Your Agent is Only as Honest as Your Signal

Summary

The article argues that ChatGPT is emerging as an intent-driven discovery and decision surface, not a place where users passively scroll. OpenAI’s platform reportedly has more than 1 billion active users and 2.5 billion daily prompts, and consumers are increasingly asking AI to solve specific problems—such as finding a dog-friendly seaside booking—without starting from a brand or search engine. Budgets follow behaviour, so advertisers have begun testing ChatGPT at the top and middle of the funnel.

The critical question for ad ops is whether the channel can be measured credibly enough to sit alongside Meta, TikTok and Google. Clicks and impressions are insufficient; finance and executive stakeholders need proof that spend produced business outcomes. The article identifies two structural measurement challenges.

First, marketers must bridge web and in-app conversions. Grubhub, advertising in ChatGPT since OpenAI opened the platform to advertisers in February, has roughly five months of campaign data and uses cross-platform measurement to view website visits, purchases, sign-ups, app installs and in-app revenue in the same dashboard as other channels. Second, most AI-driven traffic may be organic, not paid, so independent measurement must cover brand recommendations generated in AI answers, not just platform-reported paid metrics.

The broader argument is signal quality. Models and optimization engines are only as good as the data fed into them; messy or biased signals undermine trust. Deduplicated, unbiased measurement becomes more important as more of the journey moves into AI surfaces.

For ad ops decision-makers, the takeaway is to treat ChatGPT as a measurable media channel, not a small POC: define outcome KPIs, implement independent cross-platform attribution, separate paid and organic AI traffic, deduplicate conversions, and compare performance on equal footing with established channels. CMOs who get ahead will put measurement at the centre of channel decisions and scale only when business impact is proven.

Analyst Note

What's notable here isn't the ChatGPT ad inventory itself but the framing of measurement as the gate to channel legitimacy. Every new surface — retail media, CTV, short-form video — has followed the same arc: attention first, budgets second, credibility only once independent measurement caught up. AI-native discovery is compressing that timeline, with users arriving with declared intent rather than incidental exposure.

The key implication for ad ops is structural. Conversational surfaces split traffic across web and in-app paths while blurring paid and organic recommendations, and most attribution stacks were not built to reconcile either. The paid/organic distinction matters particularly: a meaningful share of AI-driven referrals are unpaid brand mentions inside answers, which platform-reported figures cannot arbitrate. That is a measurement architecture question, not a media buying one.

Timing is worth watching. Advertisers already running campaigns are generating the first cohort of benchmark data, which will set expectations for the rest of the market. Meanwhile the signal-quality argument — that optimization engines inherit whatever bias sits in their inputs — extends well beyond this channel. Competitive comparison to Meta, TikTok and Google will hinge less on reach than on whether AI surfaces can be measured on equal footing inside existing dashboards.