When the Channel Creates the Customer

ChatGPT™ hit one billion monthly active users this week. Reuters confirmed it June 2nd. The reaction from most brand teams I’ve talked to has been something like: “Yeah, we know it’s big. What are we supposed to do about it?” Fair question.

But I think most people are asking it wrong. The industry conversation about ChatGPT™ and commerce has centered on volume. Fifty million shopping queries a day, that’s the OpenAI/NBER number, based on roughly 2% of 2.5 billion daily prompts. Stackline pegs it at 84 million weekly shopping questions in the US alone. My napkin math, scaling for user growth since that 2% was measured, puts current volume at 60-70m daily. Could be higher.

OpenAI has shipped shopping research, native product recs, memory, and personalization since that original measurement. If shopping’s share of prompts ticked from 2% to 4% you’re looking at 90-120m shopping queries a day. (Those extrapolations are mine. Not sourced.) Big numbers.

But volume isn’t the interesting part. BI reported Similarweb analysis showing that a large share of users who engaged with shopping content on ChatGPT™ started their conversation with no buying intent at all. I haven’t been able to independently verify the exact percentages, the sourcing on this is muddier than I’d like but the directional finding matches what I’m seeing in practice.

So what does that actually mean? On Google, intent arrives pre-formed. Someone searches “best hat for hot, sunny Vegas summers” That person was already a shopper before they touched the keyboard. Google captures that intent. Advertisers bid on it. The whole system runs on demand that already exists. ChatGPT™ works differently. Someone opens it and types: “I’ve been exhausted lately, not sleeping well, what am I missing?” 30 questions later, they’ve worked through their sleep schedule, diet, stress levels, magnesium intake & they’re clicking through to buy a supplement they had no intention of buying when they sat down.

The conversation generated the purchase intent. That’s a different kind of channel than anything in your current media mix.

Closer to a really good podcast host who talks you into a product than to a search engine where you declare what you want.

CPA bidding on ChatGPT™ launched recently. Through Criteo, with conversion pixel tracking. Minimums dropped to $10,000. OpenAI’s Ads Manager can ingest up to a million SKUs. The infrastructure now looks, mechanically, like every other performance channel: CPA targets, pixel, SKU feeds, bid optimization. And that’s exactly the problem. Thousands of brands are about to test ChatGPT™ ads using performance marketing frameworks built for demand capture. They’ll set ROAS targets, optimize for last-click, compare CPAs to Google Shopping, and the results will look confusing because they’re measuring a demand-generation channel with demand capture tools.

Meanwhile, 70.6% of AI referrals are invisible in GA4 (Parcel Perform data). Google searches per US user have dropped roughly 20% year over year. Some of the demand “disappearing” from Google isn’t gone it’s moving to a channel that shows up as direct traffic in your analytics.

OpenAI themselves learned this the hard way. They killed Instant Checkout after Walmart pulled back because in-chat conversion was 3x lower than sending users to Walmart.com directly. Users weren’t ready to buy inside ChatGPT™. They were ready to decide inside ChatGPT™ and buy somewhere else. TD Cowen called it a “stunning admission.” The channel’s value isn’t in closing the sale it’s in shaping the consideration set before the user ever reaches a product page.

Which brings me to the question I keep coming back to: what happens to your brand strategy when the platform that distributes your product also creates the desire for it?

Google could suppress your rankings. Amazon could undercut you with a private label. Those are competitive threats with playbooks. ChatGPT™ could just… not generate the kind of conversations that lead to your product category. Or it could generate them and recommend someone else. And you might never see it in your data because the referral doesn’t carry through.