The Four Levels of Influence · Siddhant Goswami

I have started thinking about persuasion as a ladder rather than a single skill.

Not everyone who is persuasive affects us in the same way. Some people capture our attention. Some stay in our thoughts. Some change what we do. And a very small number influence us so strongly that we start influencing other people on their behalf.

I think there are four levels.

1. Attention

The first level is simply the ability to make someone stop.

You notice the person. You listen when they speak. When they enter a room, you pay attention. If they say hello, you are inclined to stop and have a conversation.

This is influence at its shallowest level, but it is still valuable because persuasion cannot happen without attention.

It is also the easiest level to manufacture.

Beauty can create it. Charisma can create it. Novelty can create it. A provocative headline can create it. A clickbait thumbnail can create it.

Someone might be extremely good at getting you to look without giving you anything worth remembering.

That is why attention alone is a weak measure of influence.

A million people can click on something and leave unchanged.

The test: Did you make me stop?

2. Reflection

The second level is harder.

You do not merely get my attention. You give me something that continues occupying my mind after you are gone.

Perhaps you ask an unusually good question.

Perhaps you tell me something I did not know.

Perhaps you give me a new way of looking at something familiar.

Perhaps we have a conversation and, several hours later, I suddenly find myself thinking about something you said.

That is a different kind of influence.

Attention lives largely in the moment. Reflection survives your absence.

This requires considerably more skill because you need more than salience. You need an idea interesting enough, surprising enough, useful enough, or unresolved enough that another person’s mind continues working on it.

A catchy headline can make someone click once.

It is much harder to repeatedly say things that people remember the next morning.

The test: Did you make me think?

3. Action

The third level is where persuasion becomes behavioural.

You do not merely make me notice something or think differently.

You make me do something I probably would not have done otherwise.

Imagine a friend telling you:

“There is an event tonight. I know you have never heard of the speaker, but trust me. Buy the ticket.”

And you buy it.

Or someone recommends a book, product, restaurant, course, investment, trip, or career decision strongly enough that you actually act.

There is an important boundary here.

Thinking is cheap.

People have thousands of interesting thoughts that never alter their behaviour.

Action introduces cost. I must spend money, time, reputation, effort, or opportunity.

The stronger the cost, the stronger the evidence of persuasion.

Getting someone to click a link is not the same as getting them to fly to another city.

Getting someone to like an idea is not the same as getting them to reorganise their life around it.

The test: Did you make me act?

4. Advocacy

The fourth level is the most interesting.

You do not merely persuade me.

You turn me into a persuader.

Imagine a musician you had barely heard of before attending their performance.

The show is so good that the following week you cannot stop talking about it.

You tell your friends.

Then you start telling them they have to attend the next show.

Perhaps one of them says they are busy and you start arguing with them:

“No, seriously. Cancel that. You need to see this person live.”

Something remarkable has happened.

The musician is no longer persuading your friend.

You are.

Their influence has propagated through you.

At Level 3, I change my behaviour because of you.

At Level 4, I start trying to change other people’s behaviour because of you.

The original source no longer even needs to be in the room.

This is what happens with remarkable artists, founders, teachers, products, ideas, political movements, communities and religions.

Their strongest believers become distribution.

The test: Did you make me persuade someone else?

The ladder

The progression looks like this:

Attention    You occupy my eyes and ears.
     |
     v
Reflection   You occupy my thoughts.
     |
     v
Action       You change my behaviour.
     |
     v
Advocacy     You change my behaviour so deeply that I begin
             changing the behaviour of others.

And perhaps the simplest way to distinguish all four is to ask: what remains after you leave the room?

At Level 1, perhaps nothing remains.

At Level 2, a thought remains.

At Level 3, a behaviour remains.

At Level 4, another persuader remains.

That last level may be the highest form of persuasion.

You have not merely transmitted an idea.

You have created another transmission node.

A field guide to the four levels

Everything above is a claim. What follows is the check.

The claim is that influence has depth, and that the test of depth is what remains after you leave the room. This guide does three things with it.

Checks each level against the psychology and communication literature, and says plainly where the claim is right, where it is unsupported, and where it needs a correction.
Gives each level one governing framework, a small set of tools, and a practice drill.
Gives one measurement per level, so you know which level you are actually operating at rather than which one you hope you are.

Read the verification first. It changes how you should use two of the most popular tools in the persuasion trade.

