Research Note
Too Slow for the Machines
The dealer layer is now being read by software that shops on a buyer's behalf.
The dealer layer carries an 8.7s median mobile LCP, which is the wait before the main content of a page appears.
For as long as anyone has cared about that number, the cost of it was a person deciding to leave. That framing is getting narrower. A growing share of early shopping happens through software that fetches pages on a buyer’s behalf, reads what it finds, and assembles a shortlist before the person has looked at anything. The page still has to be delivered before it can be considered. What has changed is who is waiting on it.
This part is worth stating plainly. The behavior of shopping agents is the argument in this note, and it is not something I measured. I have no finding about how any particular system ranks pages, how long it is willing to wait, or what it does when a page is slow. What I measured is the wait. The rest is what the wait implies about a page that has to be read before it can be judged.
What it means for a dealer deciding what to fix first
The useful property of page delivery is that it sits upstream of everything else on the list. New photography, a review widget, a chat bubble, a redesign, a fresh product catalog: every one of those improvements is behind the same wait, and none of them can be evaluated by a reader that has moved on. Fixing delivery is the one item on the list that makes the others legible.
It also has the rare quality of helping both audiences at once. The same change that puts content in front of a person sooner puts it in front of the software reading for them. Most website spending has to pick a beneficiary. This does not.
One limitation
The measurement stops at the wait. It says how long these pages take under standardized mobile testing conditions, and it says nothing about how any agent, crawler, or assistant treats them, because that was never tested here. Read this as a claim about the condition of the dealer layer and an argument about who is now affected by it. Anyone presenting it as evidence of how machine buyers behave would be borrowing authority the measurement does not carry.