Early ChatGPT Ads Lagged Direct Visits

🛒 Paid visits had weaker buying results. At 102 sites with financial tracking, ChatGPT ad visits had 84% lower purchase odds than direct visits. Purchase odds compare visits with a purchase to those without one. Average order value was $79 lower, and revenue across all visits was $2.86 lower per visit. For unpaid ChatGPT and both Google Search channels, the researchers found no statistically clear difference from direct traffic in purchase odds.

🔎 People who clicked ads browsed less. Across 387 sites with ChatGPT ad referrals, paid ChatGPT visits had 59% higher odds of ending after one page than direct visits. They averaged 1.4 fewer pages and 104 fewer seconds per visit. Unpaid ChatGPT and paid Google Search each had 39% lower odds of ending after one page than direct traffic. The routes drew different browsing patterns, and the comparison leaves open what shoppers intended when they clicked.

🇺🇸 The US comparison puts dollars on the gap. In that separate comparison, ChatGPT ad traffic brought in $0.89 per visit versus $2.45 for paid Google Search. Those are attributed revenue levels for each channel. The Buying figures above show how each channel differed from direct visits.

💵 Set a price for the test. The authors estimated about $0.26 per ChatGPT click to match paid Google Search revenue per ad dollar. They cited $1.72 per click from one publicly reported account. A cheaper click can close a revenue-per-dollar gap even if each visitor still spends less. Set your own bid ceiling using the revenue your test credits to each click and what remains after product and fulfillment costs. The $0.26 benchmark depends on the study's US comparison.

🧪 Measure orders the ads actually add. The study credits the final click before purchase. A shopper could discover a store through a ChatGPT ad and later buy after clicking a Google result. A ChatGPT click can also get credit for a sale that would have happened anyway. The data cannot connect ad exposure to a later purchase through another channel. If the campaign allows it, withhold ads from a comparable group and compare total orders or profit across channels.

📉 Watch unpaid referrals as ads roll out. In a separate analysis of market launches, the authors estimated a 5.9% fall in unpaid ChatGPT referrals over 13 weeks. The 95% confidence interval ranged from a 12.6% decline to a 1.2% increase, so the data were consistent with no change. The purchase rate among arriving unpaid visitors showed no clear change. Track unpaid visits and orders separately as paid traffic grows. Even with a steady purchase rate, fewer unpaid visits can mean fewer orders from that source.