How do you separate emotions from data when making a big decision?

A 3-Step Way to Separate Emotion From Data in Decisions

Paul Bloom's Psych explains emotions are not reflexes but cognitive appraisals, so the same racing heart reads as fear or excitement depending on the story you tell yourself, which is why naming a feeling alone changes nothing. My three-step Appraisal Split, run inside a Claude Sonnet 5 project (Anthropic's $20/month Pro plan, 1-million-token context), forces the interpretation and the facts apart before either can contaminate the other.

A founder I coach called me the night before signing a term sheet, certain the investor's aggressive board-seat demand was "just how these things go." It wasn't. When we actually separated what she felt from what was in the document, the feeling was fear of losing the deal, and the data said she had two other term sheets sitting in her inbox. That gap — between the emotion and the actual information — is where most bad decisions live, and there's a real psychological reason it's so easy to miss.

Paul Bloom's Psych lays out the mechanism plainly: emotions aren't simple reflexes triggered directly by events. They're cognitive appraisals — the same physiological arousal, a racing heart, a tight chest, gets labeled as fear, excitement, or anger depending entirely on how you interpret the situation you're in. This is the older two-factor theory of emotion that Bloom folds into the book's broader case that most of what we call "gut feeling" is actually an interpretation dressed up as a fact. The problem for decision-making isn't that emotions are irrational — it's that they arrive already fused to a story, and by the time you notice the feeling, the interpretation has already colored what you think the data says.

So here's the protocol I now run before anything consequential, in a dedicated Claude Sonnet 5 project — Anthropic's current model, released June 30, 2026, available on the $20-a-month Pro plan with a million-token context window, which matters here because you want the whole decision history in one thread, not fragmented across chats. I call it the Appraisal Split:

Name the feeling before the facts. One sentence, no justification: "I feel [X] about this decision." Write it before you open any document related to the decision. This captures the appraisal before it contaminates your reading of the evidence.
List the data with the feeling removed. In the same Claude project, paste every fact you actually have — numbers, dates, quotes, dollar amounts — and explicitly instruct the model not to interpret or recommend, only to list. This is the step people skip, because listing bare facts feels less "productive" than getting an opinion back.
Ask what a different appraisal would see. Prompt it directly: "If someone felt [opposite emotion] about this same data, what would they conclude?" This doesn't tell you which appraisal is correct — it makes the fusion between feeling and fact visible, which is the entire point.

The founder from the term sheet story ran step three and immediately said out loud, unprompted, "if I felt confident instead of scared, I'd counter." That sentence was the whole exercise working — she didn't need me to tell her the data supported a counter-offer. She needed to see that her fear, not the term sheet, had been making the decision for her.

Bloom's broader argument in Psych backs this up with more than one line of research: Paul Ekman's work on universal facial expressions shows the six basic emotions are real and cross-cultural, but the appraisal layer on top of them is what actually drives behavior, and that layer is learned, contextual, and, critically, editable. You can't stop the racing heart. You can absolutely interrogate the story you're telling yourself about why it's racing.

Where this breaks: it's slow, and it requires writing the feeling down honestly, which most people, including me most weeks, resist doing when the feeling is embarrassing — jealousy of a competitor, plain fear of looking foolish in a board meeting. I've also caught the model itself reintroducing appraisal into step two when the prompt isn't explicit enough; Claude is genuinely good at fact extraction, but it defaults to helpful synthesis unless you tell it plainly not to interpret, and a synthesized "fact list" is just an opinion wearing a bullet point. This protocol is also built for decisions with a real deliberation window — a term sheet, a hire, a pricing change. It does nothing for the fifty small calls you make on instinct every day, and it shouldn't; fast intuitive judgment exists precisely because most decisions don't deserve this much overhead.

The uncomfortable finding underneath all of this isn't that your emotions are wrong. Bloom's point, and the reason I keep coming back to Psych over flashier decision-science books, is that the emotion and the interpretation arrive as one bundled package, and the only way to make a clear-eyed decision is to force them apart on paper before you let either one drive.

Sources: Paul Bloom, Psych: The Story of the Human Mind (Ecco, 2023), particularly the chapters on emotion and cognitive appraisal; the original Schachter-Singer two-factor theory of emotion (1962), as discussed in Bloom's synthesis; Anthropic's Claude Sonnet 5 model documentation and pricing, anthropic.com (June-August 2026); Paul Ekman's research on universal facial expressions.