Verification

Claim	Status	Evidence
Persuasion is a ladder, not one skill	Supported, with a century of precedent	The AIDA model (Attention, Interest, Desire, Action) dates to E. St. Elmo Lewis around 1898, refined by E. K. Strong in 1925. The first three levels map almost exactly onto it. The fourth level, advocacy, is the real addition, and it is supported separately (see below).
Attention is the easiest level to manufacture and the weakest measure of influence	Supported	Elaboration Likelihood Model (Petty and Cacioppo, 1986): attitudes formed through peripheral cues (attractiveness, novelty, salience) are less stable, less resistant to counter-argument, and less predictive of behaviour than attitudes formed through central processing. Attention is a peripheral-route asset.
Reflection “survives your absence” and requires more than salience	Supported	ELM: central-route attitude change persists, resists counter-persuasion, and predicts behaviour. Self-generated arguments are more persuasive than received ones (Janis and King, 1954; Xu and Wegener, 2024). Curiosity requires a specific, bridgeable gap plus enough prior knowledge to notice it (Loewenstein, 1994).
Action introduces cost, and cost is evidence of persuasion	Supported	Cialdini’s commitment research: commitments that are active, public, and effortful bind hardest. The intention-behaviour gap is well documented; implementation intentions close it with a medium-to-large effect (d = 0.65 across 94 tests, Gollwitzer and Sheeran, 2006).
Advocacy: the persuaded person becomes a new source	Supported	Two-step flow (Lazarsfeld, Berelson and Gaudet, 1944; Katz and Lazarsfeld, 1955): in the original election studies, only about 5 percent of voters changed preference from direct media exposure; the rest were moved through interpersonal conversation with opinion leaders. Berger and Milkman (2012): sharing is driven by high-arousal emotion (awe, anger, anxiety), not by valence alone.
The levels should be cumulative	Supported by the structure of AIDA and hierarchy-of-effects models; not independently tested as a claim	Treat as a design principle, not a finding.
Intensity and repeatability are separate	Consistent with ELM and with the virality literature; the distinction is sound	A single viral hit is a peripheral event. Repeatable Level 2 requires a supply of central-route material.

Two corrections the research forces

Correction 1: the Zeigarnik effect is not what the content industry says it is. The popular claim that people remember unfinished tasks roughly twice as well as finished ones comes from a 1927 study. Replications from the 1960s onward failed to find the memory advantage, and a 2025 systematic review and meta-analysis found no memory benefit for unfinished tasks, only a general tendency to resume them (the Ovsiankina effect). What survives is a pull to return, not a stronger memory trace. Use open loops to create a return, not to make an idea “stick”.

Correction 2: asking people about their intentions moves behaviour, but only a little. The question-behaviour effect is real. A meta-analysis of 116 tests found d = 0.24, a small effect, larger for easy and socially desirable behaviours and smaller for hard or risky ones (Wood et al., 2016). A question is a Level 2 tool with a weak Level 3 spillover. Do not mistake it for a Level 3 tool.

One clarification

The fourth level is transmission, not loyalty. The literature agrees, and adds a mechanism: the person who transmits is persuading themselves in the act of persuading others. Role-play studies from Janis and King (1954) onward show that generating arguments for a position changes the generator’s own attitude more than passively receiving the same arguments. Advocacy is therefore not just the output of deep persuasion. It is also a cause of it. Every time your advocate argues for you, they climb higher on the ladder themselves.

Level 1: attention

Governing framework: the ELM peripheral route plus the arousal finding.

Attention is won by salience and arousal, not by argument. The ELM explains why this is cheap: peripheral processing needs no motivation and no cognitive effort from the audience. Berger and Milkman explain what wins: physiological activation. Content that produced awe, anger, or anxiety was shared more; content that produced sadness was shared less, even after controlling for how useful, surprising, or prominently placed it was.

What this means in practice:

Attention responds to contrast and activation. A claim that contradicts something the audience already believes is activating. A number that violates expectation is activating. A calm summary of a familiar point is not.
Attention does not respond to importance. The most important idea you have will not earn attention on its merits. It has to be dressed in contrast.
Attention decays on contact. If nothing follows it, nothing was transmitted.

