Someone spins a wheel rigged to stop on either 10 or 65. They are then asked whether the percentage of African nations in the United Nations is higher or lower than that number, and asked to guess the real figure. The wheel is obviously random. People know it is random. It should have no bearing on a fact they either know or don't. And yet the group that saw 10 guessed around 25 percent. The group that saw 65 guessed around 45 percent. A meaningless number, seen for a few seconds, moved the answer by twenty points.
Where the effect comes from
The experiment is Amos Tversky and Daniel Kahneman's, from their 1974 paper "Judgment under Uncertainty: Heuristics and Biases." Anchoring was one of three heuristics they described, alongside representativeness and availability, and it's the one about numbers specifically. Their explanation was anchor-and-adjust: faced with an unfamiliar estimate, people don't build an answer from scratch. They start from whatever number is already in front of them and adjust away from it. The adjustment is effortful, so people stop as soon as the answer feels plausible, which is usually well before they've adjusted enough. The anchor keeps its grip even after the adjusting is done.
A later account, selective accessibility, adds a second mechanism: once you're given an anchor, you unconsciously search your memory for reasons it could be true, and that lopsided search then biases the "independent" estimate that follows.
The honest caveat
Anchoring holds up better than most findings in this library. It has been replicated across labs, cultures, and stakes, from courtroom sentencing to real-estate appraisal to how much people tip. Where it gets nuanced is size and mechanism, not existence. Recent high-powered replications suggest some early studies overstated how large the effect is, even though the direction is solid. Expertise offers less protection than people assume: real-estate agents shown a rigged listing price still anchor to it, even when they insist their training makes them immune. Incentives (paying people for accuracy) sometimes shrink the effect and sometimes barely touch it. And anchors that are wildly implausible, like an asking price ten times too high, don't get filtered out; they take longer to reject but still pull the final estimate in their direction. The takeaway isn't "expertise or motivation cancels anchoring." It's that anchoring is closer to a structural feature of how estimation works than a fixable error.
What this means for a note
Notes are full of anchors, and most of them get written down before you have the full picture.
The first framing you give an idea in a note tends to survive long after better information shows up. Write "this is a productivity app" in your first line and every later note about the idea gets measured against that frame, even once the real shape of the thing has drifted somewhere else entirely. You're not re-deriving the idea each time you revisit it. You're adjusting from the anchor you set on day one.
The same thing happens with numbers. Jot "maybe three weeks" in a meeting note before anyone has scoped the work, and that number becomes the project's reference point regardless of what the actual estimate later turns out to be. Nobody explicitly agreed to three weeks. It just got there first.
It also happens with yourself. Re-reading old journal entries can anchor your self-assessment to a version of you that no longer exists, good or bad. A rough patch documented in detail six months ago can keep coloring how you read your present, just because it was the first full account of "how I'm doing" that you wrote and reread.
None of this means the first note was wrong. It means it was written before the rest of the evidence existed, and unless you actively revise it, it keeps voting anyway.
Working around your own anchors
A few habits reduce the pull without asking you to distrust every note you write.
- Date-stamp conclusions, not just entries. A line like "as of March, this looked like a 3-week job" marks the estimate as time-bound, so a later note can openly supersede it instead of quietly averaging against it.
- Write your own estimate before you check the reference number. If you're about to look up a comparable price, deadline, or precedent, jot your independent guess first. Checking the number before you guess anchors the guess to it.
- Tag a note's confidence, not just its content. A short "still true?" or "confidence: low" note next to an old conclusion flags it for the next review instead of letting it read as settled fact.
- Reread for disagreement, not confirmation. When you revisit an old note, look specifically for what it got wrong. The selective accessibility mechanism means a casual reread tends to surface reasons the old note was right, not reasons it wasn't.
- Let new notes stand next to old ones instead of overwriting them. A single note edited in place hides how the anchor shifted. A short trail of dated entries on the same topic (closer to a daily notes habit than a single evergreen page) makes the drift visible.
The anchoring effect isn't a flaw in your note-taking. It's a flaw in estimation generally, and a note is just where you can see it happen, one dated line at a time.