Action Identification Theory: An S-Tier Behavioral Designer’s Guide

There’s one decision every behavior-change designer makes within the first thirty seconds of any product flow, and almost nobody knows they’re making it. It’s not the button color. It’s not the onboarding length. It’s the level of abstraction at which the user identifies what they’re doing.

A user who thinks “I’m tapping a button” behaves one way. A user who thinks “I’m building a portfolio” behaves differently. A user who thinks “I’m becoming the kind of person who invests” behaves a third way entirely. The same finger, the same pixel, the same outcome on the database side, three completely different products in psychological terms.

Vallacher and Wegner figured this out in 1987, published it in Psychological Review, and named it Action Identification Theory. It’s one of those papers that quietly underwrites huge chunks of modern behavioral design and habit science. Most of us cite the descendants (Construal Level Theory, identity-based habits, self-discrepancy work) without ever opening the original.

This guide opens it. We’ll walk through what Vallacher and Wegner got right, where the theory falls apart, what the brain is actually doing when identification shifts, and how to use the dial on purpose inside the Octalysis Framework. By the end you’ll have a way of looking at your own product that will make you wince a little at choices you’ve already shipped.

Speed Run Notes

• Every action can be identified at multiple abstraction levels at once; the level a user holds in mind shapes their behavior more than the action itself does.
• High-level identification (“I’m becoming an investor”) drives persistence and meaning; low-level identification (“I’m tapping deposit”) drives execution and reliability.
• The optimality principle: people settle at the highest identification level they can sustain. Stall the user and identification falls; build their capacity and it rises.
• The Behavior Identification Form (BIF) measures this as a trait, predicting self-control, mood stability, and stickiness on hard tasks better than most personality scales.
• Construal Level Theory absorbed much of AIT’s empirical territory after 2010, but the original is still the cleanest framework for behavioral product design specifically.
• Inside Octalysis, CD1 Epic Meaning is the identity-grade dial, CD2 Development is the action-grade dial, and CD8 Loss is corrosive when it lives at the identity level.

Table of Contents

Author Credibility: Yu-kai Chou

Yu-kai Chou created the Octalysis Framework after studying gamification since 2003 — years before the term entered mainstream vocabulary. As a Human-Systems Architect & Behavioral Designer, his framework has been applied by LEGO, Microsoft, Porsche, Coca-Cola, Salesforce, and MrBeast, impacting over 1.5 Billion Users.

Chou has taught the Octalysis methodology at Harvard, Stanford, Yale, Tesla, Google, BCG, and IDEO.

His work has been cited by Harvard, Stanford, MIT, Forbes, Wall Street Journal, Wired, US Department of Energy, NIST, NSF, NCBI, US Department of Education, ClinicalTrials.gov, and Google Scholar — with 3,700+ more academic publications. Explore his books here.

TEDxLausanne · Gamification to Improve Our World17 min

I’ve spent the last two decades studying why some products make users feel like they’re becoming someone, and why others make them feel like they’re just clicking buttons. Action Identification Theory is the academic frame that explains why that distinction exists at all — and why most behavior-change products plateau at exactly the wrong identification level. The Octalysis design I’ll walk through in this guide is what happens when you treat the identification level itself as the load-bearing variable, not an afterthought.

Most behavioral theories I’ve worked with assume the action is the unit of analysis. AIT inverts that. It says the action is downstream of the identification, and that the identification is the dial designers actually pull. After mapping AIT onto a decade of Octalysis client engagements, I keep finding the same pattern: products that get one part of the eight Core Drives right but ship at the wrong identification level still fail. Products that get the identification level right tend to forgive a lot of other design sins. That asymmetry is what makes this theory worth a deep read.

What Is Action Identification Theory?

Action Identification Theory says that any single action can be mentally represented at multiple levels of abstraction at the same time, and that the level a person holds in conscious awareness governs how they execute, persist at, and feel about that action.

Consider the simple example Vallacher and Wegner used in their original 1987 paper: a person is voting. What are they doing? They’re moving a finger. They’re filling in a circle. They’re casting a ballot. They’re choosing a candidate. They’re influencing the election. They’re exercising democratic rights. They’re shaping the future of the country. Every one of those descriptions is true. Every one of them refers to the same physical event. But they sit at radically different levels of abstraction, and which one is in the person’s head while they’re doing it changes their experience of the act completely.

The theory organizes these descriptions hierarchically. Low-level identifications focus on the mechanics: what the body is doing, what specific procedure is being run, what immediate output is being produced. Mid-level identifications focus on the goal: what local outcome the action is supposed to achieve. High-level identifications focus on meaning: what kind of person performs this action, what larger purpose it serves, what story it belongs to.

Vallacher and Wegner’s central claim is that these levels are not just descriptions for an outside observer. They are real psychological states. When you ask a person what they’re doing and they say “I’m typing,” they will behave differently than when they say “I’m writing a novel,” even if the keystrokes are identical. The level changes what counts as success, what counts as failure, when fatigue kicks in, when meaning kicks in, and how much friction the person can absorb before quitting.

The 1987 paper, published in Psychological Review 94(1), 3-15, was the formal articulation of work the authors had been doing throughout the early 1980s. Their 1984 paper in the Journal of Personality and Social Psychology showed early experimental evidence that identification could be manipulated and that the manipulation had downstream behavioral consequences. The 1985 book A theory of action identification filled in the larger conceptual scaffolding.

What makes the framework powerful for behavioral designers is that it treats identification as a property of the moment, not a property of the action. The same user, doing the same workout, can be at level “lifting weights” on Monday and at level “becoming the kind of person who shows up” on Wednesday. The product mediates that. The copy mediates it. The feedback mediates it. The social context mediates it. Every surface the user touches is voting on what level they hold the action at.

The theory also predicts that the level itself is unstable. It gets pushed around by difficulty, novelty, social cues, and explicit reframing. A user who started a session at the identity level can fall to the procedural level when something stops working. A user who started at the procedural level can be lifted to the identity level by a well-timed narrative beat. Designers who understand this can manage the level deliberately rather than letting it drift.