Tools:

The contradiction opener. State a widely held belief, then the evidence that breaks it. “Most people think X. The data says the opposite.” This is the single most reliable activator in the research because it combines surprise with a small threat to the audience’s self-image as an informed person.
The viewer-as-subject frame. Make the audience the subject of the first sentence, not the topic. “You have made this mistake” outperforms “This mistake is common”.
Retellable numbers. A number is retellable if it is short, round enough to remember, and surprising. “15 minutes to 5 minutes” is retellable. “A 66.7 percent reduction in processing time” is not.
Loewenstein’s three conditions. A curiosity gap only opens when the audience already knows enough to notice a gap, the missing piece is specific, and the gap feels bridgeable. Attention hooks fail when the audience does not know enough to feel the gap, or when the gap is too vague to want closed.

Drill. Take one thing you are about to publish or present. Write five openers: one contradiction, one viewer-as-subject, one retellable number, one specific question, one plain statement of the topic. Show all five to three people for ten seconds each and ask which one they want to hear more about. The plain statement will lose. Note which of the other four wins, and why.

The measurement. Stop rate. In content: retention past the first three seconds, or click-through. In a room: did people stop the side conversation. Nothing else at this level counts.

Level 2: reflection

Governing framework: the ELM central route.

Reflection is the level at which persuasion starts to survive your absence, and ELM tells you the price of admission. Central-route processing requires two things from the audience at once: motivation (personal relevance, accountability, or a disposition to think) and ability (cognitive resources, prior knowledge, no distraction, a comprehensible message). Remove either and the audience drops to peripheral processing, and what you said will be forgotten by morning.

The payoff for meeting the price is well documented. Attitudes formed through central processing are held with more confidence, resist counter-argument, remain stable over time, and predict behaviour. This is the empirical basis for the claim above that Level 2 is where a thought remains.

What this means in practice:

You cannot make someone reflect. You can only create the conditions and supply material worth reflecting on.
The material must be theirs, not yours. Self-generated arguments beat received arguments because people generate reasons that already fit their own concerns, and because they do not mentally correct or discount arguments that came from inside their own head.
Questions outperform statements for this reason. An open question invites generation. A statement invites evaluation, which invites resistance.

Tools:

Derive before name. Present the problem first, let the audience feel it, let them attempt a solution, then supply the concept as the label for what they already understand. This is central-route processing engineered into a sequence. The concept survives because the audience did the work of arriving at it.
The unresolved question. End a segment on a question you do not answer. Per the corrected Zeigarnik finding, this will not make the content more memorable, but it will create a pull to return, which is what you want at Level 2. The question must meet Loewenstein’s conditions: specific, bridgeable, and grounded in something the audience already knows.
The self-generation prompt. Instead of “here is why this matters”, ask “why would this matter to you?” and leave silence. Health-behaviour studies found posters framed as questions (“Why would you choose healthier food?”) outperformed the same message as an instruction, with the caveat that the effect is stronger for people from individualist cultures and can reverse for collectivist audiences, who respond better to direct instruction. Know your room.
One principle per session. Central processing has a capacity limit. A catalogue of ten ideas forces the audience into peripheral mode by exhaustion. One idea, derived properly, is the only thing that fits.
The reframe. Give the audience a new lens for something familiar. Reflection is triggered when an old object becomes visible in a new way, because the audience has both the prior knowledge (ability) and a stake in the object (motivation).

Drill. After your next conversation, presentation, or piece of content, wait 24 hours and ask three people from the audience what they remember. Not what they liked, what they remember. If they remember the hook and nothing else, you operated at Level 1. If they can restate the central idea in their own words, especially if the words differ from yours, you reached Level 2. Their rewording is the evidence. Received ideas come back verbatim. Generated ideas come back paraphrased.

The measurement. Next-morning recall, in paraphrase. Ask for it directly. Do not accept “it was great” as data.

Level 3: action

Governing framework: Fogg’s B = MAP, with implementation intentions as the closing tool.

Behaviour happens when motivation, ability, and a prompt converge in the same moment. If any one is absent the behaviour does not occur, no matter how strong the other two are. This is the diagnostic. When someone agrees with you and does not act, it remains true that thinking is cheap, but Fogg tells you which of three things is missing, and it is almost never motivation.

The second half of the framework is the intention-behaviour gap. People who fully intend to act frequently do not. The most robust fix in the behaviour-change literature is the implementation intention: an if-then plan that specifies when, where, and how. Across 94 tests it produced a medium-to-large effect on goal attainment (d = 0.65), and it works by reducing ambiguity at the exact moment action is supposed to begin.

