writing — field notes
Two hundred and twenty-two posts were retired when this site was rebuilt — the instrument below runs their lights out, one by one. What publishes now is written slower: working notes on technical SEO, AI search, and building for machines that read.
corpus: 191 retired · 31 re-instrumented · 14 new · next note in the pen
the reset — 222 posts retired · 2026
still lit222
Not deleted — re-instrumented. Every light in the field running behind this site is one of the old URLs. The archive became the ambience. And the lights still burning? The 31 recordings brought back — each lamp is its link.
191 urls → /writing/ · 31 back live
the pen — what happens next
pen warm · slow by design · rss will find you
2026-11-05 · ai search
The words from episode 2 of AI Search for Beginners (GEO, AEO, LLM SEO, AI SEO, AI visibility): what each one means, which bits are new, which bits are a new invoice, and every number in the episode with where it came from and when.
2026-11-04 · ai search
Do AI Overviews show on over 80% of Google searches? On questions and buying searches it's about that, in the UK too. Across all UK searches Ofcom counts about 30%. This is the written version of Second Opinion, episode 2, with every number from the video, where it came from and when, and all 40 questions I checked.
2026-10-30 · ai search
Do you need GEO now? Not as a separate job. AI search is real, but the work is the SEO you'd be doing anyway, and pages that rank hold their place in AI Overviews longer. This is the written version of Second Opinion, episode 1, with every number from the video and where it came from.
2026-10-30 · ai search
The checklist from episode 1 of AI Search for Beginners (what AI search actually is), plus every number in the episode with where it came from and when.
2026-10-10 · technical seo
What a WordPress SEO consultant actually finds — the server in front, the builder, the plugins, the edge — with the readouts from my case files, what I ship inside WordPress, and how to tell a consultant from an agency, a freelancer or a plugin.
2026-10-10 · ecommerce seo
Shopify does more SEO for you than any platform and still leaves the half that decides rankings: collection architecture, duplicate surfaces, apps, rendering, migrations, the agent layer. What a consultant fixes, what an agency sells, and what an app can't.
2026-10-09 · technical seo
Technical SEO is the engineering that lets machines read, trust and rank a website: access, crawl, render, index, speed, structure. What it covers, what it doesn't, how it changed when the readers multiplied, and how to tell if yours is broken.
2026-10-09 · ai systems
An MCP server is a small program that gives an AI model a set of named tools it can call. What the Model Context Protocol is, what a server actually does, how to build one, what it means for a business website, and the one running behind this page.
2026-10-09 · technical seo
The technical SEO audit checklist I run on every engagement — access, crawl, render, index, speed, structured data, AI-search readiness, logs — with what each item costs you when it's wrong, and the order to do it in.
2026-10-09 · technical seo
A site migration loses rankings in three predictable ways. The SEO checklist I run on every replatform, redesign and domain move — freeze the inventory, generate and test the redirect map, watch the flip — with the evidence from migrations that went right and one that was blamed for a collapse it didn't cause.
2026-10-09 · structured data
What schema markup is, what it does and doesn't do for rankings, why most sites' structured data is a scatter of islands no machine can join, and how to write one connected graph that Google and the AI engines can read, validate and trust.
2026-10-09 · ai search
AI Overviews cite pages Google can fetch, parse and trust. What an AI Overview is, how the citations are chosen, the five things that decide whether your page is eligible, and how to measure any of it when Search Console won't separate the clicks.
2026-10-09 · ai search
GEO is the work of becoming the source an AI answer engine cites. Most of it is engineering, not copywriting. A plain definition, the four layers that decide citation, the things sold as GEO that aren't, and a sequence you can run this month.
2026-08-27 · voiced · ai search
Agents question sites now — across pages, feeds and endpoints, against everything else on the record. I expect cross-surface consistency to become a scored signal. The page becomes a witness.
rec 112026-08-24 · re-instrumented · voiced
Reddit's ChatGPT citation share fell 86% in a week and nobody can fully explain it. AI search has a measurement problem because it has an instrumentation problem — and the industry keeps shipping dashboards anyway.
rec 132026-08-17 · re-instrumented · voiced
The AI-search conversation happens at the answer layer while the actual decisions happen in the plumbing — retrieval, indices, protocols, registries. A map of the infrastructure underneath the citations.
rec 152026-08-10 · re-instrumented · voiced
A client wanted a five-year-old news mention 'off Google.' Three removals later, the piece maps what suppression actually costs in an AI-visibility world — and what to do instead.
rec 202026-08-09 · re-instrumented · voiced
The prompt-screenshot slide is in every SEO deck: fire prompts, count brand mentions, call it visibility. Why prompt tracking measures the wrong thing — and what the models actually consult.
rec 212026-08-06 · re-instrumented · voiced
Agents are being framed as a new audience to optimise for. The real story is infrastructure economics: who pays for the compute, the crawls, and the open web that agents consume without landing on it.
rec 142026-08-03 · re-instrumented · voiced
Nearly two decades of SEO assumed the page is the unit of optimisation. AI search retrieves entities, not URLs — and that quietly rewrites what building authority even means.
rec 282026-07-22 · re-instrumented · voiced
GEO, AEO, AIO, now Brand Sovereignty — new acronyms for one problem: nobody owns how machines describe your business. The case for treating knowledge governance as an operating function, not a campaign.
