Market Pulse

Market Pulse

What a 1984 Airline Screen Rule Would Ask of ChatGPT's Unpaid Answers

ChatGPT has started showing ads from businesses, and that brings back an argument the airline industry had in the 1980s. Back then, travel agents booked through reservation systems that airlines owned, and one of those systems found that 93 percent of agents' bookings came from the first screen of flights. Whatever appeared on that screen had a large advantage. This essay sets the two cases side by side and asks: if your business isn't among the advertisers, how would you know whether payment had shaped ChatGPT's ordinary, unpaid answers?
What a 1984 Airline Screen Rule Would Ask of ChatGPT's Unpaid Answers
ChatGPT has started showing ads from businesses, and that brings back an argument the airline industry had in the 1980s. Back then, travel agents booked through reservation systems that airlines owned, and one of those systems found that 93 percent of agents' bookings came from the first screen of flights. Whatever appeared on that screen had a large advantage. This essay sets the two cases side by side and asks: if your business isn't among the advertisers, how would you know whether payment had shaped ChatGPT's ordinary, unpaid answers?

Where Capable Agents Still Fail
SOPBench: Evaluating Language Agents at Following Standard Operating Procedures and Constraints
In both directions. 65% of GPT-5-mini's remaining errors refused allowed actions; 63% of Qwen3.5-2B's executed forbidden ones.
Once the rule check is correct, how often does the agent block forbidden actions, and how often complete allowed ones?
Where Capable Agents Still Fail
The Hard Part Comes After Search: Benchmarking Web Agents on Synthesizing, Organizing, and Displaying Knowledge
Often in formatting and layout. Visual errors left files unusable even when much of the content was right.
When a vendor quotes a score, does it count finished deliverables, or partial credit on work nobody could ship?
One Number

Uber says AI agents built by other companies booked four rides during Halloween week. Uber handled roughly 320 million trips that same week. Plenty of people read that as proof nobody wants a bot hailing their car.
Maybe. But if an alert told me four requests succeeded, my first question would be: out of how many? A success count with no attempt count is fucking useless. Uber didn't say how many people had an agent that could book a ride, how many tried, or how many failed.
So four could mean people don't want this. It could also mean the doors for agents were barely open. This number can't tell you which.






