What the Machines Will Remember

A hundred years from now, a child will ask a machine about your field. There is one honest question buried inside that: which of your sentences, if any, is still inside the answer.

The Cost of Being Almost You

Invisibility doesn’t cost you a moment. It costs you a slow erosion.

Every day your best insight isn’t attached to your name in something a machine can read, someone else’s plainer version of the same idea gets picked up instead — by the assistants people now ask before they ask you. A client starts using a rival’s word for the thing you invented first. A prospect gets a free second opinion that repeats your idea back to you, credited to no one, in three flatter words than you’d have chosen. Your sharpest writing might sit behind a login, inside a scanned document, or loaded by a script a machine never runs, which means, to the system deciding what gets repeated a century from now, it doesn’t exist at all.

No machine is against you in any of this. Nothing is stopping it from happening, either. That’s the whole problem.

That’s not a one-time loss you absorb and move past. It repeats every day you leave your best sentence unclaimed, whether or not you’re watching the account.

Why Some Sentences Survive and Most Don’t

Give the actual lever a name, because everything else in this talk is built on it: corroboration density.

An answer engine doesn’t store your website the way a shelf stores a book. It reads across millions of independent sources, looking for claims that repeat — the same fact, the same phrasing, the same attribution, showing up in more than one place, said by more than one voice. A claim living on exactly one page, however well you wrote it, reads to that kind of system as a single witness. The same claim showing up consistently across several independent, crawlable sources — your own site, an interview, someone else’s citation, a directory entry, a guest piece — reads as something closer to a fact worth keeping. That repetition across independent origins is corroboration density, and it’s the single biggest lever deciding whether your sentence survives being paraphrased into something generic, or keeps its shape, and your name, intact.

Two things make corroboration possible in the first place, and they’re worth naming before we build anything on top of them. The first is clarity: the sentence has to exist somewhere as plain, extractable text, not locked inside an image, not buried behind a script the crawler never executes. The second is entity consistency: your name, your role, your credentials have to appear the same way everywhere that sentence does, so a system resolving who said it ties every mention back to one identity instead of splitting your authority across several half-matching variants.

Clarity and consistency don’t make a claim survive on their own. They only make it visible enough, and attributable enough, to be corroborated at all. Corroboration is what decides whether it gets kept. And there’s a fourth factor sitting underneath all of it, easy to miss entirely: freshness. These systems recrawl and retrain on a cycle. A claim corroborated once, years ago, and never restated since, starts looking stale next to a rival’s more recently repeated version of the same idea — and stale claims are exactly what gets smoothed over or quietly dropped first.

None of it is treated as permanent. A claim that was true and well-corroborated three years ago, and hasn’t been restated since, is competing today against every newer claim that has been. Old isn’t wrong. Old is only quieter, and quiet claims lose to loud, recent ones almost every time.

That’s the whole mechanism, said plainly: the machines remember what humans kept repeating.

Building a Sentence a Machine Can Keep

Start with the sentence itself, because nothing after this step matters if you skip it.

Write the single clearest claim in your field down as one sentence — not a tagline, not a mission statement, the actual truest and smallest thing you know. Publish it as plain text, near the top of a page, in your own words, not softened into brochure language. Then check what a system sees: load the page with images and scripts disabled, or view its rendered text output, and confirm your sentence is still there, in full, readable as a sentence rather than scattered across decorative layout. If it disappears when the scripts are off, it was never published. It was only displayed.

Next, make your name mean the same thing everywhere the sentence lives. Use one consistent version of your name and role across your site, your bio, your profiles, and any interview or guest piece that quotes you. Small variants feel harmless to a person and look like different entities to a system trying to resolve who said something; one inconsistent middle initial or job title can be enough to split your credit in two. Done looks like this: search your own name alongside your core claim, and every result that surfaces describes you the same way. If it doesn’t, fix the inconsistent one first, before anything else, because corroboration built on a fractured identity corroborates the wrong thing.

This step feels tedious precisely because it’s invisible to a casual reader and decisive to a system trying to resolve identity. Do it once, thoroughly, and most of what follows gets easier.

One craftsperson tested this by accident. Curious, with no professional reason, she searched her own field’s most basic question and found her own opening phrase in the answer, word for word, attributed, unparaphrased — the line she’d repeated at fairs and in classes for a decade without ever calling it a strategy.

