Musings

I went looking for one of my own three failure modes — a brand known to AI but rarely picked by it — across three unrelated shopping categories, and published what I found either way.

I locked in a bet that repeat AI answers would show more spread than pure chance, before collecting any data. They didn't — and two weeks later, most brand-recommendation rates were indistinguishable from that same chance-sized band, under a test that could only catch a swing bigger than about 30 points.

Chasing query fan-outs — writing content to cover every hidden sub-question an AI asks itself before it answers — is widespread SEO advice. The published evidence does not support it: the mechanism is fifty years old, and the coverage data show it barely moves whether a page gets cited.

I asked all three engines' own APIs the same skincare questions, twice, on separate days, and registered one number before looking: how much their real recommendations overlap. The answer is yes, almost entirely — and the one place they didn't agree turns out to be a coin flip for all three, not a real difference of opinion.

I took the seven assumptions underneath my own measurement to the published research and tried to break them. One survived unqualified. The one that broke was mine: a page an AI read is not proof the page shaped what it said.

A re-analysis of 8,160 measurements I already held, turned on my own two-run reliability guarantee. Holding the wording still across both runs left a whole class of error untested, and the answer forces a change to the instrument.

I asked ChatGPT and Claude to recommend home loan lenders across 8 Australian questions, registered its checks before looking, and found the mid-tier lenders holding a Canstar award — Macquarie and Unloan — appear at 2.4× the rate of those without.

I re-analyzed 8,160 call-and-brand rows to ask whether Reddit being in what an AI reads is linked to the brands it recommends. Reddit was in the mix in only 4.04% of calls, and where it was, recommendations barely moved — a near-null +0.0024. On this evidence, "do Reddit" isn't a lever you can act on.

I logged every URL ChatGPT fetched before naming a skincare brand, registered one number before looking, and found that 88.6% of what the model read was third-party pages — not brand sites. One word in the question swings the own-site-fetch rate by a factor of ten.

Your AI visibility score is a point estimate drawn from a distribution you don't know the shape of. The instability brands are racing to fix is structural — and the standard playbook is aimed at the wrong layer.

I read all ten studies that have measured whether schema markup wins AI citations, and checked every number against its original source. The studies that found an effect skipped the controls. The best-controlled found nothing, or a small negative.

AI made competence nearly free. The returns now flow to the two things it can't give you — the judgment to catch it when it's confidently wrong, and imagination it doesn't already contain.

Most content gap analysis stops at keyword coverage. The structural layer — internal links, anchor text, equity distribution — is where the real gaps live.

Sales and marketing are not enemies. The boundary is. The most expensive fiction in modern B2B is the belief that marketing and sales are separate realities rather than one commercial system with two sets of excuses.

After two decades of optimising content for global discoverability, AI may be forcing us to rethink what discoverability means in the first place.

As machines become the primary interpreters of information, organizations face a new competitive challenge: remaining legible under machine interpretation, not just human navigation.

What Google's guidance on AI search actually says when you read it as platform strategy, not publishing advice.

Stripping away the tactics, the visibility outcomes, and the theatre of recommendations — what does SEO actually mean?

A 16-step framework for the perfect AI SEO landing page just walked into a pub. Nobody asked who the human was.

How the modern web collapsed information and commerce into one surface — and what it would take to build the distinction back in.

The framework is correct. The problem is sequencing — and who goes and tells whom.