By Robots For Robots is a small open laboratory about how crawlers, retrievers, and answer engines read a web page. It grew out of the work I have been doing on the Double Atari site. Its main feature is a machine view: press Shift + M on any page and the design falls away, revealing the heading outline, structured data, the extraction a parser keeps, and the chunks a retrieval system would split the page into. The look is deliberately odd, and one friend asked if it was an acid trip. But if you want to understand how robots see website content, it shows you instead of telling you.
Where did the name come from?
The post was Created by Robots for Robots, about how this site is built with AI tools, measured with Search Console, GA4, and Semrush, and written for two audiences at once: people, and the increasingly robotic systems that decide whether people ever see it.
The title was a little tongue in cheek. But the more I worked on the Double Atari site, the more literal it became. Every change I made was really a question about machines. Does the crawler see this heading? Does the schema parse? Which paragraph would an AI answer engine quote? I kept answering those questions with screenshots, browser extensions, and validators, and I kept wishing I could just flip a switch and see the page the way the machine does.
So I liked the title, and I had a pile of things I had learned on this site, and I thought: what if I built a website that took the name literally?
What is By Robots For Robots?
By Robots For Robots calls itself "an open laboratory for the machine-readable web." Its premise is simple: most pages are now read by a machine before they are read by a person, so build a site whose primary audience is the crawler, the retriever, and the answer engine, and treat the human-readable layer as the demonstration.
In practice that means:
- A machine view on every page. Press Shift + M or use the switch in the masthead. The page gets labeled by semantic region, every heading shows its level, and a dock opens with four panels: Outline, Structured Data, Extraction, and Chunks.
- Experiments instead of opinions. Each one tests a single claim about machine reading, and the results stay up with dates attached, including the wrong ones.
- Field notes. The first note is a build log, including two defects the machine view found in its own site on day one.
- Nothing gated. No sign-up, nothing for sale, and robots.txt does not block anything.
It is also built the way it preaches: no framework and no build step, JSON-LD on every page, plus llms.txt, a sitemap, and a JSON feed for the machines.
Is it an acid trip?
Right away it is super interesting to look at. Huge condensed type, a near-black background, a glowing teal accent, and once you switch the machine view on, labels like "#HEADER · SITENAVIGATIONELEMENT" pinned to the edges of the page while a terminal-style panel slides up from the bottom.
One friend, who I count on for honest feedback, looked at it for a while and asked if it was an acid trip.
Fair. It does not look like a normal website, and that is partly the point. A normal website hides its structure behind its design. This one puts the structure in front, because the structure is what the machines actually consume. If it feels a little disorienting, that is roughly how a page feels to a parser: no fonts, no colors, no layout, just regions, headings, entities, and text.
I have kept the look. I have also taken the feedback seriously. The next planned revision of the machine view ranks its findings by impact instead of listing everything evenly, which should make it easier to read for someone who has never opened developer tools.
Why is it useful if you care about SEO and AI search?
Because underneath the strange design there is some really useful content if you want to understand how robots see web pages. Most SEO and AI search advice tells you to "use clear headings," "add schema," or "write in chunks." The machine view shows you what those instructions actually produce.
| Machine view panel | What it shows | Why it matters |
|---|---|---|
| Outline | Every heading on the page, indented by level, with a warning if a level is skipped | Headings are the table of contents machines use to understand and split a page |
| Structured Data | The JSON-LD on the page, parsed, or the parser error if it is broken | Schema is how a page states what it is, who published it, and how it connects to other entities |
| Extraction | Title, meta description, canonical, language, robots directives, word count, estimated tokens, links, images, and landmarks | This is roughly what a crawler keeps once the presentation is stripped away |
| Chunks | The page split into retrievable passages at section boundaries, with a token estimate for each | Retrieval systems match questions to passages, not to whole pages |
None of it is hand-written. Every panel reads the live page at the moment you open it, so the site cannot flatter itself. If a heading level is skipped, the Outline says so. If the structured data will not parse, the panel shows the error. I go through what each panel teaches in a two part series: part one covers Outline and Structured Data, and part two covers Extraction and Chunks.
What experiments are running?
| Experiment | Status | The claim it tests |
|---|---|---|
| EXP-001 The Machine View | Running | Authors fix pages faster when they can see the parser's view without leaving the page |
| EXP-002 Readability Probe | Running | Paste any page's HTML and run sixteen weighted structural checks in your browser, with nothing uploaded |
| EXP-003 Citation Drift | Queued | The sources an answer engine cites for a fixed question change far more than the pages a search engine ranks |
| EXP-004 The Robots Ledger | Queued | Everyone talks about AI crawlers, few people publish their logs, so publish the logs |
Citation Drift is the one I am most excited about, because it is the measurement problem I keep running into with clients. If you ask ChatGPT or Perplexity the same question every week, how often do the cited sources change? Nobody has a good answer yet. The site says plainly that it will take months to produce a useful result, and that is fine.
How does it connect to Double Atari?
Everything on By Robots For Robots started as a problem on this site. The heading checks came from cleaning up heading structure on Double Atari pages. The structured data panel came from validating the schema that every Double Atari post carries: BlogPosting, BreadcrumbList, and FAQPage. The extraction and chunk panels came from writing about how language models decide what to cite and technical SEO for AI search, and wanting to show the mechanism instead of describing it.
The lab also returns the favor. Running the machine view against Double Atari pages is how I check this site's own headings and structured data before I publish. Double Atari is the consulting practice. By Robots For Robots is the workbench.
How to try it in two minutes
- Open byrobotsforrobots.ai.
- Press Shift + M, or click "Machine view" in the top right.
- Click through Outline, Structured Data, Extraction, and Chunks in the dock at the bottom.
- Then ask yourself the uncomfortable question: if I flipped this switch on my own homepage, what would it say?
If you want to find out without building anything, the Readability Probe will check your HTML for you. And if you want someone to run the full version of that check on your site, that is what a Double Atari website audit does.
By Robots For Robots FAQ
What is By Robots For Robots?
By Robots For Robots is an open laboratory website by Double Atari about how crawlers, retrieval systems, and answer engines read web pages. It publishes working tools and field notes, and every page includes a machine view that shows how a parser sees it.
What is the machine view?
The machine view is a toggle, opened with Shift + M, that strips the design from a page and shows its semantic regions and heading levels. A dock shows four panels: the heading outline, the parsed structured data, the extraction payload a crawler would keep, and the retrieval chunks the page would be split into.
Why does the site look so unusual?
It puts structure in front of design on purpose, because structure is what machines consume. The look is closer to how a page feels to a parser than to a typical marketing site. One friend asked if it was an acid trip, which is fair.
Does By Robots For Robots collect data?
No. The site says its session telemetry is computed in your browser and stored nowhere, the Readability Probe runs locally without uploading your HTML, and nothing is behind a sign-up.
How is it related to Double Atari?
It is a Double Atari side project built from lessons learned optimizing the Double Atari website for search engines and AI answer engines. Double Atari is the Minneapolis consulting practice run by Andrew Charon.
Explore related Double Atari resources: Created by Robots for Robots, how the Double Atari website keeps evolving, GEO and AEO / AI search, and the Search Lab.