A note on process

People ask two questions before they hire me. What do I actually get, and do I write this myself. Here are both answers, plainly.

Yes, I use AI. I would be a strange person to teach AI visibility and pretend otherwise.

Here is the line I hold. The eighteen years in the jewelry industry are mine. The judgment is mine. Every fact, every number and every story is real and checked before it goes out. AI helps me draft inside rules I set, and I am the editor with my name on all of it.

The standards everything has to clear

These are not aspirations. They are gates, and work that fails one does not publish.

  • Every number is verified at its original source. Not at a summary of the source, not at a document that quotes it, and not at something I wrote down earlier. If a figure cannot be traced to where it was first published, it gets cut rather than hedged.
  • Every piece carries something first-hand. A test I ran, a result I can show, a decision made at the bench. If a piece contains nothing that could only have come from me, it has no reason to exist and it does not go out.
  • Nothing is invented. No composite clients, no illustrative numbers, no experience I did not have. If I do not have the real thing, I say so or I leave it out.
  • Client results are evidence, never a template. What worked for one studio is reported with its dates and its caveats. It is never reused as a promise to somebody else.
  • Uncertainty gets stated. A single test is called a single test. A number from one month is dated. Where I do not know, the work says so instead of rounding up to confidence.

Where AI fits, and where it does not

AI drafts. It works inside constraints I set, on material I supply, and everything it produces is edited by me before anyone sees it.

What it does not do is decide. It does not choose what is worth saying, it does not supply the eighteen years, and it does not get to be the reason a claim is in a piece. The research, the tests on my own store, the client work and the judgment about what is true and what is worth your time are mine, and they are the part you are actually paying for.

It is a tool, the way a bench pin is a tool. The hands and the work are still mine.

How you can tell the human is still driving

Anyone can say they review everything. The honest evidence is what gets caught and corrected, so here is what that looks like in practice.

Recently I stopped a piece that had drifted into summarizing someone else's research rather than saying anything of my own, and it went back until it carried my read and my own testing. In the same week I cut a statistic that had been carried forward from an internal document without being checked at its source. It turned out to be real, but it was credited to two studies when it came from one, and I had nearly published that.

Corrections run in both directions here, and they run often. That is the part that does not show up in a finished page, and it is the part that makes the finished page worth reading.

What that looks like from the outside

In August 2026 I ran Google's Gemini over two of my articles using its Non-Commodity Check, an unscripted originality assessment. I did not write the prompt and I did not tell it what to look for. Here is part of what came back about one of them.

Gemini, Non-Commodity Check, August 2026

What makes this page uniquely valuable, and unlearnable from AI Overviews alone, is its transparent behind-the-scenes methodology, including documented rate-limit gaps, temporal variance between July and August test runs ... its candid tone, vulnerability regarding personal visibility failures, and focus on practical insight over keyword stuffing.

The interesting part is what it picked out. Every single thing it named as the reason those pages cannot be replicated is a passage that made the article less impressive when I wrote it.

What it credited

"documented rate-limit gaps"

What it actually was

Reporting a failed engine call as a blank rather than as a zero. A failure and a real absence are different facts.

What it credited

"temporal variance between July and August"

What it actually was

Refusing to present one test run as a trend.

What it credited

"critical breakdowns of vendor bias"

What it actually was

Cutting a statistic I could have used, because every source I could reach for it sold the thing it measured.

What it credited

"disclosing study limitations"

What it actually was

Saying out loud that my best source was sixteen months old.

What it credited

"unaggregated client metrics"

What it actually was

Real client numbers with the caveat attached that the project was a full rebuild, not the one thing I was writing about.

None of that was written to score well. Each one is a place where the honest version was the smaller claim. An evaluator with no idea any of it had been argued over picked exactly those as the reason the work holds up.

That is the whole method in one paragraph. The restraint is not the cost of doing this properly. It is the product.

What this means if you hire me

You get work with my name on it and my accountability behind it. When I show you a result, it happened. When I give you a number, it is dated and I will tell you what it does not prove. When something I did failed, you hear it from me first.

If you want to know more about how any specific piece of your work was produced, ask me. I will tell you.

Andrea LiFounder, Red Pin Geek