The un-achieving problem
The strangest data in behavior change isn't who fails — it's who succeeds and then reverses. The dieter who loses the ten kilos and regains twelve. The saver who builds the fund and drains it. The founder who finally systemizes, then quietly re-absorbs every process within a year. The goal was met. And then, like a held breath, released.
This is the signature of working at the wrong layer. A goal is a destination; reaching it tells you nothing about whether the traveler changed. If the person who arrives is still, in their own private accounting, "someone with no discipline who got lucky for three months," then the result and the self-image are in contradiction — and one of them has to give. It's almost never the self-image. I watched this so many hundreds of times across a decade that it became the foundation of how I work: change that lasts happens at the identity layer, and everything above it is weather.
Why behavior finds its level
The pull has a name in the research: self-verification. William Swann's work documented something stranger than low self-esteem — people actively, unconsciously arrange their behavior and even their relationships to confirm their existing self-image, including when that image is negative and the confirmation is costly. A coherent self-story, it turns out, is something the psyche prizes above a flattering one.
Now watch what that does to behavior change. Every action inconsistent with your self-image — the run, when you "know" you're not athletic; the calm reply, when you "know" you're hot-tempered — generates a low hum of dissonance: this isn't me. Each instance is easy to override. But the hum is constant, and willpower isn't, and over weeks the system resolves the tension the cheap way: behavior drifts back into agreement with belief. Like water finding its level, behavior finds its identity. Which is why the most leveraged question in all of behavior change isn't "what do I want to do?" It's "who do I believe is doing it?"
You do not rise to your goals. You settle to your self-image — and until the image moves, every achievement is a loan against it.
The vote mechanism
So how does a self-image actually move? Not by affirmation — your brain audits claims against evidence and rejects the unfunded ones instantly. Identity updates the way any belief updates: by accumulating proof it can't dismiss.
This is the vote model, and it's the most practically useful frame I know: every action is a vote for a particular kind of person. Ten minutes of running casts a vote for "runner." Writing three sentences votes for "writer." The skipped session votes too — for the old incumbent. No single vote decides the election; the pattern does, and the pattern is just arithmetic over time.
Three properties of votes worth engineering around:
- Small votes count fully. The identity ledger records that you ran, not your pace. This is why the two-minute version of a habit (see the habit timeline) punches so far above its weight — it's casting daily identity votes at a cost your worst Tuesday can afford.
- Frequency beats magnitude. Thirty small votes outweigh one heroic gesture — elections are won on turnout. The marathon proves less to your self-image than the four-runs-a-week record does, because the record speaks to who you are on ordinary days, which is the only question identity cares about.
- Undeniability is the whole game. The vote must be concrete enough that the old identity can't argue with it. "I tried to eat better" is contestable. "I cooked dinner four nights this week" is a fact. Track the votes somewhere visible — the chain of facts is what the belief eventually surrenders to.
The words that cast votes too
Language turns out to be a voting booth of its own. In consumer-behavior research by Patrick and Hagtvedt, people who refused temptation with "I don't" (skip dessert, miss workouts) were dramatically more persistent than those who said "I can't" — because the two phrases encode different speakers. "I can't" is a rule imposed on you, and rules invite negotiation. "I don't" is a description of who you are, and descriptions don't negotiate. Non-smokers demonstrate the end state: they don't resist cigarettes. There is no struggle to win because there's no smoker left to overrule.
The practical move: pick the identity phrasings and use them in low-stakes public moments. "I don't drink on weeknights." "I'm a morning trainer." Each utterance is a small public vote — and the self-image, which listens carefully to what you say about yourself to others, counts it.
The method, in order
- Name the operator, not the outcome. Translate the goal into a person: not "lose 10kg" but "someone who cooks and moves daily." Not "build a $5M business" but "an owner who builds systems instead of fighting fires." The outcome becomes a side effect of the person.
- Define the smallest daily vote. The version executable on your worst day. Ten minutes, three sentences, one vegetable. You're engineering turnout, not performance.
- Cast and record. Do the thing; mark the fact. A visible chain of undeniable votes is the evidence file your old identity slowly loses the argument against.
- Adopt the phrasing early — one size too big is fine, evidence-bound is mandatory. After a week of votes, "I'm a runner" is a defensible reading of real data, not a costume. Act → evidence → belief. The order is the method.
- When you slip, protect the identity, not the streak. The vote model's kindest feature: one lost vote changes nothing unless you let it trigger a recount. "Runners miss days; I'm a runner who missed a day" keeps the identity intact. "See, I knew I wasn't really..." is the old incumbent contesting the election — name it as that, and cast the next vote.
Stop asking "how do I stay motivated to do this?" and ask "what would the person I'm becoming do right now — and what's the smallest undeniable way to do it?" The first question runs on feelings, which expire. The second runs on evidence, which compounds.
The dark side: identities that hold you back
The same machinery that can carry you runs in reverse just as faithfully. "I'm bad with money." "I'm not a numbers person." "I'm just someone who can't sleep." Each began as a description of a season — often someone else's description, often decades old — and calcified into a self-verifying engine that has been arranging your evidence ever since. The brutal part: you've been casting votes for it daily, mostly by narrating it. (This is self-sabotage's actual engine more often than any fear of success.)
The repair is the same method pointed backwards: catch the identity claim mid-sentence ("I'm just not—"), demand its evidence file, and notice how much of it is old, inherited, or circular. Then start the counter-file: small undeniable votes for the alternative. Identities feel like facts from the inside. They're elections — and elections can be lost by incumbents. I've watched it happen thousands of times, and it is, every single time, the moment the whole change finally holds.
Find out which identity is running the show.
Seven questions, about a minute. See which layer — behavior, system, or identity — your pattern actually lives on.
Take the Free Assessment →Frequently asked questions
What is identity-based change?
Targeting self-image instead of outcomes: "I am a runner" rather than "I want to run a marathon." Behavior that expresses identity is self-maintaining; behavior that contradicts it generates friction and snaps back.
How do I change my identity to change my habits?
By evidence: small, frequent, undeniable votes (ten minutes counts), recorded visibly, plus identity phrasing ("I don't" beats "I can't"). The self-image surrenders to the accumulating record.
Why do I always go back to my old ways?
Self-verification: people unconsciously arrange behavior to confirm their self-image, even at cost. If the result changed but the believer didn't, the believer wins. Relapse is coherence, not weakness — change the believer.
Isn't "fake it till you make it" just self-deception?
The honest version is evidence-bound: act first in small verifiable doses, then claim the identity as a conclusion from real data. Act → evidence → belief — voting, not faking.