Somewhere in every anime fan group there is one person with a folder. It holds screenshots, timestamps, a running tally of who has hurt whom, and a list of statements pulled from chapters that will never appear on screen. They built it because the shows themselves refuse to hand out a number, and because “I think he wins” loses an argument in about four seconds flat. Power scaling is the habit of constructing that number yourself. It works more often than it should, and it fails in a handful of specific, predictable ways that are worth learning before you argue with anyone.
This is a method guide rather than a ranking. By the end you should be able to take two characters, work out what is actually on the page, and reach a conclusion you could defend out loud.
Where the whole thing came from
Power scaling is the practice of working out how strong a character is by comparing them to other characters from the same story. That is the whole definition, and the logic underneath it is old enough that you already use it in your head every day. If A defeats B, and B defeats C, then A is stronger than C. That chain is a transitive relation, and it is the entire engine of the practice. The same logic runs the other direction as well: if A can lift cars and B has demonstrated they are more powerful than A, it is safe to assume B can also lift cars. Nothing about that is anime-specific. Scaling a character is borrowing statistics from a neighbour who is standing closer to the same finish line. Scaling a character is borrowing statistics from a neighbour who is standing closer to the same finish line.
So why bother? Because fight choreography in shonen is deliberately imprecise, and imprecision is not the same as no information. A character shown taking three hits before landing one is still a data point, even though nobody announced a number. The trouble is that a fight only shows you the parts the author needed you to see, which is why scaling became a discipline with written rules rather than casual flexing: raw feats alone misrepresent almost everyone. The wiki is blunt about it, noting that relying purely on feats would rate a string of characters the story treats as planet-level as wall-level, and would place famous fighters below characters plainly shown to be their inferiors. That is the whole method in one sentence.
Ranked lists, tier lists and best-of polls are the internet’s default way to have an opinion about fiction, and scaling is that habit for stories where the fighting is the point. Knowing the rules gets you a fair shot, which matters given the vocabulary is unwritten and everybody arrived with a different rulebook.
Attack potency is not the size of the crater
Attack potency, usually shortened to AP, is how much energy a character can put behind an attack and how powerful that attack is when it lands. The word people misuse constantly is “how.” Attack potency is not a measurement of the crater. The clearest correction available comes straight from the method’s own documentation, which notes that in fiction a character’s attack potency is not always equal to how much they destroy or the area their attack covers, because attacks generally do not have to obliterate a city in order to output the energy equivalent to doing so.
Dragon Ball Super supplies a clean illustration. During the ceremony where the Gods of Destruction arrive on Earth, Beerus annihilates an entire alternate universe, and the survivors inside it are the weakened versions of Goku and Vegeta pulled out of their stronger forms. The area is enormous; the energy per person is not. Scaling that as city-destroying AP delivered to a target is a category error, and it is the same mistake made with a shockwave filling the screen in a later shonen fight. A technique scattering buildings across six blocks is not six times more potent than a tight punch that puts one fighter through a wall.
Concentration is the lever. A focused attack carries disproportionate force because none of the output is spread across a volume. The community’s documented statistics treat attack potency, striking strength and lifting strength as separate columns for exactly this reason, and hold that someone who survives a planet detonating can still be hurt by a punch that sends them through a single building. Both figures coexist, because surviving a planetary blast and having your ribs broken are different quantities of force applied differently. When an argument turns on the crater, ask what the attack was doing to whom instead.
Speed gets its own axis
Speed is the statistic that generates the most confident arguments and the least informed ones. The reason is that fans use one word for four separate measurements. A character who dodges a bullet has shown reaction speed, which is not the same claim as moving at the speed of a bullet. A character who crosses a continent in a moment has shown travel speed, which says almost nothing about what happens in a duel. Combat speed is what actually matters, and a character can be excellent at it while being hopeless at the other two, or the reverse.
Transitive logic applies to speed in the same way it applies to strength, and the wiki uses exactly that example. If A outraces light and B speed blitzes A, then B is faster than light, and the chain extends as far as you can carry it. That is a legitimate inference, and it is also the point at which people stop reading. A blitz is a specific claim: the second character moved faster than the first could react to. A dash that looks fast because of frame timing is not a blitz, and a cutscene that implies speed through motion blur is not a blitz either. A character who steps sideways while a punch passes through them has demonstrated evasion, not velocity.
The trap that catches almost everyone is the teleport. Instant movement defeats an attack whether or not the character is fast, because the attack arrives where they already left, which is why conventions like versatile speed exist to stop people reading that as a speed feat. Where a story is obsessive about frame counts, the speed claims tend to hold up. My Hero Academia, Jujutsu Kaisen and Chainsaw Man generate more speed arguments than they can resolve, and in all three the answer turns on whether the debate is about the animation or the stated abilities.
Durability decides how long a fight lasts
Durability is what a character can take, and it is the statistic that decides whether a fight is a conversation or a single exchange. If one fighter can only absorb a car crash and the other can be dropped by a thrown car, the fight has one round in it regardless of how much more damage the second fighter can output. Fighters who can go the distance keep arguing, and fighters who cannot are usually gone before the audience has finished reading the flashback.