The version of AIT that has been most useful to me in client work is this compressed reading: every behavioral product makes an implicit promise about what level it expects the user to hold the action at, and that promise is encoded in the interface long before any “value proposition” copy gets written. The interface either invites identity or invites execution. Both are legitimate choices. Choosing without knowing you’re choosing is where most products lose.

The Three Principles That Power the Theory

Vallacher and Wegner organize the theory around three principles. Each one looks simple in isolation. The interaction between them is where the predictive power lives.

Principle 1: Action is maintained with respect to the prepotent identity

The prepotent identity is the one currently in conscious awareness. Whatever level the person is holding in mind right now is the level that governs how they execute the next chunk of the action. This sounds obvious until you realize the implication: the same behavior can be performed competently at one level and incompetently at another, and the difference is purely a function of what description is loaded.

Wegner, Vallacher, Macomber, Wood, and Arps showed this experimentally in 1984. Subjects performing motor tasks at a high-level identification were more variable in their execution and slower to recover from disruption than subjects at a low-level identification, but they showed stronger long-term motivation. The prepotent level is doing real work. It isn’t a label sitting on top of behavior; it is part of the behavior itself, shaping which feedback the person attends to and which they ignore.

Principle 2: When higher-level identifications are available, they take over

The second principle is the upward gravity of meaning. If a higher-level identification is psychologically available, the person tends to drift toward it. Available here means: cued by context, narratively coherent, plausibly attached to identity. People prefer to think they are doing something important rather than something trivial, all else equal.

This is why narrative onboarding works. A new user shown a high-level frame (“you’re starting your investing journey”) will, given the chance, hold the higher frame in mind. The frame is psychologically more attractive than the lower one. Designers who fail to offer a high-level frame are not preventing the upward drift; they are forcing the user to invent one themselves, and most won’t. They will simply hold the lowest available description and behave accordingly.

Principle 3: When action cannot be maintained at the current level, identification shifts lower

The third principle is the safety valve. When execution starts to break down at the prepotent level, identification falls to a more concrete level where the action can still be performed. A musician who was identifying their playing as “interpreting Bach” will, when the piece becomes technically demanding, drop to “playing the notes correctly.” A user who was identifying their session as “building a portfolio” will, when the interface confuses them, drop to “finding the deposit button.”

This downward shift is adaptive in the short run because it lets the person continue acting. But it has long-term consequences. A user who has shifted down rarely shifts back up on their own. The product has to do work to lift them. Most products don’t, which is why so many users abandon at the procedural level even though they originally signed up at the identity level. The shift happened. Nobody noticed. Then the meaning was gone, and so were they.

The Optimality Principle — Why People Settle Where They Settle

The three principles combine into a single emergent property that Vallacher and Wegner call the optimality principle. It is the most useful single claim in the theory for behavioral product design, so it’s worth pulling apart carefully.

The optimality principle states that identification stabilizes at the highest level the person can maintain effectively. Not the highest level they aspire to. Not the highest level the product invited them to. The highest level they can actually sustain given current task demands and current capacity. That stable point is the optimal level for the moment, and it is where identification settles.

This means identification level is not chosen; it is found. Each user, in each session, is running an implicit optimization. They want the meaning of the high level. They need the executability of the low level. The compromise point sits somewhere in between, and it moves around. Add difficulty and the optimal point drops. Add capacity (through training, scaffolding, familiarity, feedback) and the optimal point rises.

The deep insight here is that designers do not get to set identification level by fiat. They can only set the conditions under which the optimization happens. A product that wants users to hold the identity level has to make the identity level executable, which means stripping enough friction out of the procedure that the higher description doesn’t collapse under task load. A product that drops identity-grade copy on top of a procedurally painful interface is doing the worst possible thing: it is pulling the user upward into a level they cannot sustain, then leaving them there to fail.

I see this constantly in client engagements. A team has a beautiful brand message at the identity level. The onboarding is sharp. The first session loads the user with a high frame. Then the third screen is a form with eleven fields, and the user collapses to the procedural level so fast they don’t even notice the drop. The brand message stops doing any work the moment the form loads. The user is no longer “becoming a creator.” They are filling in fields. And once they’re at the field-filling level, the brand message starts to feel like marketing rather than like truth.

The optimality principle predicts this. It also predicts the cure. If you want the user to hold a higher identification, you cannot just paste the higher description on the surface. You have to make the surface support it. That means reducing procedural complexity at every step where you want identity to be load-bearing, and offloading complexity to steps where you have explicitly decided to let identity drop.

The principle also predicts a less obvious phenomenon: when capacity grows, identification climbs back up on its own. A user who has been using your product for six weeks does not need to be reminded of the identity-level frame as often as a new user, because the procedural level has become automatic. The optimization point has moved. This is why mature users of well-designed products often describe their experience in much higher-level terms than new users describe the same product. It isn’t the brand getting better at marketing. It’s the user no longer paying execution tax.

The Behavior Identification Form — Measuring the Dial

If identification level is the dial, the question becomes: can we measure it? Vallacher and Wegner answered that question in 1989 with the Behavior Identification Form, published in the Journal of Personality and Social Psychology 57(4), 660-671.

The BIF is a 25-item instrument. Each item presents a single action and gives the respondent two ways to describe it, one high-level and one low-level. The respondent picks whichever description feels more natural. “Making a list” can be described as “getting organized” (high) or “writing things down” (low). “Locking a door” can be described as “securing the house” (high) or “putting a key in the lock” (low). The total number of high-level choices across the 25 items positions the respondent on a continuum from low-level identifiers to high-level identifiers.

What Vallacher and Wegner found, and what subsequent research has replicated repeatedly, is that BIF scores predict a set of outcomes you would not expect from a 25-item self-report. High-level identifiers persist longer at difficult tasks. They show stronger self-control in delay-of-gratification paradigms. They report more stable mood. They are more resilient under failure. They are better at integrating disparate experiences into a coherent self-narrative.

The instrument is operationalizing what looks like a stable dispositional trait: some people habitually identify their actions at the meaning level, others at the procedure level, and the trait is consequential. But here the theory gets subtle: the trait is not destiny. State manipulations override trait scores in the short run. A low-level identifier put in a context that strongly cues high-level identification will, for the duration of the cue, behave like a high-level identifier. The trait describes the default, not the ceiling.

