OpenAI and Criteo’s announcement that ads are coming to ChatGPT has sparked a predictable reaction from the industry. Performance marketers are excited about a powerful new intent signal, brand safety teams are nervous and everyone is asking whether this changes search.

But here’s the thing: ads inside AI search aren’t a new business model; they’re a familiar one in an unfamiliar context. Search has always been about selling intent. What is different here is the interface, and that changes everything that happens after the click.

Chat already surfaces recommendations as if they’re impartial. A user asks which CRM to buy, which agency to hire and/or which software to evaluate. The model responds.

Yet the moment ads enter that experience, that impartiality is gone, and users may not know it. Increasingly, they’ll ask a question, evaluate options and take action without ever leaving the conversation.

That is a powerful new performance surface. But it also opens the door to three downstream problems that most stacks are not built to handle.

Problem 1: The models hallucinate.

Unless the large-language model (LLM) powering your service uses forms, you’re looking at pre-populated lead data, which presents its own risks.

Social networks did this years ago. LinkedIn pre-fills your job title. Facebook pre-fills your email. It reduces friction and improves conversion rates.

AI-powered ad surfaces will almost certainly do the same thing. In theory, this is great. Unfortunately, models hallucinate, which means that even though they may get most of the context right, it’s not uncommon for them to render names incorrectly.

Now scale that across a demand-generation program running thousands of impressions per day.

Hallucinated lead attributes are a new category of bad data. They are not obviously wrong, and they are not flagged as errors. They look clean in your CRM.

A misspelled name or incorrect job title is annoying. A hallucinated company domain or fabricated email address is a compliance problem and a wasted sales cycle. Your stack needs to validate the information before any of that data touches anything downstream.

Marketing Technology News: MarTech Interview with Stephen Howard-Sarin, MD of Retail Media, Americas @ Criteo

Problem 2: Have opt-in, but will it travel?

Compliance rules such as the General Data Protection Regulation (GDPR) and California Consumer Privacy Act (CCPA) were built around form-based consent flows. A user clicks a checkbox. The timestamp, IP address and consent language get logged. Auditors can trace it, and legal can prove it.

Consent captured inside a chat interface wasn’t designed for GDPR or CCPA. It falls in a legally ambiguous territory. The interaction is conversational. The data handoff is opaque.

If a user effectively consents to being contacted by asking ChatGPT to connect them with a vendor, does that consent carry downstream to your CRM, marketing automation platform or sales development rep who calls them three days later?

Nobody has answered that yet, but regulators will.

The companies that assume chat-native, opt-in maps cleanly onto their existing compliance infrastructure are building legal exposure they cannot see. The opt-in attribution has to travel accurately, or you have a problem sitting inside your CRM waiting to surface at the worst possible time.

Problem 3: Most data stacks aren’t built for AI chat volume.

ChatGPT is not some niche channel. It already has more than 300 million weekly active users. Naturally, ads in that surface will scale fast, which means the volume of inbound intent signals hitting your pipeline is going to increase exponentially.

As a result, identity resolution, deduplication, enrichment, consent controls and routing logic all have to work reliably at higher volume and with a new data format. If they do not, brands will pay for high-intent traffic and still lose it through poor match rates, duplicates or slow follow-up.

The reality is that most stacks were not designed to support this pace, and the gap between media spend and revenue outcomes becomes very visible, very fast. A bad lead through traditional displays wastes the budget. A bad lead through an AI-optimized channel trains the model to repeat the same mistake at scale.

Modern infrastructure, not media buying, rules

The next phase of performance marketing will be defined by data infrastructure. Winning in AI-powered channels requires lead infrastructure that can handle the intent that follows.

Clean data. Enforced validation rules. Reliable consent attribution. Accurate routing between systems. These attributes can be the difference between a new channel that generates measurable pipeline and one that adds noise.

Everyone is focused on whether to buy ChatGPT ad inventory. The smarter question is whether your stack is ready to receive what it sends.

The modern data stack requires a data firewall between your lead sources and your CRM. This enables you to validate, normalize and enforce compliance before bad data touches anything downstream.

Is your stack modernized for the AI era?

About Convertr

Convertr is the enterprise data integrity layer that ensures clean, compliant, and complete data flows across systems and teams to support AI-readiness

Marketing Technology News: From MarTech Stack to MarTech Fabric: Weaving Brand, Content, and Conversion Into One Thread