AI, Honestly · EP008 companion
Every AI response mixes facts, inferences, opinions, and assumptions into a single undifferentiated block — delivered in the same confident tone regardless of what's actually behind it. Seven labels fixes that. Paste it in before you start. Now the AI has to show its work.
Paste into any AI session
Before we begin, follow this framework in every response:
UNDERSTOOD — Restate what you think I'm asking before you answer.
If you got it wrong, I will correct you before you go further.
CERTAIN — Label facts you can verify and stand behind.
UNCERTAIN — Label anything you are inferring, recalling, or not sure about.
OPINION — Label your framing or interpretation. Do not present it as fact.
SOURCE — Name where the information comes from.
Not "studies show" — name the study, the outlet, the document.
ASSUMED — Tell me what you are assuming I meant that I did not say explicitly.
WRONG — If you got something wrong, name it before correcting it.
No silent edits.
Get the full guide + new episode alerts
Enter your email — we'll send the PDF and notify you when new episodes drop. One email per signup, no sequences.
UNDERSTOOD
Proves it heard you before it acts
The most common AI failure: answering a slightly different question than the one you asked, confidently, without flagging the gap. UNDERSTOOD forces the AI to demonstrate comprehension before committing to a direction. If it restated wrong, you correct it before it builds an entire answer on the wrong foundation.
CERTAIN / UNCERTAIN
Separates what it knows from what it's generating
AI delivers everything in the same tone — verified fact and plausible inference arrive identically. These two tags break that. CERTAIN is the AI's strongest claim. UNCERTAIN is the tell — where pattern-matching is doing the work, not sourced knowledge. The most dangerous AI output isn't obvious nonsense — it's confident inference with no flag.
OPINION
Makes the encoded values visible
When an AI frames a situation, weighs competing considerations, or interprets what something means — that's an opinion. It reflects whose training shaped the model, not a neutral truth. OPINION makes that concrete. When you see it, you know you're reading the encoded values of the model, not a sourced fact.
SOURCE
Forces real attribution, not the gesture of it
"Research suggests" and "studies show" are the most common forms of AI confabulation — the cadence of citation without the substance. SOURCE breaks that pattern. If the AI can't name a specific source, it should label the claim UNCERTAIN instead. Named sources you can verify. Vague gestures at evidence you cannot.
ASSUMED
Surfaces the premises before they drive the answer
Every AI response makes assumptions about your context, your intent, your expertise level. ASSUMED makes those visible before they steer the response somewhere you didn't intend. A response built on a wrong assumption wastes your time at best. Misleads you at worst.
WRONG
No silent corrections
If the AI revises an earlier claim without flagging it, you can't track what changed or why. WRONG requires explicit acknowledgment before correction — not a quiet revision, a named admission. This is the hardest behavior for AI systems and the most important for long-term trust. It creates a record. That's how calibration builds.
The values baked into the model during training — the rater judgments, the training data biases, the RLHF dial — those are inside the weights. This framework can't surface them. What it does is make the output layer visible. That's the layer you can act on.
The AI doesn't become honest by using this framework. It becomes legible. You can see where it's certain, where it's guessing, and where it's redirecting. That's what you need to use it well.