For behavioral designers, the BIF has two practical uses. The first is research. If you are testing whether your product successfully invites high-level identification, administer a behavior-specific BIF before and after a session. Movement in scores is a clean signal of whether the interface is doing the lifting you hoped it would. The second is segmentation. Users high on the BIF will engage with identity-grade framings; users low on the BIF will bounce off identity-grade framings and engage more reliably with procedural framings. Treating both segments the same is leaving conversion on the table.

One caveat worth flagging: the BIF as originally constructed is somewhat dated in its specific item content (some items reference activities that are less universal than they were in 1989), and the instrument blurs the line between trait and state more than the authors initially acknowledged. Researchers since have proposed shorter versions and domain-specific variants. The conceptual contribution of the BIF, however, that identification level is measurable and predictive, remains intact and remains useful.

What Vallacher and Wegner Got Right

Forty years after the original work, the parts of Action Identification Theory that have held up are the parts that move it out of pure cognitive psychology and into something a designer can use. Three claims in particular have aged well.

The action is the variable, not the goal

Most behavioral theories before AIT treated the action as a constant and the goal as the variable. Goal-setting theory, for instance, asks how different goals change behavior on a fixed action. AIT inverts this. It says the action itself is the variable, because the same physical action is psychologically many different actions depending on how it’s identified. This inversion is what made AIT useful for product design specifically. Designers don’t usually get to pick the user’s goals. They almost always get to pick the framing of the actions the product asks for. AIT made that framing a first-class design surface.

Identification level is a designable surface

The 1987 paper was careful not to make strong design claims, but the implication has been confirmed in dozens of studies since: identification level is responsive to context. Copy, narrative framing, social cues, feedback timing, and reward structure all push the level around. This means a behavioral product is not just delivering an action to the user; it is delivering an identification of the action, and that identification is something the designer chooses (consciously or not). The contribution here is reframing the design space itself. Once you see the identification level as a knob, you cannot unsee it.

The optimality principle predicts collapse before it happens

The optimality principle is the part of the theory that has the most predictive bite. It says identification will fall whenever the user can’t maintain the current level. This gives designers an early-warning signal. If your users are dropping out at a particular step, look at whether identification has collapsed at or just before that step. The collapse usually precedes the dropout by several seconds or several screens. That gap is the window where a well-designed intervention can lift them back up. Most retention work focuses on the moment of dropout. AIT says the real work happens upstream of it, at the moment identification slipped.

Where Action Identification Theory Falls Apart

Honest engagement with the theory means naming what doesn’t work. Three weaknesses matter for behavioral designers.

The trait/state distinction in the BIF is murky

Vallacher and Wegner present BIF scores as measuring a dispositional trait, but the same instrument is sensitive to short-term state manipulations. Administer the BIF after priming subjects with abstract or concrete language, and scores move. Administer it under cognitive load, and scores fall. This is theoretically interesting but methodologically awkward, because it means the instrument is measuring something whose stability is itself a function of context. For designers running BIF-style measurements on users, the interpretation problem is real. A high BIF score after a session might mean the user is a high-level identifier, or it might mean the session successfully lifted them, or both. The instrument can’t fully distinguish.

Construal Level Theory has subsumed much of the empirical territory

Trope and Liberman’s Construal Level Theory, formalized in their 2010 Psychological Review paper, picked up the abstraction-level question and tied it to psychological distance (temporal, spatial, social, hypothetical). CLT has generated a larger research base than AIT in the last fifteen years and has subsumed many of AIT’s empirical claims under a broader framework. The intellectual debt is acknowledged but the field has largely moved on. For designers, this raises a question: why use AIT at all when CLT is more current? The answer, in my view, is that CLT is broader but less specific to action. AIT remains the cleaner tool for the specific design problem of how a user holds the action they are currently performing. CLT is better for thinking about future selves and distant events. The frameworks complement each other.

The theory underspecifies how identification actually shifts

The three principles describe when identification shifts but not how. What is the mechanism? What’s the latency? Are shifts gradient or discrete? The original work is largely silent on these questions, and the empirical literature has not fully filled the gap. This matters for design because the answers determine intervention timing. If shifts are slow and gradient, designers can intervene whenever they notice them. If shifts are fast and discrete, designers need to anticipate them. The truth is probably both, depending on which direction the shift is going, but the theory itself doesn’t say. This is the area where neuroscience has the most to add, which is why the next section matters.

What’s Really Happening Inside the Brain

The original AIT papers predate functional neuroimaging as we know it. Vallacher and Wegner could describe identification levels behaviorally but not anatomically. The neuroscience that has emerged since gives us a much sharper picture of what’s actually happening when identification shifts, and it largely vindicates the original framework.

The Default Mode Network is the place to start. Buckner, Andrews-Hanna, and Schacter’s 2008 review in the Annals of the New York Academy of Sciences consolidated a decade of work showing that the DMN, a set of regions including the medial prefrontal cortex (mPFC), posterior cingulate cortex, and parts of the angular gyrus, activates during self-referential, abstract, and meaning-oriented thought. When a person identifies an action at a high level (“I’m becoming someone who exercises”), DMN activity is elevated. When the same person identifies the same action at a low level (“I’m doing a squat”), DMN activity drops and activity in motor and procedural regions rises.

The mPFC specifically does heavy lifting on self-referential processing. Lieberman’s 2007 review in the Annual Review of Psychology integrated dozens of fMRI studies showing that mPFC activation tracks the personal relevance and identity-loadedness of stimuli. Actions framed in identity terms engage mPFC more strongly than the same actions framed in procedural terms. This gives us a neural correlate of the upward pull described in the second principle: the brain has a region whose activation rises when meaning is available, and that region’s activation is what high-level identification feels like from the inside.

The anterior cingulate cortex (ACC) is the third critical region. ACC has been shown across multiple studies to track discrepancy between current performance and current goal. When the user’s procedural execution falls behind the level required by their current identification, ACC fires, signaling the mismatch. This is the neural footprint of the downward shift: ACC discrepancy detection precedes the conscious recognition that the current level isn’t sustainable, and that recognition is what triggers the drop to a lower identification.

