When my own website hadn't caught up with who I'm becoming

I spent a week making my own site tell the truth about who I am now, and kept finding places where it still described who I used to be.

For about five years my training page led with Microsoft Office and parked WordPress underneath, as if Office were the headline and WordPress an afterthought. That was true once. It stopped being true a long time ago, and I kept not noticing, the way you stop seeing the crooked picture frame in your own hallway. This week I straightened it. Then I spent the rest of the week finding other places where my own site was still describing a version of me I have been quietly leaving behind.

That is the real project underneath the housekeeping. I am a WordPress person who has spent thirty years building for the web, and I am becoming someone whose main work is teaching, and teaching about AI in particular. I did not announce that with a banner. I am doing it the way most real changes happen, one honest edit at a time. The trouble with editing yourself in public is that the old version is still sitting there in a hundred small places, and some of those places are machine-readable.

Reordering a training page so the real work leads

So I rewrote the training page from the headline down. WordPress leads now, and Microsoft training sits beneath it as its own track, present without apologizing for itself and without competing for a visitor's first thirty seconds. While I was in there I corrected the day rate, which now starts at $1,800 CAD. That is what an in-person training day actually costs once you account for prep, materials, travel, and the fact that the person in the room has shipped this work on real sites rather than reading from a vendor deck.

Changing one number meant chasing it through the schema, the seeders, and a handful of copy references that still quoted the old figure. Half-measures on price are worse than honest pricing. Anyone who runs the math on your day rate against your bio will spot the gap, and the gap reads as either carelessness or a discount you did not mean to give.

The smaller change matters more than the reorder. The page used to describe what I sell. It now describes, a little more, what I want to be known for. Those are two different pages, and the difference between them is the whole shift in miniature.

A stances post type, because opinions are not tutorials

I have opinions. About code quality, about how you walk a junior developer through their first production incident without making it the worst day of their career, and increasingly about how people should and should not use AI to do their thinking for them. For years those opinions lived inside ordinary blog posts, shoulder to shoulder with how-tos, with no signal that they were stated positions rather than instructions.

So I built a Stances post type. Each stance is one stated position on one topic, and the post meta tracks three things: the position, my confidence in it, and the date I last reviewed it. That last field matters most. An opinion from 2022 I have not revisited is not the same thing as one I would defend over coffee today, and the block pattern shows the freshness date right where a reader can see it.

I care about that surface more now than I would have a year ago, because several of the stances are about AI. If I am going to build a teaching practice around learning to learn with AI, the honest starting point is a public record of what I actually think, dated, so anyone can check whether I have updated it as the tools changed. You can read those positions on the stances page as they roll out. Ten are seeded and being reviewed against one question before they go live: would I still send a friend here if they were trying to make a real decision on this topic.

The finding that was about my own name

Then I pointed the same audit at my own schema that I would point at a client's, and it came back with two findings I did not enjoy. One was easy. Two plugins were each emitting a business object for the same address, quietly contradicting each other, so I deleted one and let the other own the node.

The second one was about who I am, and it stung.

My Person schema and my business schema cited the same employer in two different formats, one with the full legal name and one with the brand name. To a human those are obviously the same company. To a knowledge graph trying to merge two records into one identity, they are two different strings, so it either picks one and discards the other or declines to merge them at all. Either way the signal that says this is one person, with this history, gets weaker.

It is the same problem as changing your name and updating some accounts but not others. The bank still knows the old you, the new registry knows the new you, and for a while any system trying to line the two up decides you are two people who happen to share an address. I had done exactly that to myself, in the one format that is supposed to state plainly who I am and what I have done.

Here is why it landed harder than a normal bug. The whole point of structured data is to tell a search engine, and now an AI, who you are clearly enough that it does not have to guess. The clients I most want are the ones who check, and while they will not read two blocks of JSON themselves, the systems they trust to summarize me will, and a contradiction in your own data about your own identity is exactly the thing that makes a summary hedge.

I centralized the employer name so every emitter calls one function and there is one canonical string to be right. It took twenty minutes. It had sat wrong for years, on a site I look at every day, because I trusted a copy-paste from long ago and never re-read it the way a machine would.

That is the strange part of changing what you do for a living while you are still doing it in public. You update the copy first, because copy is where you think. The metadata, the schema, the old page orders, all the machine-readable sediment underneath, they lag. For a while the truest description of who you are becoming lives in your intentions and your newest edits, while the oldest version of you is still on file, still being read, still quietly insisting you are someone you have decided to stop being. The work is not glamorous. It is mostly finding those places one at a time and making them agree with the person you actually are now. If you would rather have someone find the place your own structured data contradicts itself before a prospect's AI assistant does, that is part of the work I do.

This is the same shift I keep circling in my writing on learning and working with AI, most directly in the two ways teaching with AI goes wrong. The website is just where the lag shows up first.

If you are somewhere in your own version of that shift, the teaching is the part I have moved to the front. Helping people learn to work with AI is the work I do now, and it started, the way these things do, with one honest edit.