AI StrategistRich Schefren · Strategic Profits

Glossary

Imprint

Your judgment, captured. Not a description of how you decide, written afterwards. The decisions themselves, kept as you make them.

Imprint (noun). A body of rulings belonging to one person. Each entry records a decision that actually happened, the situation that produced it, and the reason behind it. Entries are captured at the moment of the call. Nothing inferred is admitted.

The word does two jobs at once, which is why it survived a long list of alternatives. An imprint is what a form leaves in a surface, so it carries the sense of something transferred by direct contact rather than reproduced from a photograph. It is also the mark of a specific origin, the way a printer's imprint identifies who set the type. Both meanings are load bearing here.

Most collections get better as they grow. This one does not work that way.

Take any other asset a business accumulates. A customer list, a document library, a set of case studies, a codebase. Add a weak item to any of them and you have a slightly weaker collection. The good items still work. The weak one sits there, mildly annoying, doing limited damage.

An Imprint has a different failure shape, and it is the single most important thing to understand about it.

One guessed entry does not make it slightly worse.
It makes the whole set unauditable.

Here is the mechanism. Every entry either came from a decision you made or it came from something inferring what you would probably decide. Both kinds read the same. A good inference sounds exactly like you, because that is what makes it a good inference. It uses your vocabulary, matches your usual pattern, and lands in your voice.

So once inferred material is in the set, you have lost the ability to tell the two apart. Not just for the inferred entries. For all of them. You are now looking at a body of judgment where any given item might be yours or might be a plausible construction, with nothing on the surface to distinguish them.

And the damage compounds, because later inferences get drawn from earlier ones. A guess about your pricing standard informs a guess about your positioning standard, which informs a guess about which client you would turn down. Three layers in, the system is confidently reasoning from a person who was invented one step at a time.

The only recovery from that state is to throw the set out and start again, because there is no way to sort it. Which is why the admission rule is strict at the point of entry rather than corrected later. It is cheaper to keep the set small and clean than to clean a large one, and after a certain size it stops being possible at all.

What actually goes in one

The material is more ordinary than people expect. Judgment does not show up as a philosophy. It shows up as five small events that happen all day.

Each of those makes something visible for a few seconds that is otherwise invisible, including to you. The rejection is the most valuable of the five and the most reliably lost, because rejected work leaves no artifact. It goes in a folder nobody opens, or it never gets made at all.

Notice what is not on the list. Nowhere does anybody sit down to explain their judgment. That is not modesty about the format. It is that the explanation, given in the abstract, would be a different and worse thing: a rationalisation composed after the fact, which is a description of your judgment rather than an instance of it. The work you are already doing is the only reliable source.

The property that makes it worth having

An Imprint cannot be bought, and this is not a marketing line. It is a structural fact with a specific cause.

Everything else in an AI stack has a supplier. The model has a vendor. The tools have vendors. Memory, retrieval, orchestration, evaluation, all of it can be procured on a Thursday afternoon by anyone with a card. Your competitor can assemble an identical stack to yours, and increasingly will, because the components keep getting better and cheaper.

The rulings have no supplier, because they only exist as a byproduct of one person having made those specific calls over years. There is no vendor who has them, no dataset that contains them, and no model that can generate them, because generating them would mean inventing them, which lands you back in the failure described above.

That is an unusual position for a business asset to be in. It is also the only part of the stack where being early actually accumulates into something, rather than being overtaken by the next release.

What an Imprint is not

Not thisBecause
An archiveAn archive holds outputs. An Imprint holds the calls behind them
A chat historyA transcript records what was typed, and most rulings were never typed
A persona or voice modelThose reproduce how you sound. This holds how you decide
A style guideA style guide is self-report, which caps at what you can say about yourself
A memory storeMemory records what happened. This records how you ruled on it
Something you can finishIt accumulates while you work. There is no completed state

Where the idea has been sitting unbuilt

The want is old and well documented. Bill Carr, who ran Amazon's digital media business and reported to Jeff Bezos, described Bezos working for years on how to inject his lens of thinking into meetings he could not attend.

Amazon then built the most thorough set of judgment carriers any company has produced: leadership principles, six-page narratives, tenets, working-backwards documents. Those carried a great deal, and they all shared one ceiling. They could hold only what Bezos managed to put into words, in advance, away from the moment he was actually deciding.

The gap was never the intent or the effort. Nothing existed that could capture a judgment at the moment it was used. That is the part that changed.

Frequently asked

What is an Imprint?

A body of rulings. Each one records a real decision a specific person made, the situation in front of them at the time, and the reason behind the call. It is not a profile, a voice model, or a store of documents. It answers a different question from any of those: not what did this person say, but what would this person decide here.

How is an Imprint different from a prompt library or a set of instructions?

An instruction is something you wrote down because you could say it. A ruling is captured at the moment you made a call, including calls you could not have described in advance. The two overlap at the easy end and diverge completely at the end that matters, which is the part of your judgment that only shows up when a real case is in front of you.

Can I build an Imprint from my existing files and chat history?

You can build the base layer from them, and it is worth doing. What files cannot give you is the ruling. A finished document shows the option that survived. It does not show the three that were killed, or the reason, and no amount of reading it backwards recovers them. That part starts accumulating from the day something begins capturing it, and not before.

Does an Imprint work if someone else fills it in for me?

No, and this is the one property that has no workaround. An assistant can describe how they think you decide. That description is their model of you, which is exactly the guessed material an Imprint exists to keep out. The value comes from the entries being yours, which means the decisions have to be yours.

What happens to an Imprint when I change my mind?

The ruling is superseded rather than deleted, and anything that was resting on it gets re-examined. That operation has a name, belief propagation, and without it a captured judgment slowly becomes an accurate portrait of a person who no longer exists.

The mechanism is public

Imprint is open source, local first, and provenance preserving, which means the admission rule described above is enforced in code rather than promised in copy. The tool ships empty. What fills it can only come from one person's decisions, which is the whole point and also the reason giving the tool away costs nothing.

The category this belongs to is defined at What is Imprinted AI.

Related terms

Where the term comes from

This is one entry in the vocabulary of a longer argument. The full glossary has fifteen terms. The case they belong to runs about 23,000 words, it is free, and there is no email gate on it.

Read The A.I. Business Manifesto

Nothing on this page is for sale. Quote it, argue with it, or pass it on.

Last updated: 28 July 2026