Putting these together gives us a working neural model of action identification. The DMN holds the high-level frame in awareness. The mPFC tags the action as self-relevant under that frame. Motor and procedural regions execute. ACC monitors for discrepancy. When discrepancy crosses threshold, the high-level frame is dropped, DMN activity falls, and the action becomes pure execution. This model isn’t fully validated end-to-end, but each component has independent empirical support, and together they explain things AIT alone couldn’t.

One implication for designers: if you want to keep users at a high identification level, you are essentially trying to keep their DMN engaged. DMN engagement is fragile. It collapses under cognitive load, under high stimulus drive, under sustained attention to fine procedural detail. This is why most apps optimized for short-burst engagement (notifications, streaks, micro-rewards) actually push users toward low-level identification even when their brand messaging is identity-grade. The micro-stimulation regime is incompatible with the neural substrate of high-level identification. You can have one or the other. You cannot have both.

This is also why mindful or contemplative product moments, a deliberate pause, a slow animation, a long-form reflection prompt, have outsized effects on identity-grade engagement. They give the DMN room to do its work. Most products are too anxious to allow that room. They fill every empty second with stimulus, and in doing so they collapse the identification level they are trying to maintain.

There’s a second neural layer worth naming: the role of dopaminergic prediction-error signaling in habit formation and the way it interacts with identification level. When the user’s prediction-error signal is consistently rewarded at the procedural level (the deposit button worked, the task completed, the streak ticked up), the dopamine system reinforces procedural identification. The same physical action, rewarded at the identity level instead (a moment of recognition that the user has become someone), reinforces identity identification. Most products reward at the procedural level by default because procedural rewards are easier to time and measure. The accidental consequence is that the product trains the user to identify at the lower level, even when the brand is doing identity-grade work elsewhere.

Action Identification Theory vs Other Theories

AIT lives in a crowded neighborhood. To use it well, it helps to know where it overlaps with and diverges from its closest cousins.

AIT vs Construal Level Theory (Trope & Liberman)

Construal Level Theory shares AIT’s core insight that the same content can be represented at different levels of abstraction, and that the level shapes downstream cognition and behavior. The key difference is the independent variable. CLT ties abstraction level to psychological distance: distant events are construed abstractly, near events concretely. AIT ties identification level to action maintenance and the optimality principle. The two frameworks generate similar predictions in many cases (both predict that a distant goal will be held at a higher level than an immediate task), but they diverge when distance and action demands point in different directions. For behavioral product design specifically, AIT is usually the more useful frame because the unit of analysis is an action the user is actually performing right now. CLT is more useful for thinking about how users represent their future selves or how they evaluate hypothetical scenarios.

AIT vs Self-Determination Theory

Deci and Ryan’s Self-Determination Theory focuses on the three basic psychological needs (autonomy, competence, relatedness) and how their satisfaction or frustration drives intrinsic versus extrinsic motivation. AIT and SDT are doing different things at different levels. SDT explains why an action is motivating; AIT explains how the same motivating action can be psychologically held at different levels. The two frameworks compose cleanly. An identity-grade identification of an action is more likely to satisfy autonomy and competence needs than a procedure-grade identification of the same action, because identity-grade engagement makes the action feel chosen and meaningful. SDT tells you the soil quality; AIT tells you how deep the roots go.

AIT vs Goal-Setting Theory (Locke & Latham)

Goal-Setting Theory says specific, difficult goals produce better performance than vague or easy goals. AIT complicates this. Specificity in Locke and Latham’s sense usually pushes identification toward the procedural level, which improves short-term execution but undermines long-term persistence and identity formation. The classic SMART-goal framing is a low-to-mid-level identification regime. It’s optimized for measurable output, not for meaning. AIT predicts (and the data largely confirms) that highly specific goals work well for short-horizon, well-bounded tasks but underperform for long-horizon identity-formation tasks. For habit work, vague identity goals (“I’m becoming a writer”) often outperform specific behavioral goals (“write 500 words a day”) in the long run because they protect against the brittleness that comes from procedural focus.

AIT vs Identity-Based Habits (James Clear via William James)

James Clear’s identity-based habits framework, popularized in Atomic Habits, draws an explicit line back through William James’s 1890 Principles of Psychology. The argument: change happens when you stop trying to do the action and start being the kind of person who does the action. This is the high-level identification principle of AIT restated for a popular audience. Clear’s contribution is the practical operationalization (cast a vote for the identity with each small action) rather than the underlying theoretical claim. AIT predates this framing by decades and gives it the rigorous experimental backing that Atomic Habits mostly leaves implicit. If you’ve read Clear, you’ve encountered AIT without the academic apparatus.

Action Identification Theory in the Real World

Theory is only useful if it survives contact with reality. The four domains below are where I’ve seen AIT pay off most consistently in client work and where the published literature has the strongest applied evidence.

Workplace and leadership — the mission-vs-task lever

Most leadership thinking distinguishes between transactional management (assign tasks, monitor execution) and transformational leadership (inspire identity, connect to purpose). AIT gives this distinction a precise mechanism. Transactional management operates at the low-to-mid identification level; transformational leadership operates at the high level. Both are necessary. Pure transactional management leaves employees procedurally proficient but identity-flat, with the long-term consequence that they bounce when better procedural conditions appear elsewhere. Pure transformational leadership lifts identification to the identity level without giving employees the procedural scaffolding to sustain it, which leads to the collapse pattern: people who believe in the mission but can’t execute on it eventually feel like frauds and disengage.

The leaders I’ve seen do this well move people up and down the identification ladder deliberately. Strategy review meetings load the high level. Sprint planning loads the mid level. Daily standups load the low level. The cadence is the design. When the cadence breaks (only standups, no strategy review for six months), identification calcifies at the low level and the team loses its sense of why. When the cadence inverts (constant strategy talk, no concrete execution rhythm), people feel inspired but ineffective. The best leaders manage the rhythm of identification levels, not just the content at any single level. This rhythm is what most management consultants miss: they audit the content of each meeting, not the identification level it produces, and so their recommendations leave the underlying problem intact.