rec 032026-07-20 · re-instrumented · voiced
Impressions climb while clicks flatten; trackers report volatility Google denies. The measurement layer of SEO has decoupled from the reality it describes — the instruments are lying, and the industry pretends they aren't.
rec 092026-07-13 · re-instrumented · voiced
Two trade-press stories nobody reads together: WebMCP momentum and agent-liability questions. The playbook for the agentic web is measurement plus liability, not optimisation alone.
rec 102026-07-06 · re-instrumented · voiced
Apple shipped an MCP server for Safari — agents get the live DOM, network waterfall, and accessibility tree. Covered as a dev tool; actually a preview of how agentic browsing reads your site.
rec 242026-07-04 · re-instrumented · voiced
Two investigations into a marketing firm vanished from Google after anonymous DMCA complaints citing unrelated material. Fake takedowns are now an SEO attack surface — cheap, effective, and barely policed.
rec 262026-06-30 · re-instrumented · voiced
Prompt-counting studies infer ChatGPT's machinery from its outputs. Watching the actual network traffic shows the retrieval routing underneath — and why brand strength decides which route you get.
rec 062026-06-29 · re-instrumented · voiced
Pichai promised agentic search and outbound traffic in the same breath. Only one of those can dominate. Why brand becomes the compounding asset when agents complete the task and never hand you the visit.
rec 182026-06-27 · re-instrumented · voiced
Google now classes manipulating generative AI responses as spam. The deeper story: retrieval poisoning works because agent architectures trust what they fetch — a flaw policy can't patch.
rec 082026-06-22 · re-instrumented · voiced
Cloudflare, Shopify, Stripe, Google, and Netlify all rebuilt for non-human visitors in the same six months. The agent-ready web is not a trend forecast — it already shipped, and this maps the moves.
rec 312026-06-18 · re-instrumented · voiced · ai search
llms.txt in plain terms: what the file is, what it can and cannot do for AI visibility, and a copy-ready implementation that costs ten minutes — no magic, one real problem solved.
rec 302026-06-18 · re-instrumented · voiced · ai search
ChatGPT has no ranking algorithm to game. What it actually looks at when it surfaces sources, what you control, what you don't — and where to spend the effort.
rec 122026-06-15 · re-instrumented · voiced
Sixty-eight percent of US Google searches now end without a click anywhere. The composition underneath the headline — branded vs transactional — decides whether that number should scare you.
rec 072026-06-01 · re-instrumented · voiced
AI bot traffic grew 187% while human traffic grew 3.1%; publisher referrals collapsed. The operational playbook for UK businesses serving machine media — what to instrument, publish, and stop doing.
rec 252026-06-01 · re-instrumented · voiced
EntityMap joins schema extensions, NLWeb, and llms.txt on the shelf of structured-data fixes for AI attribution. The problem is generation, not data cleanliness — and cleaner markup can't solve it.
rec 192026-05-26 · re-instrumented · voiced
An arXiv paper found subjective superlatives capture 100% of the optimisation effect in LLM tool registries. Registries are the new search index — unregulated, gameable, and already being gamed.
rec 042026-05-25 · re-instrumented · voiced
The three most useful AI-search measurement releases this year came from a browser company, a CDN, and an academic team — not the SEO industry. What outsiders see that vendor dashboards don't.
rec 292026-05-23 · re-instrumented · voiced
Mike King is running real experiments, publishing data with confidence intervals, and accusing Google's GEO guidance of being naive and self-serving. The industry nods and moves on. That's the wrong response.
rec 172026-05-23 · re-instrumented · voiced
GEO advice keeps treating machine visibility as a content problem. It's structural: rendering, markup, extraction cost. If a parser can't cheaply read the page, the writing never gets its chance.
rec 162026-05-16 · re-instrumented · voiced
Google says AEO is still SEO; Ahrefs' controlled tests say schema doesn't move citations; GA4 quietly adds an AI Assistant channel. The citation trap: measuring substitution as if it were traffic.
rec 232026-05-13 · re-instrumented · voiced
1,400 tickets in 18 months, almost none shipped, team laid off for 'nothing being done.' The SEO implementation gap isn't a prioritisation problem — it's organisational design.
rec 222026-05-12 · re-instrumented · voiced
Conductor surveyed 250+ executives: the top AI-search strategy at every maturity level is scaling AI content generation. That consensus is a coordination failure dressed as a strategy.
rec 052026-05-11 · re-instrumented · voiced
The SEO loop was always page-in, ranking-out. Protocols — MCP, commerce rails, agent registries — are eating that stack from below. The 2026 map of what replaces the page-shaped job.
rec 022026-05-04 · re-instrumented · voiced
Your website is no longer a destination — it's a source. Why the source thesis is the most important argument in search right now, and what it demands of everyone still optimising destinations.
rec 272025-11-12 · re-instrumented · voiced
A simple AEO/GEO reporting system for SMEs: what to measure weekly, which numbers are real (assistant referrals, cited pages, lead source notes), and which vanity metrics to refuse.
rec 012022-02-03 · re-instrumented · voiced
Buried on Google's no-results page is a Yeti with a fishing rod. How to trigger the search engine's most charming easter egg — and why the empty state got the best design.