Essays on voice, craft, and scaling without sounding like everyone else.
Opinionated, occasionally long, never generic. New posts every couple of weeks.
May 14, 2026
Claude vs ChatGPT for content writing in 2026: an honest side-by-side
Claude and ChatGPT are different writing tools in 2026. Different default voice, different system prompt adherence, different refusal patterns, different context window behavior. The honest answer to which is better for content writing is conditional on use case. Here is the design-decision-level side-by-side, plus the writer-side use-case mapping.
May 14, 2026
AI detection tools tested: what Originality.ai, GPTZero, ZeroGPT, Copyleaks, and Winston AI actually catch in 2026
AI detection tools in 2026 are caught between a real use case (catching unedited AI-drafted content) and a real failure mode (false-positive flagging long-form essayists and AI-edited human writing). Originality.ai, GPTZero, ZeroGPT, Copyleaks, and Winston AI each claim high accuracy, each catches a subset of what they claim, and the false-positive problem is the central honest observation. Here is the skeptical-honest read.
May 13, 2026
How to avoid the AI tells: a writer's checklist for 2026
How to avoid the AI tells in your writing in 2026 is the remediation companion to the diagnostic. Nine canonical tells become nine active-avoidance practices, each with constructed before/after examples. Em-dash density, AI vocabulary cluster, symmetric two-clause hook, the not-just-X-but-Y frame, beige bullet middle, generic closing CTA, symmetric paragraph rhythm, voice-flat coherence, missing taboos. Plus the two-minute pre-publish scan.
May 13, 2026
Can your audience tell you're using AI? An honest 2026 analysis
Can your audience tell you're using AI to write your content? The honest answer in 2026 is conditional, and the conditional answer is the article's contribution. Audiences detect at three different levels (explicit, implicit, unaware), care at different levels (trust-degradation patterns, AI-assisted vs AI-drafted), and the asymmetry between the levels is what matters operationally. No fabricated detection-rate percentages; directional language throughout.
May 12, 2026
How to train AI on your writing voice: the technical breakdown
How to train AI on your writing voice depends on which technical approach you use. Three categories: prompting a general LLM with your writing samples (cheap, weak, hits a ceiling by paragraph three), fine-tuning an open-weight base model on your corpus (expensive, partial, hard to operate), or voice profiling on a multi-signal training corpus across the 10 signals of voice (the approach that actually produces output in your style). Side-by-side technical comparison, when each is worth doing, and the ceiling each one hits.
May 12, 2026
Why all AI-written tweets sound the same (and how to actually fix it)
The reason why AI content sounds generic is mechanical, but the operating reason most explanations skip is that general-purpose AI tools are optimizing for helpful-assistant output, which is the opposite of voice. The five-line prescription for actually fixing it: stop trying to prompt your way out of it, train a dedicated voice model on your full profile, document a voice doc and taboo list, score every generation against your baseline, and use the tool as a partner.
May 12, 2026
The words AI overuses (and how to ban them from your writing forever)
AI vocabulary is the second-fastest tell after the em-dash. Here is the full list of words AI overuses in 2026 (leverage, delve, unlock, navigate, harness, foster, elevate, embark, robust, seamless, comprehensive, holistic, plus the frame openers and bridges), a substitution table for each, a three-tier taboo system you can install in your drafting workflow, why banning the words is necessary but not sufficient, and why the list keeps changing as new models ship.
May 12, 2026
How to spot AI-generated content in 2026: the em-dash and 8 other tells
How to spot AI content in 2026: the em-dash is the canonical tell, but it is one of nine. Here is the full diagnostic: eight more vocabulary, structure, and rhythm signs of AI writing, two common false positives, why you should not stop using em-dashes to dodge detection, and the byline-removal test that catches the rest.
May 12, 2026
Accessible images on X, voice-first: the accessibility floor under alt-text and why voice-first creators ship it consistently
Image accessibility on X is bigger than alt-text. Color contrast, in-image text, video captions, and screenshot legibility all matter for blind and low-vision readers and for AI assistants reading the post. The voice-first commitment to specificity in writing extends naturally to specificity in image accessibility. Here's the floor.
May 11, 2026
Alt-text on X: the AEO move most creators skip, done in voice
Alt-text on X serves two audiences: visually impaired readers and AI assistants indexing the post. Most creators skip it. A small minority keyword-stuffs it. Here's the voice-first version that serves both audiences without doing either job badly.
May 11, 2026
How to use AI for tweet writing without losing your voice
To use AI for tweet writing without losing your voice, keep the AI in the ideation, outline, and editing roles and never let it write the post from a blank prompt. Bring your own idea, generate options and rewrite aggressively, check your cumulative output for drift weekly, and reach for a voice-trained tool (with a voice match score) once you are posting at volume. The working playbook, the prompts that help, and the drift test.
May 11, 2026
Grok on X: what it does well, what to use somewhere else
Grok is the only AI assistant with native real-time access to X. That's genuinely useful for research and trend reading. But the things you'd actually want an AI for, drafting in your style, are the things Grok is worst at. Here's the honest review.