Education — why over-scaffolding produces compliance instead of mastery

Modern education is heavily scaffolded. Every learning activity has rubrics, success criteria, step-by-step guides, and explicit feedback loops. This is procedurally excellent and identification-poor. Students under heavy scaffolding identify their learning at the procedural level (“I’m filling in this worksheet”) rather than at the identity level (“I’m becoming someone who understands this”). The procedural identification is executable on autopilot, which produces the metrics teachers want, but it never forms the identity that produces independent mastery.

The educators who get exceptional results tend to scaffold less than seems prudent. They throw students into problems above their procedural ceiling and let the struggle force a higher identification. This looks inefficient in the short run (more confusion, more failure, slower visible progress) but produces durable mastery because the student has been identifying themselves as a problem-solver, not as a worksheet-completer. AIT predicts this and the educational research literature is loaded with corroborating evidence. The implication for digital education products is uncomfortable: most are too scaffolded to produce mastery, and the metrics they optimize against actively select for the wrong identification level. The companies that figure out how to maintain identity-grade learning at scale will own the next decade of edtech, while the ones still chasing completion rates will produce a generation of credentialed but unconfident learners.

Marketing and UX — why “feature lists” sell software but never sell identity

Most software marketing pages live at the mid identification level. They list features, describe outcomes, and invite the user to evaluate whether the procedural promises match their procedural needs. This sells software. It does not produce loyalty, advocacy, or identity-grade engagement. The brands that achieve the latter speak at the high identification level. They tell the user who they are becoming by using the product. Apple, Patagonia, Notion at its best, certain fitness products — all of them refuse to sell features and insist on selling identity. The data backs this up. Lifetime value is markedly higher for users acquired through identity-grade messaging than for users acquired through feature-grade messaging, even when conversion rates at the acquisition step look similar.

The deeper UX implication is that the identification level invited by the marketing has to be sustained by the interface. A landing page that promises identity transformation followed by an onboarding that drops the user immediately into feature explanation produces a worse cohort than a fully feature-grade funnel would have produced. The mismatch itself is the damage. The user noticed they were lifted, then noticed they were dropped, and the experience of being dropped is what gets remembered. Consistency across the identification level of every surface is the single biggest source of compounding gains in product UX over time. I’ve audited dozens of high-spend marketing funnels where the gap between landing-page identification level and onboarding identification level was the largest single factor explaining poor cohort retention, and almost nobody on those teams had a vocabulary for what was happening.

Healthcare and behavior change — identity framing for medication adherence and recovery

The healthcare literature on behavior change has been one of the most productive applied territories for AIT. Medication adherence, for instance, is notoriously poor when patients identify their behavior at the procedural level (“I’m taking a pill”). Adherence improves substantially when the same behavior is identified at the identity level (“I’m someone who manages my condition”). Several randomized trials in chronic disease management have shown 15-30% improvements in adherence from interventions that explicitly target identification level rather than procedural reminders. The pill itself doesn’t change. The identification of taking the pill does, and the procedural adherence follows the identification.

Addiction recovery is the most striking case. The shift from “I’m not drinking today” to “I’m someone in recovery” is exactly an upward identification shift, and the recovery literature has known for decades that this shift correlates with sustained sobriety far better than procedural willpower does. Twelve-step programs are doing AIT without naming it. The identity-grade language is the active ingredient, not the meeting attendance per se. Digital recovery products that miss this and focus on procedural tracking (days sober, check-ins) tend to underperform older models that load identity first. The data is clear and yet the design pattern persists, because procedural tracking is easier to ship and easier to instrument. The harder, slower work of identity formation is where the real outcomes live.

The Elephant in the Room — The Identification Tax Nobody Talks About

Here’s what most discussions of identification level skip: high-level identification is not free. Every time you lift a user into the identity grade, you’re installing a tax that will be paid the next time the behavior fails. And behavior will fail. People miss workouts, blow budgets, skip meditations, eat the cake. When the behavior fails at the procedural level, only the procedure failed. When the behavior fails at the identity level, the identity failed with it.

This is the asymmetric cost of high identification, and it’s the reason designers should be more thoughtful about when to install it. A user who identifies “missed today’s run” at the procedural level shrugs and runs tomorrow. A user who identifies the same miss at the identity level concludes that they are not, in fact, a runner. The high identification amplifies both the upside of success and the downside of failure, and the downside is heavier than the upside.

The data on streak-based products bears this out. Streaks lift identification to a fragile high level: “I’m someone who hasn’t missed a day in 47 days.” When the streak breaks, the identification collapses catastrophically. A meaningful proportion of users who break a long streak don’t come back. The product accidentally installed an identity, then accidentally destroyed it, and the user mourns the destroyed identity by leaving the product that hosted it. Most streak-based products lose more lifetime value to streak breaks than they gain from streak motivation, but the analytics rarely surface this because the loss is invisible in the moment. It shows up as a cohort decay curve that nobody can fully explain, and the team blames product-market fit instead of the identification design.

The reverse problem is just as real. Products that keep users permanently at the procedural level avoid the failure-amplification problem but never produce identity formation. Users stay engaged longer (no catastrophic identity collapse) but never become advocates, never deepen their use, never integrate the product into their self-concept. They use it; they don’t become anything because of it. These products have lower churn and lower lifetime value, and almost no defensibility against a competitor who figures out how to do identity work safely.

The design implication is that identification level is not a knob you should turn all the way up. It is a knob with an optimum, and the optimum is below the maximum. The skilled designer installs just enough identification to produce identity formation while preserving the option to recover from failure without identity collapse. This is harder than it sounds. It requires building in narratives of fallibility (“the kind of person who runs sometimes misses runs and gets back to it”), normalizing recovery, and reframing failures as identity-consistent rather than identity-disconfirming. Most products do none of this. They install identity at acquisition and let it shatter at the first failure, and they wonder why retention is so volatile.

