Articles, written slowly.
Engineering notes, design rationales, and the occasional opinion about AI, bots, and the open web. Updated when we have something worth saying.
What does Concept211 write about?
Two things, and nothing else. The first is case studies of software this team built and still operates, written from the inside rather than from a brief: CarSnipe, ConceptSEO, SocialBotify, ClaudeView and the Hyperliquid Guide. Each one names what the product does, the decision that shaped it, and the numbers it produces, including the ones that moved the wrong way. The second is measured research on how AI search decides what to cite, run against our own properties first and client work after that.
Nine articles are published here as of 25 August 2026: five case studies, three pieces of research, and one engineering field report on the eight data APIs behind that research, written up with the integrations that broke. There is no posting schedule and there will not be one. A page goes up when there is a measurement behind it, which is why the archive is short and why every figure on it carries the window it was taken over.
What is in these articles that is not in everyone else's?
First-hand measurement, published with its limits attached. The longest piece here reports what a weekly AI-search audit cadence produced across eight sites over 20 weeks: 1,683 recommendations implemented and verified against live URLs, with 41.0 percent of that work landing in content and internal linking rather than in the technical checklist most guides lead with. Crawler access, llms.txt and server-rendered HTML passed on 100 percent of the sample with zero variance, which makes them preconditions rather than differentiators.
A separate reading of Cloudflare's edge analytics across 14 properties over 4 to 10 August 2026 counted 74,236 requests carrying a known crawler user agent, then split them by path, and found that the headline refusal rates every AI-visibility tool reports are dominated by credential scanners borrowing crawler names. On one property, 1,171 of 1,180 requests claiming to be Applebot asked for files the site has never published.
None of that is a survey of other people's advice and none of it is modeled. Where a dataset cannot support a claim, the article says so and stops, and where a number came out unflattering it is printed the way it came back.
Every SEO data API we wired into production, and which ones are actually usable
Field notes on eight SEO data APIs behind a running platform: what Search Console gives you that the UI does not, why PageSpeed Insights times out, and what breaks that the docs never mention.
What 20 weeks of weekly AI-search audits across 8 sites produced
A first-party benchmark: 1,683 SEO and GEO fixes implemented across eight live sites in 20 weeks, with the method, the null results, and what the data does not prove.
Case study: building ClaudeView, a native Windows app for Claude Code
Why we built a WPF desktop app for the Claude Code CLI instead of an Electron wrapper or a VS Code extension, and what a 120 MB three-pane workspace buys you.
Case study: building SocialBotify, an AI social post generator with a strategy behind it
Inside SocialBotify: deriving brand voice from a customer own site, planning posts against a 70/20/10 framework, and publishing to ten networks from one approval screen.
GEO vs traditional SEO: what actually changes when you optimize for AI search
What changes when you optimize for AI search: what engines cite, the signals that win a citation, how to measure it, and when to hire a GEO partner.
Case study: building ConceptSEO, an AI platform that runs SEO every week
Inside ConceptSEO, our Laravel and OpenAI/Anthropic platform that audits sites weekly, writes engineering-ready fixes, and tracks AI-search citations across multiple domains.
Case study: getting an independent crypto guide cited in AI search
How the Hyperliquid Guide, an independent reference with 50,000-plus readers, became a source AI engines quote, using the GEO checklist we run for clients.
Case study: building CarSnipe into a Marketplace alert engine
The engineering behind CarSnipe: a Chrome and Firefox extension that watches Facebook Marketplace from the buyer's own browser, paired to a Telegram bot, checking every 3 minutes on Pro.
How pages get cited in AI search: the six levers that decide it
What decides whether ChatGPT, Perplexity, or Google AI Overviews quote a page: crawler access, server-rendered HTML, liftable passages, schema, snippet caps, and llms.txt.