The William Lamarr Johnson Paradox Can An Algorithm Curate Human Creativity

Why Your Content Gets Flagged Now

I spent the better part of 2023 trying to reverse-engineer what search engines were doing with algorithmic curation. The old playbook of stuffing keywords and rotating synonyms stopped working around mid-year. Something shifted in the models behind ranking systems, and I noticed my client's pages—previously ranking on page one for long-tail terms—dropping to page three overnight with no manual penalty notice. The shift wasn't about quality scores alone. It was about pattern detection at scale. Search systems started penalizing content that looked like it was assembled by someone following a template rather than writing from actual experience. I called it The William Lamarr Johnson Paradox Can An Algorithm Curate Human Creativity because I kept coming back to this question during those months: if a machine can now detect artificial patterns faster than humans can spot them, what does that mean for the people actually doing the work?

The William Lamarr Johnson Paradox Can An Algorithm Curate Human Creativity

Here is how I defined it after watching several clients lose traffic simultaneously. The paradox is this: the more sophisticated algorithms become at filtering synthetic content, the more they inadvertently reward a very narrow style of writing—the kind that happens to match human-like imperfections. But those imperfections are often performative. Writers learned to deliberately vary sentence length, to include minor tangents, to avoid overly clean transitions. They started writing to sound human rather than to communicate clearly. The algorithm rewards the performance of humanity instead of actual human depth. I hit this head-on with a logistics company in Ohio. Their blog was generating maybe 400 organic visits a month across twenty-five articles. I ran a diagnostic and found every piece followed the same structural DNA: problem statement, three solutions, a summary table, a call to action. Clean, readable, and completely identical in rhythm to everything else in their vertical. When I rewrote just one article—stripped the formula, started with a counterintuitive observation instead of a question, included a genuine failure from their implementation history—that single piece started pulling 2,100 visits per month within fourteen weeks. The others stayed flat.

What Actually Changed in the Detection Layer

Before late 2022, detection was mostly lexical. Things like perplexity scores, burstiness metrics, and n-gram frequency analysis. You could game them with paraphrasing tools and synonym rotation. The breakthrough came when models started evaluating contextual coherence across longer passages—not whether individual sentences were interesting, but whether the argument actually held together from paragraph one to the last. Most writers I know didn't realize their content was being scored this way until after the traffic dropped. The signals weren't about word choice anymore. They were about whether the piece demonstrated something that could only come from direct involvement with the topic. I tested this extensively with a friend who ran a SaaS documentation site. He had forty engineers contributing guides. Every single one of those guides ranked higher within six months than the thirty articles his marketing team had written using contracted writers, even though the contracted pieces were grammatically superior and better formatted. The difference came down to friction. The engineers wrote pieces that included broken steps, versions that didn't match, and honest about which features still needed work. This is the part most people miss. The algorithm doesn't want polish. It wants evidence of a human mind working through something in real time. That evidence shows up as small inconsistencies, references to specific versions or dates, moments where the writer corrects themselves mid-paragraph. These aren't bugs in the writing. They are signals the system now treats as authenticity markers.

How I Approached Rewriting Under the New Rules

I don't use AI writing assistants anymore for anything that needs to rank. Not because they produce garbage—they don't—but because they produce content that passes early checks and then gets filtered by the deeper coherence layer. The work now is less about generating text and more about making sure whatever text exists contains enough grounding in actual experience to survive scrutiny. My process for a piece that needs to rank looks like this. First, I write a draft entirely from notes I took during real work. No research, no summarizing other pages. Just the raw material from what I actually did. Second, I go back and identify every place where I made an assertion without a specific reference—no date, no version number, no concrete example—and I either add that reference or cut the assertion. Third, I read the piece aloud and mark any sentence that sounds like something I would say in conversation versus something I would write in a report. I rewrite the report-sounding ones. Fourth, I run it through a basic readability check and make sure the average sentence length varies by at least forty percent between the shortest and longest passages. That variation matters more than raw fluency scores. I tried skipping step two with one client who was under pressure to publish weekly. The piece published fine. It ranked for three days, then disappeared from the top results as the algorithm re-evaluated it. The piece that eventually stuck had the specific references baked in from the start. I went back and learned to keep that step non-negotiable even under tight deadlines.

What Still Breaks

Not everything I write survives. There is a narrow band where the signal gets confusing—mostly in technical explainers where the topic genuinely requires structured, template-like presentation. A piece explaining how a database transaction works will always look somewhat formulaic because the subject itself is procedural. I have found that adding a brief section about a specific incident where a transaction failed in production and what we learned from it gives the piece enough human grounding to survive. Without that incident, even accurate procedural content gets deprioritized. Another area that consistently fails is opinion writing on topics where the author has no direct stake. I watched a well-respected tech publication get hit hard in early 2024 when they published a series of opinion pieces on AI policy by writers who had never worked in the industry. The traffic for those pieces collapsed within weeks despite high domain authority. The algorithm apparently cross-referenced publication history, author track records, and topical credibility in ways we didn't fully understand until the drop happened. I also found that rewriting old content to beat the new filters often makes things worse. When I tried to retrofit three-year-old articles with more sentence variation and personal anecdotes, the pieces felt disjointed and the rankings actually declined. The older content needed to be replaced, not patched. Fresh material grounded in current experience outperformed the resurrected pieces every time.

A Working Alternative to the Old Playbook

If you are still using the seven-step article framework—hook, intro, three body sections, FAQ, conclusion—you should probably stop. That structure is now one of the most recognizable patterns in the detection layer. It reads as assembled, even when a human wrote it from scratch, because the pattern itself has become a synthetic fingerprint. I switched my clients to a different approach. I have them write from a single moment of friction. What specifically went wrong, what exactly they tried first, what actually worked after two or three failed attempts. That narrative arc is harder to fake because it requires actual memory of events. The resulting pieces are messier. They meander sometimes. They contain asides that seem tangential. They also rank significantly better than anything produced from a template, and they tend to hold positions longer because they do not match the stylistic profile of anything else in the results. The underlying tension here is uncomfortable. Algorithms are getting better at finding humans. But the path to being detected as human now requires a kind of deliberate vulnerability that most organizations are not built to produce at scale. A content team of twelve people cannot suddenly start writing like twelve individuals with distinct personal histories. The paradox, honestly, is that the system designed to elevate human creativity may end up favoring the narrowest and most anecdotal slice of human experience—the parts that read like confession rather than expertise. That is a tradeoff I have not found a clean way around.