Machine layer

Your site has a new kind of reader.

Someone asks an assistant about your company, your project, your work. It answers — frequently without ever loading your page. What it says is assembled from whatever it can reach and whatever it already assumed, and you get no vote in that unless your site is built so it can be found, read, and checked.

What this is

A record of doing that to one website: what moved, what did not, and what turned out to be the wrong problem entirely. The central finding is that being found, being understood and being believed are three separate problems with three unrelated fixes.

reach: did the system get your page at all?
understanding: having got it, did it parse what you meant?
credence: having parsed it, did it treat your claims as true?

They fail in that order, and only two of them are worth your time. Understanding turned out not to be a problem at all — every model that actually reached the page parsed it correctly, every time. Reach is mostly infrastructure and mostly not yours. Credence is almost entirely writing, and it is where the returns are.

The rest is the working out: the markup, written out; the loop that produces findings of your own; five things built and then removed; and the evidence, dated, with what it does and does not support.

One site, over three weeks. The numbered blocks are observations, recorded at the time. Everything around them is what was made of those observations — a weaker kind of claim, and written as ordinary prose so it never borrows the authority of a measurement. Nothing here is offered as a rule of the web.

Start with reach — why a cautious answer about your site usually means nobody read it.