Standing disclosure
How I use AI, and where I don't.
A firm that sells verified facts should say plainly how its own material gets made. Nobody asked me to publish this page. It exists because not publishing it would be the sort of thing I'd criticise a founder for.
The short version
Machines help me write. They don't decide what's true.
I use AI agents to research, draft, and publish. That's not a confession, it's a working method, and given what I advise on it would be strange if I didn't. But the division of labour matters, so here it is in plain terms.
What AI does here
- Searches for what moved in the market each week
- Drafts posts and article sections from my thesis
- Re-checks every published figure against its primary source
- Cuts video into shorter clips
- Publishes the weekly signal board on a schedule
- Reads my own past work back to me so I catch myself repeating a number I never verified
What it never does
- Advise a client, in any form
- Reply to you as me, anywhere, ever
- Decide what I believe or what position I take
- Publish a number I haven't seen
- Speak on anything sensitive, or on any named third party
- Apologise on my behalf
If you're ever in a conversation with Get Reality and you want to know whether you're talking to a person, ask. You'll get a straight answer, and the answer is nearly always that it's me, because the only thing I'm actually selling is that a real person told you the truth.
Synthetic video and voice
If it's a synthetic version of me, it says so.
I may use an AI likeness and a cloned voice for some short videos and narration. When I do, three things are always true.
It's labelled. On the video, in the description, and using the platform's own disclosure setting where one exists. No exceptions, no fine print.
The words are mine. A synthetic presenter reads a script I wrote and approved. It doesn't generate opinions and then attribute them to me.
It's never the sensitive stuff. Client counsel, apologies, commentary on named people or companies, and anything where being wrong costs somebody money - those are me, in person, permanently.
I'll also say the quiet part, since this page is the wrong place to be coy. Almost no law currently requires this of me. New York's synthetic performer statute exempts recognisable real people. California's transparency act binds the tool vendors, not someone using them. The EU's rules reach further, and I intend to meet them, but the honest reason this page exists isn't compliance.
It's that I'd have a hard time selling you an honest read on your company while quietly running a synthetic version of myself.
The rule that matters most
No agent of mine publishes a number it couldn't trace.
Every figure I publish carries its primary source and its date. If a number can't be traced back to the source in one hop, it doesn't run.
When something can't be re-verified, the rule is keep it exactly as it was, with its original date, or remove it. Never re-estimate, never round, never quietly refresh a date I didn't confirm. That applies to the automated weekly update without exception, because the automation publishes without me reading it first.
I learned this the hard way and in public. In April 2026 I published research in which, on re-checking, two out of three figures turned out to be wrong - seven unsupportable, eight misstated, one overstated by roughly 170 times because I'd read token trading volume as protocol revenue. The correction log is public, and the weekly board shows the weeks my thesis gets weaker as prominently as the weeks it gets stronger.
That's the whole reason for the cadence. A thesis nobody re-checks is just an opinion with footnotes.
Found something wrong?
Tell me. That's the service working.
If a figure on this site doesn't survive your own check, I want to know, and I'll correct it in the open with the date attached. No number here is too load-bearing to lose.
This page is reviewed whenever the method changes. Last updated July 31, 2026.