DE
Research and notes
Articles on agentic SEO, connecting AI with Claude, llms.txt and machine-readable infrastructure.
This blog documents measurements and projects: how readable websites are for AI agents, what llms.txt, structured data and crawler rules actually do, how I connect AI to my own tools via MCP, and what a case study with real numbers shows. Some posts come from the SOVP research. I stopped developing the protocol in October 2026; those posts stay online as an archive.
Every post is a log entry with a running number, date and reading time. Each one opens with a short summary for readers and machines. When I correct a published post, the correction appears in the text with its date instead of disappearing quietly. Use the categories to filter by topic.
What an MCP server is and how I connected two in-house SEO tools to Claude: guardrails inside the tool, one evening from 49 to 68 keywords for about USD 0.23, and two bugs in my own code that surfaced along the way.
[READ PROTOCOL]What an MCP server is and how I connected two in-house SEO tools to Claude: guardrails inside the tool, one evening from 49 to 68 keywords for about USD 0.23, and two bugs in my own code that surfaced along the way.
A new open-source bridge translates a SOVP attestation result into a signed AgenTrust TRACE v0.2 Trust Record. Ed25519-signed, Level 0 TRACE conformant, with a documented tamper-and-reject test in CI.
42% of 400 measurable domains returned zero valid JSON-LD; 82% of 559 measurements served no llms.txt. InfraCheck is an open-source CLI that checks any domain for five machine-readable signals — JSON-LD, llms.txt, robots.txt, HTTP headers, SSL/TLS — and returns a binary PASS/FAIL in seconds.
llms.txt is a curated Markdown file at your domain root that gives LLMs a machine-readable map of your key content. The format under specification v2, a full example, how it differs from robots.txt and sitemap.xml, and how to validate that it actually works.
SOVP batch scan of Germany's top 100 online retailers (EHI/ECDB 2025). QuickScan (June 2026 scoring): 0/100 CERTIFIED, 14/100 PARTIAL. FullScan: 0/100 CERTIFIED. AI Readiness cluster avg: 14. First ECOM100-DE baseline, June 2026.
The starting position of teleschmie.de on June 1, 2026: SOVP scan with Sovereign Score 60 (NOT_CONFORMANT) and a GSC baseline of 37,871 impressions, 310 clicks and 0.82% CTR. The sprint ended on August 18, 2026. An evaluation has not been published.
Google I/O 2026 confirmed what deterministic infrastructure validation already measured: llms.txt belongs at Layer 0. SOVP now scores it as a verified bonus parameter in the new agenticReadiness cluster.
draft-litzki-sovp is live on the IETF Datatracker. The protocol specifies decentralized validation of entity integrity for the agentic web. Python reference implementation is public on GitHub under Apache 2.0.
Google's experimental Web Bot Auth protocol replaces self-declared User-Agent strings with HTTP Message Signatures and JWKS directories. The same cryptographic verification model SOVP applies to infrastructure identity.
An llms.txt file signals intent. SOVP validates actual infrastructure integrity. The difference is CERTIFIED versus invisible.
Agentic commerce refers to commercial processes executed by autonomous AI agents. This article explains the technical foundations, enterprise system architecture requirements, and protocol standards such as SOVP.
No protocols in this category yet.
For other editors I write, among other things, about briefs, expert articles, technical infrastructure and AI agents. The list with publication, date and topic is on my profile page under Publications.