The deeper move, which almost no product I’ve audited has implemented well, is to load identification at two levels simultaneously and rely on whichever one the user can sustain in the moment. Identity is available when the user has capacity for it. Procedure is available when they don’t. Both are legitimate, both are honored, neither is contingent on the other. Users move between the levels as their week, their energy, their circumstances allow. The product is robust to identification collapse because the lower level catches them, and it’s also capable of identity formation because the higher level remains accessible when capacity is high. This two-layer design is rare in practice and is the thing I push hardest for in advanced Octalysis engagements. Once you see a product that does this well, every other product looks brittle by comparison.

How to Apply Action Identification Theory with the Octalysis Framework

The Octalysis Framework organizes behavioral design around eight Core Drives (shorthand: CD1 through CD8). Each Core Drive operates most naturally at a specific identification level, and mapping each Core Drive to its native AIT level is one of the cleanest ways to design with the identification dial on purpose. The acronyms below refer to the same eight Core Drives throughout — Core Drive 1 (CD1) Epic Meaning, Core Drive 2 (CD2) Development, and so on through Core Drive 8 (CD8) Loss.

Core Drive 1 (CD1): Epic Meaning & Calling sits at the highest identification level the framework supports. CD1 is the identity-grade Core Drive. When you engage CD1, you are inviting the user to identify their action at the level of meaning, purpose, and self-narrative. The classic Game Techniques here, Calling #67 (the user is the chosen one, the appointed steward, the only person who can do this), Narrative #10 (the action is part of a larger story the user is living), and Heroes #82 (the user is positioned as a hero archetype within their own life), are precisely the surfaces that lift identification to the identity level. A product that lacks CD1 entirely will struggle to install identity-grade identification at all, because the Core Drive that does that work isn’t engaged. The brand might gesture at meaning, but without the structural support of CD1 game techniques, the gesture doesn’t take.

Core Drive 2 (CD2): Development & Accomplishment operates at the mid identification level. CD2 is goal-grade and action-grade. When users engage CD2, they identify their actions as steps in a progression toward a measurable outcome. Step-by-Step Tutorial #13, Achievement Symbols #3, Progress Bar #4, and Boss Fights #19 all work at this level. They make the action feel like an achievement rather than like mere execution, but they don’t reach the identity layer. Core Drive 2 is the workhorse of most behavioral products and the source of most short-term engagement. A product that runs only on CD2 produces accomplishment without becoming, which is why so many CD2-heavy products feel hollow over time. The user finishes the tutorial, fills the progress bar, beats the boss, and still doesn’t feel like they have become anyone. The Core Drive is doing its job. The job just doesn’t include identity formation.

Core Drive 3 (CD3): Empowerment of Creativity & Feedback operates at the low-level execution layer by default, but feedback specifically has the power to lift identification. Real-Time Feedback #57 keeps the user grounded in the moment of action; Plant Picker #44 (and its cousins, the open-ended creative tools) lets the user experience agency in execution. The interesting move with CD3 is that high-quality feedback can elevate the user’s identification without forcing the issue. A creative tool that responds gracefully to the user’s choices subtly invites them to identify as a creator, which is an identity-grade frame, even though the immediate action is procedural. Core Drive 3 is the Core Drive that bridges levels most naturally, and it’s the one most underused for identification lifting because designers tend to think of feedback as informational rather than identity-shaping.

Core Drive 4 (CD4): Ownership & Possession requires high-level identification to function fully. You don’t fully own what you can’t conceptualize as yours, and conceptualizing something as yours is an identity-grade cognitive operation. Inside Core Drive 4, Avatars #1 work at the identity level (the avatar is a representation of self), Status Points #2 work at the mid level (the points represent accumulated progress), Choice Perception #56 works across levels (the perception of choice is itself an identification-lifting move). A product that wants to engage CD4 fully has to invite identity-grade identification, because ownership at the procedural level is shallow and doesn’t drive long-term retention. The user might collect items, but they won’t feel like the items are part of who they are, which is the actual psychological mechanism CD4 is trying to engage.

Core Drive 5 (CD5): Social Influence & Relatedness is the social anchor for identification level. People tend to identify their actions at the level their peers identify them. If the people around you call what you’re doing “investing,” you call it investing. If they call it “messing around with stocks,” you call it that. Mentor #41 elevates identification because the mentor frames the action at the level the mentee aspires to. Group Quests #22 lock in identification at the group’s level. A skilled product designer uses social context to lift identification level, especially for users whose individual default is lower than the product wants. This is why Core Drive 5 community is so often the missing ingredient in behavioral products: without peers identifying the action at the identity level, individual users can’t sustain it on their own.

Core Drive 8 (CD8): Loss & Avoidance deserves the most careful handling in identification-level terms. CD8 used at the boundary, at the low level, is fine. Countdown Timer #65 creates execution-grade urgency. Status Quo Sloth #87 amplifies the cost of changing what you’re already doing. These work at the procedural level and don’t threaten identity. But CD8 at the identity level is corrosive. A streak that frames itself as “you’re not the kind of person who breaks streaks” — the Duolingo green-owl pattern, the Snapchat fire emoji, the Apple Fitness ring closure — installs identity-grade loss aversion, and when the streak breaks, the user loses not just the streak but the identity. The product accidentally designed the user’s self-disconfirmation. Identity-grade Core Drive 8 should be used sparingly, with explicit recovery paths, and almost never as the primary engagement driver. Most products that lean hard on CD8 are mortgaging long-term identity formation for short-term retention metrics.

Two Core Drives I haven’t mapped here, CD6 Scarcity & Impatience and CD7 Unpredictability & Curiosity, can operate at any identification level depending on how they’re framed. Scarcity that says “only 24 hours left” is procedural. Scarcity that says “this is the moment that defines what kind of person you become” is identity-grade. Unpredictability that delivers a random reward is procedural. Unpredictability that delivers a moment of insight or transformation is identity-grade. The choice of level for these two Core Drives is one of the most consequential design decisions a behavioral product makes, and it should be made deliberately rather than left to whichever framing happened to come up in the copywriting meeting.

The applied principle: audit your product Core Drive by Core Drive, and ask what identification level each Core Drive is currently operating at. Identify the mismatches. A CD1 message followed by a CD2 funnel followed by a CD4 onboarding that asks for ownership before the user has any identity context is a chain of mismatches that will produce exactly the collapse pattern AIT predicts. Align the levels across the user journey. Let the levels rise and fall in a deliberate rhythm. That alignment is where Octalysis and AIT compound into something more useful than either framework alone, and it’s the kind of work that separates products that produce identity from products that just produce engagement metrics.