What this means in practice:

Your persuasion has not failed if the person is convinced and idle. Your design has failed. Check ability first (is it easy enough?), prompt second (is there a cue at the right moment?), motivation last.
Cost is evidence, as I said above, but cost is also friction. The Fogg model and the claim above are in tension here, and the resolution is sequencing: lower the cost of the first action, then raise the cost of subsequent commitments once identity is engaged.
Commitments bind in proportion to three properties: active (the person does something, ideally writes), public (someone else sees it), and effortful (it cost something). A fourth property, per Cialdini, is perceived free choice: if the commitment feels coerced or bribed, it does not bind.

Tools:

The B = MAP diagnostic. When action does not occur, run the three questions in order. Was there a prompt at the moment they could act? Could they do it right then, with what they had? Did they want to enough? Fix the first failure you find. Most failures are prompt and ability failures dressed up as motivation failures.
Implementation intentions, imposed from outside. Do not ask “will you do it?” Ask “when, where, and what is the first step?” Get the answer stated aloud or written. “If it is Tuesday at 6pm, then I open the file and write the first paragraph.” This is the single highest-yield tool here.
The written, public, effortful commitment. A form the person fills in themselves, visible to at least one other person, with a small cost attached. A commitment device with money attached (a deposit, a fee) converts cost-as-evidence into a mechanism.
The small first yes. Foot-in-the-door: a small compliant act makes a larger related act more likely, because the person now has a self-image to stay consistent with. Use it honestly: the small act must be one the person understands and would have chosen.
The pre-committed definition of done. Before work starts, define the binary, externally verifiable state that counts as done. This is an implementation intention applied to an outcome rather than a moment, and it removes the escape hatch of self-reported progress.

Drill. Pick one thing you want one specific person to do. Before you make the ask, write down: the prompt (what will cue them, and when), the ability (what will make it take under five minutes to start), and the implementation intention you will extract from them (if X, then Y). Make the ask. Then check in 48 hours. If they did not act, identify which of the three was missing. Do not conclude they were not persuaded.

The measurement. Verified behaviour with a cost attached. Registrations are weak evidence; attendance is stronger; paid attendance is strongest. Count the most costly thing, not the easiest thing.

Level 4: advocacy

Governing framework: two-step flow, plus self-persuasion through argument generation.

The two-step flow model found, against the researchers’ own expectations, that mass communication mostly did not move people directly. It moved opinion leaders, who then moved everyone else through conversation. Opinion leaders tend to be similar to the people they influence, have an intense involvement with a specific topic, and add their own interpretation to what they pass on. The observation that the original source no longer needs to be in the room is the two-step flow stated from the persuader’s side.

The second framework explains why advocates deepen rather than dilute. Role-play and self-generated argument studies show that constructing a case for a position changes the constructor’s own attitude more than receiving the case would have. Aronson’s summary: the most durable persuasion is the persuasion no one is seen to attempt. When your advocate argues for you, they persuade themselves in the same act. This is why advocacy compounds and why it is the top of the ladder rather than a side effect.

What this means in practice:

You do not create advocates by asking for referrals. You create them by giving them something they can retell, a reason to retell it, and a moment in which retelling is easy.
Retellability is a design property. It requires a short, surprising claim (Level 1 material), an idea the advocate has made their own (Level 2), and evidence from their own behaviour (Level 3). Advocacy is cumulative because the transmission needs all three.
Sharing is driven by arousal. Per Berger and Milkman, awe and anger travel; contentment and sadness do not. The advocate needs to feel something activating about your idea, not merely approve of it.
Identity is the strongest binder. Cialdini’s seventh principle, unity, holds that shared identity (“we”) pulls harder than mere liking or similarity. Advocates who see themselves as members of something, not customers of something, transmit on their own initiative.

