Transparency
This page is about how Evergreen Data Harbor runs its own business — not about the workflows we build for clients. If we're going to advocate for careful AI adoption, our own use of it should be legible.
AI slop is real, and it's worth being precise about where it comes from. It isn't that the models are incapable. Output quality tracks input quality — shallow input, shallow output — and slop only ever reaches you when a person steps out of the loop between the two.
Slop is a publishing failure, not a generation failure. Somewhere, someone decided that something was good enough to send without reading it closely. That's a choice, and it's the choice we've organized our work to avoid.
A person writes the brief and edits the result.
The ideas originate with us. AI helps draft, sharpen, and pressure-test — it does not decide what we think or what we're willing to put our name on.
Nothing publishes unread.
Every published word has a human behind it who can defend every line. Where AI materially helped produce something, we say so — the way the byline on our LinkedIn posts already does.
AI never initiates contact with you.
AI may help draft an email or organize notes. But a person decides who to contact, what to say, and when to send it. There is no agent, no automation, and no sequence on the other end. If you hear from Evergreen Data Harbor, Alan sent it.
AI-generated imagery is labeled.
If an image on this site was generated or materially altered by AI, it says so on the image. We hold ourselves to the same standard we'd set for a client publishing a rendering.
No exceptions
We do not read, catalog, or analyze your files with AI.
Not for triage. Not for sorting. Not for “just seeing what's on here.” When that work genuinely needs doing — verifying a restore, confirming we recovered what you actually asked for — a person does it. The work still happens when it's needed. It just doesn't happen with AI. We have turned down work rather than make an exception to this.
One honest exception, because we'd rather explain it than hide it. Recovery work sometimes needs tools whose output includes filenames and file metadata — rebuilding damaged photos, for example, works by referencing the EXIF data of intact ones. When we need help diagnosing a failure like that, those logs may pass through an AI tool. We don't scrub them first, because editing files on a drive we're trying to recover is how data gets destroyed. What never goes through is the content itself: your photos, documents, and messages are not read, uploaded, or analyzed by any AI system.
Our own business records — scheduling, invoicing, service notes, marketing drafts — may pass through AI tools, and that sometimes includes a client name or contact detail. We don't save client information into any AI tool's memory, project storage, or saved context; it stays in our local business files. We make no claims about how the platforms themselves retain data, which is why we keep what reaches them to a minimum.
A mix of AI tools — Anthropic Claude, OpenAI GPT, Google Gemini, Microsoft Copilot, and an assortment of other niche models — all on paid business plans rather than free consumer accounts. We make no claims about how any of these platforms handle data internally. The commitment above is simpler and entirely within our control: your information never reaches them in the first place. When a client engagement involves specific tools, we tell that client which ones, why, and what data they touch, before anything begins.
Last updated 2026-08-21. This page reflects how we work today. If any of it changes, this page changes first, and the date changes with it.
Questions about any of this
If something here is unclear, or you read that line above and aren't sure whether we're the right fit for what you need, that's a good conversation to have. We'd rather over-explain than have you guess.