Practical Steps: Designing the Identification Level on Purpose

Here are seven concrete moves you can make on Monday morning that will materially change the identification level your product invites.

1. Audit the first thirty seconds of your user flow for level mismatch. Open your product fresh and write down, for each screen in the first thirty seconds, what level of identification the user is being invited to hold. Identity, goal, action, or pure procedure. If the level moves around without a deliberate rhythm, you’ve found your first fix. Bring the levels into a coherent arc, usually high to low or high to mid, never mid to high to mid. The arc itself is the thing you’re designing, and most teams have never explicitly designed it because nobody on the team had the vocabulary to describe what they were looking at.
2. Rewrite three pieces of in-product copy at the identity level and run a structured comparison. Pick three high-value moments (onboarding completion, first habit logged, milestone hit) and write the copy at the identity level instead of the procedural level. “You logged a workout” becomes “You showed up like someone who runs.” Measure not just immediate engagement but retention at 14 and 30 days. The lift, if it appears, will be larger at 30 days than at 14, because identification-level work compounds. Procedural copy delivers a one-time hit; identity copy installs a frame that keeps paying off.
3. Install a two-layer narrative for failure recovery. Every failure surface in your product should have copy that honors the identity level and the procedural level simultaneously. “Missing a day doesn’t change who you are; here’s how to get back tomorrow” is doing both jobs. The identity sentence prevents identity collapse. The procedural sentence makes recovery executable. Most products write only the procedural version and accidentally let the identity shatter. Two sentences instead of one is the cheapest high-impact change you can ship this week.
4. Add a deliberate pause to your highest-value moment. If you have an onboarding completion screen, a major milestone, or a celebration moment, add an explicit slow beat: a long animation, a full-screen quiet message, a reflection prompt. This is the DMN room I described in the neuroscience section. Without it, the moment passes too quickly for identity-grade processing to happen. With it, the moment loads. Slow design at the right moments is identity-grade design, and most products are too anxious about engagement metrics to slow down where slowing down is exactly what creates the lasting memory.
5. Map every Core Drive in your product to its current identification level and find the mismatch. Use the Octalysis-AIT mapping from the previous section. Walk through each Core Drive your product engages and ask which level it’s operating at. Then ask whether that level is the right level for that Core Drive given the user’s stage. The mismatches you find are usually easier to fix than they look. A small copy change or feedback timing change can move a Core Drive’s effective identification level by a full step. The exercise itself is worth more than any single fix, because it builds the habit of looking at design choices through the identification lens.
6. Build a downshift-resistance mechanic into your high-effort moments. When the user is about to do something hard (a long lesson, a deep practice, a difficult decision), pre-load the high identification level explicitly. A single sentence reminding the user who they are becoming by doing this thing makes the downshift less likely under task load. The optimality principle says the level will fall when execution gets hard. The countermove is to make the high level more sticky right before the difficulty arrives, not after. Pre-load, then execute, rather than executing and trying to recover the level you already lost.
7. Measure identification level changes over a user’s lifecycle and use them as a leading indicator. Administer a short behavior-specific BIF (5 items, not 25) at sign-up, at 14 days, and at 30 days. Track movement. Users whose identification level rises between day 1 and day 30 retain at materially higher rates than users whose level falls. This metric will predict churn weeks before behavioral signals do, because identification collapse precedes dropout. Use it as an early warning system, and use it to evaluate whether product changes are moving the identification dial in the direction you intended.

Action Identification Theory Was the Beginning, Not the End

Vallacher and Wegner’s 1987 paper does something most behavioral theory papers don’t do: it survives. Forty years on, the central claim is intact, the predictions still hold, and the framework still gives designers a useful map. That’s rare. Most behavioral theories of comparable vintage have been folded into larger frameworks, demoted to historical interest, or quietly contradicted by better data.

What’s happened to AIT is more interesting than any of those outcomes. The original framework pointed at something that newer theories have built on. Construal Level Theory took the abstraction-level insight and tied it to psychological distance, generating a much broader research program. Identity-based habits work (James Clear’s popular framing, but rooted in William James) took the identity-level identification piece and built a habit-formation practice around it. Implementation intentions (Gollwitzer’s work, refined by Sheeran) took the procedural-level identification piece and showed how specifying the low level reliably increases action completion. Each of these is a child of AIT in some meaningful sense.

The reason to still read the original is that the children are each specialized. CLT is better at distance, identity-based habits is better at long-horizon self-change, implementation intentions is better at single-action completion. AIT is the framework that holds the whole picture: the relationship between levels, the optimality dynamics, the upward and downward shifts. If you want the full theory of how a user holds an action they are currently performing, AIT is still the cleanest source.

For behavioral designers specifically, this matters because product design is almost never about a single level. It’s about managing the relationship between levels over time. A great product moves users up and down the identification ladder in a deliberate rhythm. Onboarding lifts them. The first action grounds them. The first failure protects them. The first comeback re-lifts them. The fifth week elevates them again as capacity grows. The product is conducting an identification-level symphony, and AIT is the score that lets you read it. The children of AIT each tell you a single instrument’s part beautifully. The original is the only one that shows you the whole composition. That’s why it’s still required reading, and why I expect it will still be required reading in another forty years.

Frequently Asked Questions

What is Action Identification Theory?

Action Identification Theory, proposed by Vallacher and Wegner in 1987, says that any action can be mentally represented at multiple levels of abstraction simultaneously. The level a person holds in conscious awareness shapes how they execute, persist at, and feel about the behavior. Low-level identifications focus on mechanics; high-level identifications focus on meaning. The same physical action becomes psychologically different actions depending on the level held in mind.

Who created Action Identification Theory?

Robin Vallacher and Daniel Wegner developed the theory across the early 1980s and formalized it in their 1987 Psychological Review paper. Their 1985 book A theory of action identification provided the broader conceptual scaffolding. Wegner is also widely known for his work on thought suppression (the “white bear” effect) and the illusion of conscious will. Vallacher continued developing the theory through subsequent collaborations into the 2010s.

