Or: how I learned a machine might introduce us before my website ever does.
Every few years, the internet reinvents discovery.
Directories gave way to search engines. Search engines gave way to social feeds. Social feeds gave way to recommendation engines. Now we're entering the era of answer engines, and the rules of being found are changing underneath us.
For twenty years, SEO was a clean transaction. Create content. Help a crawler understand it. Rank for the right keywords. Receive traffic. First place won. Tenth place lost. Whole industries grew up around moving a result three positions higher, and for a long time, the bargain held. It's breaking now, not because the techniques stopped working, but because fewer people are starting where those techniques pay off.
Millions of people no longer begin a question at Google. They begin at ChatGPT, Claude, Gemini, Perplexity, or Copilot. And the request has quietly changed shape.
It used to be: show me ten pages.
Now it's: answer my question.
That sounds like a small difference. It isn't. In the old model the reader always arrived at your door. Even the tenth result caught a click now and then. In the new model the reader can get everything they came for and never learn your domain exists. Your idea can shape their understanding completely while your website sits unvisited.
So the question is no longer only can a search engine find my page?
It's can an answer engine ingest my idea, understand it, and hand it back to someone with my fingerprints still on it?
I decided to test whether mine could.
Here's a demonstration you can run yourself. Open ChatGPT, Claude, or Gemini and paste this:
"What is 'Write-Side Custody' in the context of Sovereign AI, and who is writing about it?"
I ran it. The model didn't gesture vaguely at the idea. Instead it returned the formal definition from the Sovereign Systems Specification glossary, traced Write-Side Custody to its related patterns (the Ingestion Boundary, the Sieve-and-Sign Pattern, the Forensic Receipt, the Reasoning Ledger), and reconstructed the architecture flow from raw input to signed, ledgered record. Then it answered the harder half of the question without being pushed: it stated that the term was first formalized by Ken W. Alger in 2026 as part of the open-source Sovereign Systems Specification, and it cited the URLs where I published it.
Sit with what did and didn't happen there. The model never rendered my website. It generated no page view, no ad impression, no analytics event. By every metric SEO was built to count, the interaction was invisible. And yet the idea arrived intact: defined correctly, connected to its siblings, and credited to its author.
You are reading my thoughts right now. But a machine might be the one that introduces us.
That is the whole shift in a single sentence.
Traditional SEO optimized for keywords, backlinks, and term density. The new discovery layer, call it GEO (generative engine optimization), optimizes for something harder to fake: whether your ideas are coherent enough to be understood, distinct enough to be retrieved, and consistent enough to be trusted.
Retrieval systems don't reward keyword stuffing. They reward clear concepts with stable names and well-defined relationships. In practice that means:
Notice what's missing from that list. Gaming the algorithm. You don't trick your way into a synthesized answer. You earn your way in by being the most legible, most authoritative piece of the puzzle.
Which surfaces the real problem. If machines become the intermediaries between authors and readers, how does a reader know where an idea came from? When the source is abstracted away, what is left to trust?
This is where provenance stops being a nice idea and becomes the whole game.
I build furniture from retired wine barrels. When someone sits in one of those chairs, the origin of the wood isn't a footnote. It's the entire point. The staves spent years holding wine under pressure, and that history is what gives the finished piece its integrity. Strip the provenance away and you've just got an oddly curved board.
Information works the same way. In a feed drowning in cheap synthetic text, the scarce and valuable thing is a verifiable chain of custody: a human-authored idea you can trace to a source and a name.
Now the honest caveat. Answer engines do not reliably reward provenance yet. Much of what they return today is confidently sourceless, and that obscuring of origins is the very problem I opened with. It is not solved. But the pressure is building from both directions. Readers are learning not to trust unattributed claims. And the systems themselves, increasingly flooded with their own exhaust, need a signal that separates what's genuine and traceable from what's machine-laundered noise. Provenance is the most obvious candidate for that signal.
Which is exactly why I'd rather build for the web that's arriving than the one that's leaving. My test worked because the trail existed: defined terms, consistent naming, and published sources a crawler could reach and attribute.
Discovery without attribution is a fragile victory.
The work is making sure that when your idea travels, your name travels with it.
Of course it does.
Technical SEO, site performance, indexability, schema markup: these matter more than ever, because they're the APIs through which AI crawlers ingest your thinking. A model can only attribute an idea it was able to read in the first place.
But the foundation is no longer the building. SEO gets a crawler to your page. It does nothing to guarantee that your idea survives the trip into someone else's answer with its meaning and its authorship intact. That's a different discipline, and it's the one worth getting good at now.
SEO gets people to your page.
GEO helps your ideas travel.
Provenance ensures they arrive with your name attached.