What I Actually Use AI For — and Why It's Not What You Think

In 1996 I started a folder on my desktop. I called it something forgettable — probably just "Health" or "Reference" — and I started filling it with articles, course notes, online material, anything that seemed clinically interesting or useful. The information superhighway was a novel concept then. AI was something I'd studied in a second-year psychology module in 1989 — symbolic logic, expert systems, the stuff that felt like science fiction even in the lecture theatre.

That folder became a collection. The collection became a practice. The practice became a methodology. And now, in 2026, the methodology has become a clinical website, four published books, a 42-tool clinical suite, a blog library of over 120 posts, and a testing framework built around five functional labs that together give a more complete picture of human biology than anything available in conventional medicine at the time that folder was started.

AI helped build parts of that — specifically, the version of AI that has emerged in the last few years as a genuinely capable collaborative thinking tool rather than the symbolic logic experiments I studied in first-year university. I want to be specific about what that means, because the discourse around AI tends toward two unhelpful poles: uncritical enthusiasm on one side, unfounded suspicion on the other. Neither maps onto what I've actually experienced.

What AI Is in This Context

The AI I work with is a large language model — a system trained on an enormous breadth of text that can engage with complex clinical, technical, and creative questions with a level of coherence and depth that is genuinely useful for a solo practitioner building a knowledge infrastructure without a team behind them.

It is not a replacement for clinical judgement. It does not see the client sitting across the video call, does not hear the hesitation in how someone describes their fatigue, does not integrate the ten-year history of someone's gut symptoms with the fact that they moved into an older property five years ago and the bloating started six months later. That integration is clinical work, and it remains mine.

What it can do is serve as a thinking partner — a collaborator that can hold a large amount of information simultaneously, help me articulate a clinical framework I've been developing implicitly over decades, build the written infrastructure that makes that framework accessible to people who need it, and challenge my thinking in ways that sharpen rather than replace it.

"The relationship between the clinician and the tool is the thing. The tool doesn't think for you. It thinks with you — and what comes out depends entirely on the quality of what you bring to the conversation."

The detective-health.com site — with its clinical diagram pages, the cascade diagrams, the Learn section, the GI-MAP interpretation guides, the DUTCH landing pages, the new specialist testing pages on SIBO and mycotoxins and environmental toxins — was built in that mode. I brought the clinical framework, the pattern recognition, the 37 years of case experience, the understanding of what questions matter. The AI helped translate that into written infrastructure at a scale and pace that would have taken years of solo writing to produce.

The 1996 Folder, Grown Up

The "font of all knowledge" concept I've been building since that desktop folder is what the site is becoming. Not a content mill. Not SEO-optimised health generic content. A clinical reference that reflects how I actually think about human physiology, functional testing, and the gap between what conventional medicine can offer and what the body is actually asking for.

Two recent clients found me through ChatGPT. Not through a Google search, not through social media, not through a referral. Through an AI system that was answering a question they'd typed — probably something like "functional medicine practitioner Edinburgh" or "GI-MAP testing UK" — and surfacing detective-health.com as a relevant resource. That is the long-form content strategy working in a way that didn't exist when that original folder was started.

It's also a validation of something I've believed since the beginning: that depth of clinical content, written clearly and honestly, will find the people who need it. The AI-mediated search channel is just a new mechanism for that finding to happen.

What Critical Thinking with AI Actually Looks Like

The folic acid post that generated 88,000 impressions on X in early July is a good example of the distinction I'm drawing. The argument in that post — that mandatory folic acid fortification fails people with MTHFR variants because it adds synthetic folic acid to a food supply without accounting for the significant proportion of the population who cannot efficiently convert it — is not an AI argument. It's a clinical argument I've been making in consultations for years. The AI helped me articulate it in a format that worked for a general audience and a social media thread. The clinical insight, the evidence base, the specific mechanism — those came from 37 years of practice and the clinical literature I've been reading since before most of the people who shared that post were born.

That's the distinction that matters. AI as a tool for expressing and distributing clinical knowledge is legitimate and useful. AI as a replacement for the clinical knowledge itself is something else entirely — and it's something I'm neither doing nor recommending.

The same principle applies to the test interpretation tools I've built — the Nutrition Integration Engine, the blood chemistry interpreter, the DUTCH pattern recognition layer. These encode my clinical frameworks. The AI helped build the infrastructure. The frameworks are mine, developed over decades, and they require clinical oversight to apply correctly. A tool without the clinician is just a number generator.