What is the optimality principle in Action Identification Theory?

The optimality principle states that identification stabilizes at the highest level the person can maintain effectively given current task demands and capacity. When execution becomes difficult, identification falls to a more concrete level where action can still be performed. When capacity grows, identification climbs back up. This emergent property is what makes the theory predictive: designers can anticipate where the identification level will settle by examining the procedural complexity of the action.

What is the Behavior Identification Form (BIF)?

The BIF is a 25-item self-report instrument developed by Vallacher and Wegner in 1989 and published in the Journal of Personality and Social Psychology. Each item describes an action and asks which of two descriptions, one high-level and one low-level, the respondent prefers. The total score positions someone on a continuum from low-level identifiers to high-level identifiers. BIF scores predict persistence, self-control, mood stability, and resilience under failure.

How does Action Identification Theory differ from Construal Level Theory?

Both theories deal with levels of mental representation. The key difference is the independent variable. Construal Level Theory ties construal to psychological distance: temporal, spatial, social, hypothetical. AIT ties identification level to action maintenance and the optimality principle. CLT is broader and more current; AIT is more specific to the action being performed in the moment. The two frameworks are complementary and often generate similar predictions, but they diverge when distance and action demands point in different directions.

Can you change your action identification level?

Yes. Identification level is not fixed. Difficulty, novelty, social cues, and explicit reframing all push the level around. Trait BIF scores describe a default, not a ceiling. State manipulations override trait scores in the short run. For designers, this is the central practical claim of the theory: the identification level your interface invites is itself a design surface, responsive to copy, feedback timing, narrative framing, and the social and procedural structure of the experience.

Why do high-level identifiers persist longer at difficult tasks?

High-level identifiers attach the action to identity and meaning, so each instance of the behavior accrues value beyond the immediate outcome. A bad run still counts as something because it was performed by someone who runs. Low-level identifiers, by contrast, evaluate each action purely on procedural success. When procedure fails, the action feels like a complete loss. The asymmetry compounds over time, which is why BIF scores predict long-horizon persistence better than most personality measures.

How does Action Identification Theory apply to habit formation?

Habits form most reliably when the action is identified at a low enough level to be executable on autopilot, but the identity layer remains accessible for moments when motivation dips. This is the two-layer structure I recommend in advanced Octalysis engagements. The procedural layer keeps the behavior reliable. The identity layer keeps the behavior meaningful. James Clear’s identity-based habits framework is the high-level identification principle of AIT restated for habit design specifically.

What does Action Identification Theory mean for behavioral design?

It means the identification level your interface invites is the highest-leverage design decision most designers make without realizing it. Copy, feedback, narrative, and rewards each anchor the user at some level. Choosing that anchor deliberately, and managing the rhythm of level changes across the user journey, produces materially better retention, engagement, and identity formation than any single-level approach. Octalysis Core Drives map onto identification levels and provide the operational vocabulary for this design work.

Is Action Identification Theory still relevant today?

Yes, and arguably more relevant now than when it was published. Newer theories like Construal Level Theory, identity-based habits, and implementation intentions build on AIT’s core insight that the same action can be represented at different levels. The original framework remains the cleanest map for product designers thinking about identity and execution together. Forty years on, the central claim is intact, the predictions still hold, and the framework still gives designers a more useful map than any of its descendants alone.

References

1. Vallacher, R. R., & Wegner, D. M. (1987). What do people think they’re doing? Action identification and human behavior. Psychological Review, 94(1), 3-15.
2. Vallacher, R. R., & Wegner, D. M. (1985). A theory of action identification. Erlbaum.
3. Vallacher, R. R., & Wegner, D. M. (1989). Levels of personal agency: Individual variation in action identification. Journal of Personality and Social Psychology, 57(4), 660-671.
4. Trope, Y., & Liberman, N. (2010). Construal-level theory of psychological distance. Psychological Review, 117(2), 440-463.
5. Wegner, D. M., Vallacher, R. R., Macomber, G., Wood, R., & Arps, K. (1984). The emergence of action. Journal of Personality and Social Psychology, 46(2), 269-279.
6. Buckner, R. L., Andrews-Hanna, J. R., & Schacter, D. L. (2008). The brain’s default network. Annals of the New York Academy of Sciences, 1124, 1-38.
7. Lieberman, M. D. (2007). Social cognitive neuroscience: A review of core processes. Annual Review of Psychology, 58, 259-289.
8. Locke, E. A., & Latham, G. P. (1990). A theory of goal setting and task performance. Prentice-Hall.
9. Deci, E. L., & Ryan, R. M. (2000). The “what” and “why” of goal pursuits: Human needs and the self-determination of behavior. Psychological Inquiry, 11(4), 227-268.
10. Gollwitzer, P. M., & Sheeran, P. (2006). Implementation intentions and goal achievement: A meta-analysis of effects and processes. Advances in Experimental Social Psychology, 38, 69-119.
11. James, W. (1890). The principles of psychology. Henry Holt.
12. Bandura, A. (1997). Self-efficacy: The exercise of control. W. H. Freeman.
13. Higgins, E. T. (1987). Self-discrepancy: A theory relating self and affect. Psychological Review, 94(3), 319-340.
14. Wood, W., & Neal, D. T. (2007). A new look at habits and the goal-behavior interface. Psychological Review, 114(4), 843-863.
15. Vallacher, R. R., & Wegner, D. M. (2012). Action identification theory. In P. A. M. Van Lange, A. W. Kruglanski, & E. T. Higgins (Eds.), Handbook of theories of social psychology (Vol. 1, pp. 327-348). Sage.

Related Reading

• Self-Determination Theory (Deci & Ryan): Autonomy, Competence, Relatedness
• Goal-Setting Theory (Locke & Latham): SMART Goals and Beyond
• Self-Discrepancy Theory (Higgins): Actual, Ideal, Ought Selves
• The Octalysis Framework: 8 Core Drives of Gamification

WOULD YOU LIKE YU-KAI CHOU TO WORK WITH YOUR ORGANIZATION?