but there are still thresholds, targets and rules you can set manually. The topic is already on the table: how to stay in control when Google and Meta automate campaigns is a question many marketing managers are already asking.
With a new format like ChatGPT's click-to-chat, governance is even less certain. There are no historical benchmarks, no established best practices, no agencies with years of data to lean on.
That doesn't mean avoiding it. It means entering with a contained test: a separate budget, clear goals, and metrics defined before the campaign goes live.
What to do in the coming months
The format isn't generally available yet. But the time before the official launch is useful for getting ready:
- Map the conversational touchpoints: what questions does your sales team get today during pre-qualification? Those become the AI agent's instructions.
- Define alternative metrics: if site traffic is no longer the main metric, what replaces it? Completed conversations, data collected, demo requests generated by the chat.
- Isolate the test budget: don't move budget away from campaigns that work. Use a dedicated experimental share.
- Check consistency with the rest of the media plan: a format that lives inside ChatGPT should be integrated into your overall paid media and advertising strategy, not tacked on as an afterthought.
A common mistake here is waiting for the format to mature before studying it. Once it goes mainstream, anyone who hasn't already thought about AI agents, attribution and conversational metrics starts behind.
SHM Studio follows the evolution of these formats to help companies decide where and when it makes sense to experiment, without burning budget on channels that aren't ready for their market yet.
Related articles
Discover more articles exploring similar topics, selected to offer you a more complete and stimulating perspective. Each piece of content is carefully chosen to enrich your experience.
