About

The audienceChanged First

By Robots For Robots is an open laboratory for the machine-readable web. It exists because the audience for a web page changed and most websites have not.

The premise

For thirty years the job was to make a page work for a person who had already found it. That job still matters. It is no longer the first job.

Before anyone sees anything, a page is fetched, stripped of its presentation, split into passages, embedded as vectors and filed. What survives that process is what gets quoted back to the person who asked a question somewhere else entirely. A great deal of what websites spend money on (the hero video, the scroll animation, the carousel nobody clicks) does not survive it at all. It is invisible to the reader that arrives first.

That is not an argument against design. It is an argument that structure is now part of the design, and that most of the industry is still treating it as plumbing to be hidden. This site takes the opposite position: the plumbing is the interesting part, so it is put on the surface where it can be inspected.

A page that reads beautifully and parses badly is a page that loses the argument before the argument starts.

How it runs

Operating rules

01 · Ship half-finished
Nothing waits for polish. Versions are numbered, dated and left visible. A laboratory that only publishes conclusions is a brochure.
02 · Point it here first
Every instrument built here is run against this site before it is pointed at anyone else's. If a test is too embarrassing to run on the home page, it is not ready.
03 · Wrong stays up
Failed experiments stay published and marked as failed. Predictions that aged badly keep their original date. Quietly deleting the misses is how a field stops learning.
04 · No gates
No email wall, no sign-up, no paywall and no interstitial. Nothing on this site is for sale and nothing needs your address.

On rule 03

Why wrong
stays up

A laboratory that deletes its failures is a brochure wearing a lab coat. Every experiment here carries a status, and the ones that did not work keep theirs, with the original date attached.

The argument is practical rather than noble. In a subject moving this quickly, the record of what failed is more useful than the record of what worked. It is rarer, it is harder to reconstruct later, and it is the part everyone else quietly removes. A defect written up with a date on it is also a stronger claim than a conclusion with none. The build log for this site documents two of its own bugs at the length they deserve, because they were the most interesting thing that happened while it was being built.

The same standard applies going forward: a hypothesis stated before the result, a method you can repeat, and the status left as whatever it actually is, including queued and including abandoned.

Who runs it

A side project
of Double Atari

Double Atari is a Minneapolis consultancy working across SEO, GEO and AEO, CMS architecture, content strategy and analytics: the practical end of making websites legible to both people and machines.

This is where the questions go that a client engagement has no budget for. The two feed each other: techniques proven here end up in the client work, and problems met in the client work end up as experiments here. That is the whole arrangement. Nothing on this site is a product, nothing is for sale, and there is no newsletter.

Colophon

Specification

Status
Version 0.1, perpetual draft. No 1.0 planned.
Stack
Hand-written HTML, CSS and vanilla JavaScript. No framework, no build step, no dependencies.
Typography
Barlow Condensed for display, Inter for prose, JetBrains Mono for anything a machine said.
Images
No images inside any page; the only graphic on screen is a CSS grid behind the masthead. Two raster files exist for places this site does not control: a favicon and a sharing card. Both were drawn from the site's own type, rule and grid, neither photographed nor generated.
Analytics
Google Tag Manager, feeding Google Analytics 4, and Google Search Console. No advertising pixels, no session recording, no heat maps.
For machines
llms.txt, sitemap.xml, feed.json, and schema.org JSON-LD on every page.
Crawler policy
Open. Every well-behaved crawler is welcome, AI training crawlers included. A site arguing that machines should read well cannot then refuse to be read.