Articles
Walking the Walk
I sell content operations built with AI and governed by humans. It would be odd to write about that and then not use it. Everything here comes out of the same pipeline I build for clients, and every piece carries a note explaining exactly how.
How This Section Works
Every article goes through the same run: a research pass with the sources written down, a drafting pass against a written standard, then gates that check structure, claims and anything with a compliance angle before the thing can publish at all.
Then I edit it. I sign it. If there's a number on this site, I've checked it.
And that's the whole argument of this site, made in public, on a schedule, where you can audit it. The note at the foot of each piece isn't a legal nicety. It's the proof.
Pieces
17 August 2026 · 13 min read
Seven detectors read essays by non-native English speakers and called 61% of them machine-written. Then Claude started watermarking what it writes, which is a different thing altogether, and the only way that comes off a page is an editor rewriting most of it.
17 August 2026 · 13 min read
A model makes the same few mistakes in every draft, forever, so the job isn't proofreading. Three files run against every piece, and anything you fix twice becomes a rule that sticks. The six tells worth hunting, and one paragraph shown before and after.
16 August 2026 · 14 min read
Every guide to this tells you which tool to buy and which holidays to post around. None of them tell you what actually goes in the thing. Topical clusters, a site structure that doesn't sprawl, and a named reader against every slot.
Updated 11 August 2026 · 18 min read
Every regulated publisher owns a compliance document, and an incident log proving nobody read it. Three tests tell a control apart from a document. Most programmes fail the second and have never thought about the third.
Updated 11 August 2026 · 16 min read
Drafting gets cheap fast. The review you already had doesn't scale up to meet it, and that gap is the saving. With the arithmetic, the numbers worth ignoring, and the one objection I take seriously.
Updated 11 August 2026 · 11 min read
Everyone checks the jurisdiction list. Nobody checks the sentence. Four ways market gating leaks, including the one that belongs to the content team rather than engineering.
12 August 2026 · 24 min read
Google says E-E-A-T isn't a ranking factor and expert bylines don't magically help. Both are true, and everyone draws the wrong conclusion from them. Six people already on your payroll know things no writer can research, and there is a fifteen-minute red pen that gets it out of them without slowing anything down.
11 August 2026 · 17 min read
Google's policy doesn't care how the page got written, and says so in writing. What 331,000 pages show about how AI content really ranks, the three spam policies that do bite, eight things that measurably improve output, and why Claude's watermark isn't a problem.
Updated 11 August 2026 · 19 min read
Slop is not something a model does to you. It's what ungoverned generation looks like. The four files that run this site, what they forbid, how to audit an archive for it, and the two checks this article failed before you saw it.
11 August 2026 · 5 min read
Every de-AI checklist only takes things away, and run at full strength they hand you a second artificial voice that nobody is screening for yet. The first draft of this piece failed its own argument.
11 August 2026 · 5 min read
A policy with one dial can only get turned up. Far enough and it starts deleting facts that four independent sources agree on, and a page with no figure loses to a page that has one.
11 August 2026 · 13 min read
ChatGPT, Gemini and Claude Code, compared on what actually decides it for a content team, with the test to run before you commit any budget. Three lineup or price changes in three weeks is why the spec table is the wrong table.
11 August 2026 · 17 min read
I stopped hiring freelancers when the work started turning up as raw Gemini output. The choice was never human or machine. It's governed AI or ungoverned AI, and one of those you can't audit.