Field reports
Story-shaped pages from production: what happened, what it cost, what we changed. The emotional entry points of the library.
Adding a Citable Content Layer to a Live WordPress DomainThe migration that never migrated: 1 Worker, an explicit allowlist of 12 content path prefixes, 0 existing URLs changed, rollback by greying DNS [our data].AI Overviews Cut Your Blog Traffic in Half: What Happens NextWhen AI Overviews land on your queries, clicks fall while impressions hold — Ahrefs measured −34.5% CTR. A field report on turning summaries into citations.PlaceholderDoes llms.txt Work? A 90-Day Watch on 3 Sites, Zero Fetches So FarWe serve llms.txt on 3 production sites and log every request through a 90-day watch. So far: 0 fetches from major AI engine crawlers. A running result.Fact-Auditing 400+ Pages: What Three Audits CaughtThree fact audits: 445 claims checked and 40 corrected, 461 figures recomputed with 1 error, and 1 fabricated legal claim caught before publish [our data].Four Pages, Three Answers: When Your Own Library Contradicts ItselfA day-one review of 67 new pages found 4 answering 1 question 3 different ways — on the build whose whole strategy is correcting contradictions [our data].Our First AI Overview Citation: A Day-3 Glossary PageA glossary page shipped on day 3 of a greenfield build won the AI Overview for its defining term within days — brand link included. The field report.Phase Zero: Clearing Inherited Trust Liabilities Before Building AuthorityBefore the apex pointed at 1 new page: fabricated review markup removed, hidden text stripped, 4 contradictory approval-rate claims replaced by 1 [our data].Publishing 100+ Pages in a Week: What Broke and What HeldA 167-page, five-cluster content map shipped in 2 days on a greenfield build — and 74 pages launched with zero inbound links. The git-dated field report.The 2,895-Line Robots.txt We Never WroteA pre-cutover audit found Cloudflare serving a 2,895-line robots.txt blocking 8 AI crawlers — 1 day before the apex flip, while git said allow [our data].Three Brands, One Playbook: What Transferred and What Didn'tThe same contract and gates across 3 verticals and 2 countries: content maps in 2 days, 188/298/106 files — and the parts rebuilt every time [our data].We Audited Our Own Case-Study Register and Found 2 Wrong ClaimsA 2026-08-21 pass against deployed code found 2 defects in our own first-party record. Nothing wrong shipped. What drifted, and the rule we adopted.What Our Automated GSC Miner Actually FoundSearch Console returned mineable queries 4 days after first commit; by day 21 the weekly miner shipped a comparison page and 3 snippet fixes [our data].When a Platform's Own AI Fabricated Coverage ClaimsA publishing platform's own AI fabricated coverage claims for a YMYL brand — false, against case law, and matching 0 of our numeric gates [our data].When LLM Extraction Invented a Price LadderAn OEM page carried 1 lease offer at $479/month. Extraction returned a fabricated $249–$399 ladder. The gate that landed the same day [our data].