Every standard emerging between websites and agents points the same way: from the site to the agent. I haven’t found one that carries a message back.
Line them up. robots.txt says what an agent may fetch. llms.txt says what is worth reading. Schema markup says what a page is. Actions a page declares for browser agents say what can be done here. Different jobs, one direction.
That isn’t an oversight. Each one solved a problem somebody had. Sites wanted control over who crawls them, and agents needed to read pages without wading through markup. Both sides had a reason to build a way for the site to speak. Nobody had a reason to build the return.
The return would look like this: I reached step three of your flow, the form rejected a value I was allowed to send, I stopped, here is the state I was in. A failure report a machine can send and a company can receive.
The agent’s side won’t build it unprompted. Trying another source costs less than filing a report. Your side can’t ask for it, because there is no channel to ask through.
Before agents, that channel existed without anyone designing it. It was a person with a complaint. Informal, inconsistent, and universal, because every user who hit a wall was also a customer who could say so.
So a company can follow every convention on offer and still be in a strange position. It has become easier to read, and no easier to tell. The first can be finished, ticked off a list, scored as complete. The second can sit at zero the whole time, and nothing on the list will show it.
None of this argues against adopting any of them. It says what they are: ways of being understood. They don’t tell you where you fail.
Next time someone proposes a standard for agents, ask one thing before anything else: what comes back? If the answer is nothing, you’re paying to be read, not to find out where you break.