Last week I got a spam email that opened with “Hi ATLANTIC WOOD INDUSTRIES INC,” in that comma-spliced way spam always has. I’m not a lumber company. I have never been a lumber company. The same week, during a screen share at work, a Google profile tied to my account introduced me to a colleague the same way. Two systems that have never met, agreeing on the same wrong fact about who I am.
When two strangers repeat the same false thing, they’re reading from the same source. So I went looking for it.
Nearly twenty years ago, I built a website for a treated-lumber company. While the work was in progress I parked it on a subdomain of my portfolio, the kind of dev link you spin up without thinking (and kill when the project ships). I never killed it. The project ended, the client moved on, and their website kept quietly serving from my domain… for years. Nothing I could see pointed to it, so as far as I was concerned, it didn’t exist.
The machines saw it fine. Somewhere in those years, the companies that crawl the web for business records found the subdomain and drew the obvious conclusion: if this domain serves this company’s website, then this domain must BE this company. One of them still lists my portfolio as the lumber company’s official site. Others copied the record, because that’s what records are for. A shortcut I made in my twenties had hardened into a fact, held in databases I’ve never heard of, waiting.
For most of those twenty years, none of it mattered. It was a wrong row in a spreadsheet nobody I know would ever open. The old web was full of stale facts like mine, and they were mostly harmless, because a human stood between every record and every conclusion. Anyone who actually looked me up would land on my portfolio, see the work, and move on. The wrong answer existed, but it had to get past a person, and it never did.
That filter is gone. If you ask an AI-backed search about a person or a company now, it doesn’t hand you ten links to weigh for yourself. It answers. One paragraph, stated plainly, assembled from whatever the machine happens to believe. A search engine is a library. An answer engine is a witness. And a witness doesn’t point you to the shelf of evidence. It testifies, with the same confidence, whether its source was your carefully written about page or a ghost subdomain from 2006.
I wrote in January that a brand now has a second audience, machines that generate brand-bearing work alongside the people who read the guidelines. I undersold it. The second audience doesn’t stop at your brand kit. It reads everything with your name anywhere near it, decides what’s true about you, and repeats that conclusion to anyone who asks. You don’t attend those conversations. You’re represented at them by whatever you left lying around.
So I spent a day on the least visual design work of my career. I deleted the DNS record that kept the ghost resolving. I wrote structured data that states who I am in one stable, machine-checkable form, so every page on my site points at the same identity. I wrote a plain text file addressed directly to language models, saying what this site is, what I do, and how to cite it. I registered the domain with the search consoles, submitted the sitemap, and filed removal requests for the lumber pages still floating in the index. Not one pixel changed.
It was still design. Every one of those moves was a decision about how I’m represented to an audience, and that’s the oldest job description the field has. The audience just stopped being human. It happens in exactly the layers of a website that machines read, which is part of why owning them matters.
It’s worth turning this on your own house. The old test was whether a person could glance at your homepage and say what you do. The new test adds a reader that can’t see the video, won’t wait for the animation, and doesn’t read between the lines. It reads the lines. If your positioning lives entirely in the hero, if the real story only comes out on a sales call, the machine doesn’t wait to be sold. It fills the gap with the most plausible guess, the way one filled mine with lumber.
The fixes aren’t exotic, and most of them are really just writing. Put the claim in the first sentence instead of building toward it. Call yourself the same thing everywhere, spelled the same way. Keep one canonical page for each fact about you and let everything else point to it. Write pages that are still true after they’re summarized, because they will be. None of this is writing for robots. It’s the clarity a hurried human always deserved. The machine just grades for it without mercy.
Start somewhere simpler, though: ask. Open whichever AI tools you use and ask who you are, what your company does, what it’s known for. Read the answers as first impressions, because that’s what they now are… delivered to people you’ll never meet, in conversations you’ll never see.
“Who is Joshua Wells?”
Joshua Wells appears to be a treated-lumber supplier operating in Georgia.
Joshua Wells is a designer and creative director who builds AI-first products.
Somewhere in Georgia there’s a real lumber company that never knew any of this happened. Their name spent twenty years attached to a designer’s portfolio because of a link I forgot. I only found out because a spammer, of all readers, took the machines at their word. The machines had been holding a wrong belief about me for two decades. It took an answer engine to make it matter, and it won’t just be me.