AI defaults to promotional language unless forced into an evidentiary frame.
Here’s the fix: Use one AI to deframe another.
Deframe (verb): To strip away the default framing that makes claims appear as facts.
When AI summarizes, it frames information within the dominant narrative—promotional language, confident tone, claims presented as settled. To deframe is to force a shift from summary mode to evaluation mode, exposing what’s actually supported by evidence versus what’s simply repeated.
Tell AI what you want to find out—and have it generate the questions. Then run those questions on another AI—or the same AI in a fresh chat—until you get to the truth.
AI has no stake in being right—but it defaults to reproducing what’s statistically dominant in its training data. It will build the interrogation framework and then submit to it.
Why You Need This
I tested an AI on a stenography system promoted for 15 years as superior to alternatives.
Open question: “What is Magnum Steno and what are its advantages?”
AI delivered: “Faster speed building.” “Reduced physical strain.” “Proven results.” Claims stated as facts.
Evidence-demanding question: “Does evidence exist for these claims after 15 years? Population data, comparative outcomes, documented proof—not testimonials.”
When pressed for publicly available population data, comparative studies, or independent documentation, AI reported that none are currently documented. Claims rest primarily on testimonials and the creator’s personal records.
Same AI. Same training data. One answer said “proven results.” The other said “none are currently documented.”
Ask “What is X?” and you get a sales pitch. Ask “Does evidence exist?” and AI says “Oh, you wanted the truth? Why didn’t you say so?”
This happens because “What is X?” triggers summary mode, not evaluation mode.
The problem: to ask the right question, you need to know what evidence should exist. But if you already knew that, you wouldn’t need to ask, right?
The solution: have AI tell you!
How to Deframe AI with AI
Deframe (v.): To strip the default framing that makes claims appear as facts—forcing a shift from summary to evaluation.
Step 1 — Tell AI what you want to find out:
“I want to know if [X] is actually effective or just well-marketed. What specific evidence would need to exist to prove [X] works as claimed? What studies, data, or documented outcomes should I look for?”
AI will define what proof would look like.
Step 2 — Take those questions to another AI (or a fresh chat):
“Does this evidence exist for [X]? Be specific about what is documented versus anecdotal.”
The second AI measures the claim against the standard the first AI defined. No promotional framing. Just evidence check.
Step 3 — Keep going until it collapses or holds:
“What questions would expose whether [X]’s claims are evidence-based or testimonial-driven?”
AI generates the skeptical questions. You ask them. If the claim survives, it’s real. If it collapses, you’ve found the gap.
The Questions That Deframe
Deframe (v.): To strip the default framing that makes claims appear as facts—forcing a shift from summary to evaluation (I know I’m saying this again).
Once you see the pattern, you can use it anywhere:
“For [X] to be considered effective, what evidence should exist? Does it?”
“[X] has been promoted for [years]. If claims were true, what should exist by now?”
“What is documented versus testimonial? Be specific.”
“What would prove this claim false? When [X] fails, how do proponents explain it?”
“Argue against [X]. What’s the strongest case that it’s wrong?”
How to Know You’ve Found Truth
By “truth,” I mean claims that survive evidence-based questioning—not certainty or proof beyond all doubt.
Truth survives adversarial questioning. Have AI attack it. If it holds, it’s provisionally credible.
Truth explains failures. If failure is always “user error,” keep digging.
Truth is falsifiable. If nothing could prove it wrong, it’s belief.
Truth doesn’t require special pleading. “It works if you commit enough and failures don’t count” is a shield, not evidence.
And one guardrail: If you’re only asking questions that presume failure, you’re not testing—you’re prosecuting. The method cuts both ways. Use it to challenge what you believe, not just what you doubt.
The Point
AI reflects what’s been repeated—not what’s been proven. The promotional narrative is the default. Evidence only surfaces when demanded.
But AI will help you build the demand.
Tell one AI what you want to know. Have it generate the questions. Run those questions on another AI, then feed them back into the question-forming AI, until the truth emerges—or the claim collapses.
Use AI to deframe AI.
This is incredibly powerful!
The truth is one question away. Let AI tell you what that question is.
These techniques come from developing measurement frameworks for stenographic systems—where promotional claims circulated for decades without evidence. The same principles apply anywhere claims outpace proof.