Tools:

Hand them the argument, not the ad. Give advocates the contradiction opener and the retellable number from Level 1, framed as a claim they can make in their own voice. An advocate who can say “most people think X, but it turns out Y” in a sentence will say it. One who has to explain your product will not.
Ask them to make the case. Invite the persuaded person to explain to someone else why they acted. A short written or spoken account, a demo, a post. This is the role-play mechanism deployed on purpose. The act of explaining is the point; the audience for the explanation is secondary.
Make the advocate the opinion leader. Two-step flow says influence passes through people similar to the audience. Put your advocate in front of their own peers, not yours. A student explaining to future students outperforms an instructor explaining to them.
Build the “we”. Name the group. Give it a vocabulary, a shared object, and a shared enemy (a widely held belief you both reject). Unity is the mechanism by which advocacy stops needing prompts.
Engineer the arousal, honestly. The activating emotion for advocacy should be the same one that persuaded the advocate: the awe of seeing something work, or the anger at how much time the old way wasted. Do not manufacture outrage; borrow the one that was already there at Level 2.

Drill. Find one person who has acted on something you persuaded them of. Ask them to explain it to one other person, in their own words, and to tell you afterwards what they said. Read their explanation for two things: what they got wrong (a gap in your Level 2 material) and what they added (the interpretation an opinion leader supplies). Fix the first. Keep the second; it is often better than yours.

The measurement. Unprompted transmission. Not “would you recommend” but “did you, to whom, and what did you say”. A referral you asked for is Level 3 behaviour. A referral you did not ask for is Level 4.

The room test as an instrument

The question above, what remains after you leave the room, is a measurement protocol if you take it literally. Here it is, one metric per level, with the tool that moves it.

Level	What remains	Metric	Primary lever
1 Attention	Nothing, unless followed	Stop rate	Contrast, activation, viewer-as-subject
2 Reflection	A thought, in their words	Next-morning paraphrase recall	Derive before name, self-generation prompts
3 Action	A behaviour, with a cost	Verified costly action	B = MAP diagnostic, implementation intentions
4 Advocacy	Another persuader	Unprompted transmission	Argument handed over, “we” identity, role-play

Run the test in order. A Level 4 result without Level 2 evidence is almost always a peripheral-route fluke: a viral share of something no one understood. It will not repeat.

The ethics line, stated once

Every tool above has a legitimate and an illegitimate form, and the research is clear about which produces durable results. Self-persuasion works because it preserves the person’s freedom of choice; remove the freedom and reactance returns. Commitment binds only when the person perceives it as freely chosen; a tricked commitment produces the reverse. Open loops that never close produce resentment rather than return. Scarcity is the most overused principle in the set and the one most likely to trigger distrust.

The practical rule: every gap you open, close. Every claim you make, be able to defend to the person’s face. Every commitment you extract, make sure they would extract it from themselves. The ladder only carries weight when each level was earned.

Reading list, by level

Foundations

Petty, R. E. and Cacioppo, J. T. (1986). Communication and Persuasion: Central and Peripheral Routes to Attitude Change. The framework for Levels 1 and 2.
Strong, E. K. (1925). The Psychology of Selling and Advertising. The source for AIDA as usually cited; the 1898 Lewis origin is disputed in the historiography.

Level 1

Berger, J. and Milkman, K. L. (2012). “What Makes Online Content Viral?” Journal of Marketing Research.
Loewenstein, G. (1994). “The Psychology of Curiosity.” Psychological Bulletin.

Level 2

Janis, I. L. and King, B. T. (1954). “The Influence of Role Playing on Opinion Change.” Journal of Abnormal and Social Psychology.
Xu, M. and Wegener, D. T. (2024). “Persuasive Benefits of Self-Generated Arguments.” Social Psychological and Personality Science.
Wood, C. et al. (2016). “The Impact of Asking Intention or Self-Prediction Questions on Subsequent Behavior: A Meta-Analysis.” Personality and Social Psychology Review. Read this for the size of the effect, not the existence of it.
The 2025 systematic review of the Zeigarnik and Ovsiankina effects, for the correction.

Level 3

Fogg, B. J. (2019). Tiny Habits. B = MAP in full.
Gollwitzer, P. M. and Sheeran, P. (2006). “Implementation Intentions and Goal Achievement: A Meta-Analysis.” Advances in Experimental Social Psychology.
Cialdini, R. B. (2021). Influence, New and Expanded. The commitment chapter and the unity chapter.

Level 4

Katz, E. and Lazarsfeld, P. F. (1955). Personal Influence.
Aronson, E. (1999). “The Power of Self-Persuasion.” American Psychologist.
Rogers, E. M. (1962 and later editions). Diffusion of Innovations. For the opinion-leader and adopter-category apparatus.