Why Mark Playne's Book Made Me Think

I came across Mark Playne's book AI & I through a connection in the health space. The timing was interesting — I was already deep in this AI collaboration project, already thinking about what it meant that I was building clinical infrastructure with a language model at a pace that would have been impossible alone.

What Playne does that most AI-in-health content doesn't is bring genuine clinical grounding to the question of how practitioners should engage with AI tools. Most of what gets written about AI and healthcare falls into one of two failure modes: breathless technology evangelism that doesn't understand clinical practice, or defensive institutional caution that doesn't understand what the technology can actually do. Playne writes from inside the clinical world, about the specific questions that matter to practitioners who are trying to use these tools well.

The questions he raises — about maintaining clinical reasoning, about the relationship between the practitioner's expertise and the tool's capabilities, about what AI can legitimately assist with and what it cannot — are exactly the questions I've been working through in practice. Reading someone who has articulated them clearly, and done so before the current wave of AI enthusiasm made every opinion on the topic sound like either a press release or a manifesto, was useful.

The questions Playne raises — about maintaining critical thinking, about the relationship between the person asking and the tool answering, about what happens when you press past the surface-level response and keep asking — are questions worth sitting with whether you're a practitioner, a researcher, or simply someone trying to navigate a health information landscape that has become increasingly difficult to read clearly.

The Positive Case — What This Collaboration Has Produced

I want to be concrete about this, because vague claims about AI productivity are as unhelpful as vague suspicion of it.

In the period I've been working with AI as a collaborative tool, detective-health.com has grown from approximately 41 pages to over 93 published URLs. The blog has over 120 posts covering clinical topics — CoQ10, oregano, reishi, HRT, ferritin ranges, MTHFR, folic acid fortification, metabolic typing, electrolytes, the gut-brain axis, HPA axis dysfunction, thyroid conversion problems, insulin resistance mechanisms. Four books are now on Amazon. A 42-tool clinical suite is live. The Gumroad catalogue has 25 products. The Mailchimp sequence has eight welcome emails running. The clinical diagram library has six cascade diagrams. A specialist testing section now covers SIBO, Metabolomix+, MycoTOX, and EnviroTOX Complete — tests that didn't have any public-facing explanation on the site a week ago.

None of that output is AI-generated clinical opinion presented as my own. All of it encodes clinical frameworks I've developed, clinical knowledge I've built over 37 years, and clinical positions I hold and can defend in a consultation room. The AI is the publishing infrastructure. The clinical content is mine.

That distinction matters because the legitimate concern about AI in health content is real: there is a significant and growing volume of health information online that was generated by AI systems without clinical oversight, that sounds authoritative, and that is either wrong, oversimplified, or misapplied. I am not contributing to that. I am using AI to make a larger body of clinical thinking more accessible to more people — which is what the original 1996 folder was always trying to do.

On the Politics of This

Some of the engagement I've received since the folic acid post went viral has come from people who are suspicious of AI on broadly political grounds — that it represents a concentration of power, that it's a tool of control, that it's a mechanism for homogenising thought. I understand those concerns and I don't dismiss them entirely.

But the AI I work with is a tool. It has no agenda regarding folic acid fortification policy. It has no position on government intervention in food supply. It helped me write a post arguing against a specific policy decision made in 2017 under a Conservative government, on the grounds that the decision was made without adequate consideration of individual genetic variation. The political valence of that argument — which has attracted responses from people across a wide ideological range — comes from the clinical evidence, not from the AI.

The force-for-good question is the interesting one. AI as a tool for making genuine clinical expertise more widely accessible — for democratising the kind of depth of investigation that was previously available only to those who could afford direct access to experienced practitioners — is a legitimate positive case. That's what I'm trying to build. Whether it succeeds depends on whether the clinical content is good, not on whether the tool used to publish it is viewed with suspicion or enthusiasm.

The Honest Position

I am not an AI evangelist. I am a functional diagnostic nutrition practitioner with 37 years of clinical experience who has found a tool that helps him do more with the knowledge he's built than he could do alone. The tool is useful when used with clinical judgement and honest attribution. It is not a replacement for either.

The font of all knowledge concept — the idea that a deep, well-organised, clinically honest body of content can serve as a resource for people who are trying to understand their health better than conventional medicine has helped them understand it — is worth building well. AI is helping me build it faster. The quality of what gets built depends entirely on what I bring to the collaboration.

That's the honest position. It's not a dark art. It's not nefarious. It's a tool, used by someone who knows what they're doing, in service of a project that existed before the tool did.