Everything published so far
Numbered in the order it was written. The series builds, so the early ones are the place to start.
- Where AI Actually Fits in a Content Programme You Already Run 1,557 words AI belongs at the three or four genuinely constrained steps in an existing workflow — research synthesis, structural drafting, repurposing, reporting — not across the whole pipeline; maps a standard programme and marks the insertion points worth testing first.
- Building a Brand Voice Brief an LLM Can Actually Follow 1,890 words Adjective lists like 'confident, friendly, human' produce nothing; models respond to constraints, banned constructions, sentence-length rules and annotated before/after examples. Shows how to convert an existing tone-of-voice deck into machine-usable rules.
- Write a One-Page AI Usage Policy Before You Scale Any Tool 1,837 words A short policy written in week one prevents the shadow tool stack that costs six months to unpick later; covers the five decisions a content lead must make (approved tools, data classes, disclosure, sign-off, accountability) on a single page.
- Choosing Your Stack: One Model, Two Tools, No Sprawl 1,523 words Small teams get better output from one frontier model used deeply than from six specialist SaaS wrappers used shallowly; sets a selection test based on context handling, data terms and exit cost rather than feature lists.
- How Generative Engines Decide What to Cite 1,566 words Citation is a retrieval-and-selection problem, not a ranking one: explains the passage retrieval, source corroboration and extractability factors that decide which page gets quoted, and what that changes about how you write.
- Baseline Your Content Performance Before You Introduce AI 1,545 words Without a pre-AI baseline of cost per asset, cycle time and per-piece performance, every later claim about AI's impact is unfalsifiable; gives the specific twelve numbers to capture before the first AI-assisted brief ships.
- Auditing the Content You Already Have Before You Point AI at It 1,684 words AI makes content audits cheap enough to be worth doing properly, but only if you classify assets by decay, cannibalisation and update cost first — the audit determines whether new production or refresh gets the AI investment.
- Briefing AI the Way You'd Brief a Good Freelancer 1,742 words The quality ceiling of an AI draft is set entirely by the brief, and the best brief format is the one you'd hand a paid contractor: audience state, the argument, what's out of scope, the sources, the proof. Prompt tricks matter far less.
- The Repurposing Chain: Turning One Pillar Piece Into Nine Assets 1,370 words Repurposing fails when each derivative is a compression of the parent; the chain works when each asset is re-argued for its channel from the same source material, with a defined order that front-loads the highest-distribution formats.
- Structuring Pages So AI Assistants Can Lift Your Answers 1,756 words Extractable content has a specific shape — self-contained passages, claim-then-evidence order, question-shaped headings, no anaphora across sections. Shows how to restructure a typical blog post without hollowing it out for human readers.
- The Metrics That Change the Moment AI Enters Production 1,457 words Volume and cost-per-post stop being meaningful once production is cheap; the measures that start mattering are per-asset yield, editorial rework rate and share of assisted output that survives to publication.
- Disclosure: When to Tell Readers AI Was Involved 1,681 words Blanket disclosure on every AI-touched asset is noise; proposes a materiality threshold based on whether AI shaped claims, voice or persona, and shows where UK advertising and platform rules make disclosure non-optional.
- Building a Quarterly Content Calendar With AI Without Losing the Plot 1,828 words AI is good at generating candidate topics and terrible at sequencing them; the calendar should be built by human-set narrative arcs and campaign dependencies, with AI filling gaps against the arc rather than proposing the arc.
- Draft, Then Interrogate: Getting Past the First Output 1,419 words The first output is a straw man to argue with, not a draft to edit; a structured interrogation pass — challenge each claim, demand the counter-argument, force specificity — is what separates usable assisted writing from generic output.
- Your First Custom GPT: What to Build and What to Skip 1,751 words Most teams build a general-purpose 'writer' GPT that underperforms a plain chat window; the assistants that earn their keep are narrow and repetitive — brief generation, voice enforcement, metadata batching. Includes a build order for a five-person team.
- Entity and Brand Signals: Becoming a Known Thing to a Model 1,732 words Models cite entities they can resolve with confidence; consistent naming, a coherent about-surface, third-party corroboration and structured identity data do more for AI visibility than another blog post.
- Measuring Time Saved Without Lying to Yourself 1,741 words Self-reported time savings are systematically inflated because editing and verification time is invisible; gives a lightweight time-tracking method that captures the rework tail and produces a defensible net figure.
- Copyright, Training Data and Who Owns the Draft: The UK Position 1,964 words UK law's computer-generated works provision and the absence of a broad TDM exception put content teams in a specific position that differs from the US; sets out what you can safely claim ownership of and what to put in freelancer contracts.
- Using AI for Audience Research That Isn't Persona Fan Fiction 1,678 words Asking a model to invent a persona produces a composite of marketing blogs; asking it to synthesise your own call transcripts, support tickets and review corpora produces something you can plan against. The input source is the whole argument.
- Editing AI Drafts: A Subtractive Checklist for Content Leads 1,773 words Editing assisted drafts is mostly deletion and evidence insertion, not line polishing; a fixed subtractive pass — cut hedges, cut throat-clearing, cut unsupported claims, then add proof — is faster and more repeatable than freeform editing.
- Knowledge Files and Context: Keeping AI On-Brand by Default 1,657 words Pasting context into every prompt doesn't scale across a team; a curated, versioned knowledge set — voice rules, product truths, claim library, past best work — makes on-brand the default output rather than the edited outcome.
- Original Data as Citation Bait 1,840 words In a market where synthesis is free, proprietary numbers are the only durable reason for a model to cite you; shows how a two-person team can run a publishable survey or dataset analysis each quarter on a realistic budget.
- Tracking Referrals From ChatGPT, Perplexity and Copilot in GA4 1,733 words AI assistant referrals arrive fragmented and partly cloaked; gives the channel grouping, regex set and landing-page annotations needed to see the traffic as one line item, plus what the numbers genuinely cannot tell you.
- UK GDPR and Putting Customer Data Into an LLM 1,903 words Most content-team data handling risk sits in three routine acts — pasting call transcripts, uploading CRM exports, and training on customer stories; covers lawful basis, data minimisation and the vendor settings that actually change your exposure.
- AI-Assisted Pillar and Cluster Mapping at Speed 1,877 words Clustering by keyword similarity produces topically tidy, commercially useless architecture; clustering by buyer question and decision stage, with AI handling the sorting, produces structures that convert and that models can navigate.
- Feeding AI Your Own Research So Drafts Aren't Hollow 1,724 words Generic output is an input problem: the workflow that fixes it puts primary material — interviews, data, internal docs, customer language — into context before a single word is drafted, and the drafting step becomes assembly rather than invention.
- Automating the Boring Bits With Zapier and Make 1,516 words The automations that pay back for content teams are administrative, not creative — brief routing, asset filing, publication checklists, stakeholder notifications; gives five builds that remove coordination overhead without touching the writing.