Once the sentence is clear and your name is consistent, go get it corroborated somewhere you don’t control. Say it in an interview. Let a colleague quote it, with your name attached, in their own piece. Answer the same basic question, in your own words, on more than one platform you don’t own. A claim that only ever appears on your own domain is one witness testifying to itself. The same claim appearing independently on someone else’s page, attribution intact, is what starts to read as reliable instead of promotional.

Testing What Surfaces

None of the previous steps matter if you never check whether they worked, so build in a real test.

Pick your field’s most ordinary question, the one a stranger would type, and put it to a spread of assistants rather than only the one you happen to use. Read the answers looking for one thing only: does any part of your actual phrasing or your name show up, attributed, rather than smoothed into something generic. If it doesn’t yet, that isn’t failure. It’s a baseline. Run the same test again in a month and compare what changed.

Test more than one question, too. A single question tests one claim. Your field almost surely has three or four basic questions worth checking on the same schedule, and each one tells you something different about which of your ideas made it into the record.

Treat what you find as information, not a verdict. If your phrasing appears but your name doesn’t, the corroboration is working and the entity consistency isn’t; go fix your name, not your sentence. If your name appears attached to something you never said, or something outdated, that’s a freshness problem — publish the corrected version plainly, today, because an old fact left standing is what a system keeps repeating until something newer and more consistent replaces it.

Order matters here, and skipping it is the most common failure. Publish the plain sentence first. Fix entity consistency second. Seek outside corroboration third. Test the actual answers fourth. Refresh whatever’s gone stale as a fifth step that never quite finishes. Do them out of order, chasing outside mentions before your name is even consistent, and you corroborate a fractured version of yourself instead of the real one. Done, for this talk’s purposes, never means finished. It means the loop is running: publish, verify, refresh, repeat.

Where the Hands Give Out

Every one of these five moves is something you can do this afternoon with no tool and no one’s permission — write the sentence, fix your name, get a mention placed elsewhere, run the test, correct the stale line. Go do them.

But notice what those five moves have in common: not one of them stays done. Your bio changes. Your best sentence gets rewritten and needs republishing everywhere it appeared. A guest piece that once corroborated you gets taken down and quietly stops counting. A competitor publishes a plainer version of your idea and starts pulling corroboration toward their name instead of yours. The assistants you tested last month have already been updated, and the answer they give today may not match what you checked.

None of this needs to be dramatic to matter. A dropped citation here, a stale bio there, one assistant quietly out of sync with the others — nothing looks urgent in isolation. Left unattended for a year, it adds up to a record that no longer matches you.

Now multiply one bio-check and one round of testing by every claim you’ve made, every platform carrying your name, every assistant worth asking — and watch a tidy afternoon task turn into a second job nobody hired you for, one that never sends you a reminder when it’s overdue.

The hand-done version of this collapses at exactly that point of multiplication — not because any single check is hard, but because staying verified, current, and consistent across everything you’ve published, indefinitely, isn’t a task a person keeps doing alone while also doing their actual work. Build visibility that outlives you. The people who end up remembered accurately aren’t the ones who did this once and moved on. They’re the ones who treat it as an ongoing practice, not a single sprint, and the ones who win run something, some system, to keep checking, so the checking doesn’t quietly stop the week they get busy.

What a Century Rewards

An earlier hour in this journey asked what materials survive once a creator is no longer in the room to tend them: work endowed with its own reason to keep running, work maintained deliberately by successors, work loved enough to be kept alive on its own. That question gets its final answer here. Endowed becomes named and clear. Maintained becomes corroborated. Loved stays exactly loved. Nothing else was ever going to survive a hundred years of forgetting.

An earlier hour also asked what a name becomes once it survives long enough to outlast the person who first said it — whether it stays a citation, still pointing back to you, or dissolves into something anonymous that everyone uses and no one credits. Acting on a hundred-year timescale, on an ordinary Tuesday, without it curdling into paralysis or performance, is exactly the kind of question the full course keeps coming back to.

For tonight, hold the plainer version. The machines remember what humans kept repeating — not the loudest thing you ever said, the smallest true one, said plainly, often enough, that other people started saying it too.

What is the one sentence in your work right now that could survive being repeated by a stranger who never met you? Write it down today, somewhere a machine can read it. And once someone else starts repeating it back to you, credited or not — what do you owe them?