Blunt force, piercing and slashing get separated on the serious character writeups, and the split is not pedantry. A character who survives being thrown through a fortress wall has shown resistance to blunt trauma at that level, which says nothing about a thin blade finding the gap in the armour. Attack on Titan is built on that gap. The human fighters cannot take a fraction of what the titans deal out, and the tension of the show rests on blades precise enough to matter anyway. Swap in a hammer and the tactical picture changes without altering a single character stat.
The other half of durability is that it applies only to attacks actually being physically withstood. A body can be pulverised while a technique is still in effect, which is why the same character can be a fragile human in a fistfight and an overwhelming problem in a spell contest. Regeneration, immortality and a body that reconstitutes are abilities rather than statistics, and they make a character effectively durable past what the feats suggest. Keep those in separate columns, because a fight where one fighter’s durability is irrelevant is a fight decided by something other than durability.
Hax is where arguments get settled
Hax is the community’s word for abilities that skip the raw numbers entirely. Sealing, mind control, intangibility, existence erasure, regeneration, time manipulation and portal creation all work by changing the rules rather than by applying more force, and hax is the layer that decides fights the stats cannot. It is also the most misrepresented category in the discipline, usually by people who treat an ability as automatic.
The method’s documentation gives the correction directly, and it is worth reading twice. If A has city-level durability and B harms A using hax that negates durability, it is not safe to give B city-level attack potency, because B never used raw power to overwhelm that durability. B demonstrated an ability that works on a different layer. Jujutsu Kaisen is the clearest modern case. Gojo’s Infinity is built on the convergence of an infinite series and slows anything moving toward him, and Sukuna’s domain is an attack that removes the entire area of the technique itself, so the winner of their exchange is decided by who has a domain and whether the barrier functions. Neither result is a measurement of how hard either of them punches, and a scale that ranks them by force alone has described nothing that happened.
The second correction is about conditions, and it is where most published hot takes go wrong. Sealing requires a successful landing, time stop requires consciousness, and regeneration requires not being incinerated first. A fight where hax decides the outcome is a sequence, not an automatic, and the sequence is usually settled by who acts first and who was prepared. That is why a versus thread is a different problem from a scaling exercise. Scaling asks what a character can do. A matchup asks what two characters will actually do to each other, and the two disagree more often than fans expect.
How a scaling chain gets built
A chain is a sequence where each character inherits statistics from another. The documented example: A performs a city-destroying feat and shows city-level durability, and B defeats A, critically injures A, or is shown as physically superior to A. It is then safe to give B city-level attack potency and durability despite B having no direct feat of that size. Add a third link and the logic carries further, with C inheriting the city feat two steps removed from the only person who ever performed it.
The chain also runs the other way, which surprises people. If A has a city-level feat and B lost the fight but still put up a considerable struggle, harmed A, and forced A to make a real effort, then B is city level on the lower end of that bracket. A fighter who cannot scratch the other one is not sharing a tier, but one who lands hits and survives is in the same neighbourhood. The wiki frames the same point through tier widths, noting that the gap between small and large planet tiers is nearly thirty million times, so anyone landing blows on a large-planet fighter has to be in that broad bracket even while losing.
Some series hand you their own numbers. Dragon Ball publishes power levels, Naruto runs on chakra natures, and Hunter x Hunter has Nen with an exponent attached to nearly every technique, Gon losing access to it entirely after the Chimera Ant arc. When a story’s in-verse ratings are consistent they work as supporting evidence rather than being ignored. The documentation’s advice is to prefer a character’s own feats first and fall back on stated statistics only when the system holds together, and it warns against treating numbers as linear, since a value of a thousand is not twenty times the strength of fifty. The numbers rank order. They do not measure magnitude.
Longer chains are not automatically better. Each link is a chance for the writer to have been inconsistent, and the community’s rules treat consistency, chain length and how the work presents the result as the factors deciding whether a rating is defensible. They also allow a limited upscaling: a character may reach the next step up when the gap to it is small, with 1.2 times or less cited as the workable boundary. Anything wider is left alone because the chain gets fragile.
The four places a chain breaks
Chains fail in four recurring ways, and each one has a documented example that makes it much harder to argue past.
The first is an opponent who was not trying. If A has city-level stats and B takes hits from A, it is not safe to give B city-level stats when A was holding back, restricting themselves, or running on a wound, because B never met the real version. Watch for characters who fight the entire arc at partial power and then reveal a reserve in the last three episodes. The scaling you did at episode 12 and the scaling that matters at episode 300 are different documents.
The second is a fight that avoided the ability that matters. The wiki’s example is precise: A is human-level in a straight fight but wields telekinesis at building level, and B beats A physically while that telekinesis goes unused. B does not get building-level stats, because A’s body was never what B was beating. This one catches out even careful scalers, because the fight looked like a normal brawl and the entire mismatch was hidden in a power nobody deployed.
The third is an outcome the story never intended. When a character’s upper limits are consistently modest and they suddenly beat someone far above that, the community wiki files it under Outliers or Plot-Induced Stupidity and declines to scale from it. Authorial need is a real force in serialized fiction. A story will hand a protagonist a surge whenever the plot requires the protagonist to matter, and treating that surge as the new baseline leaves you with a character rated from the one episode where they were pushed furthest.
The fourth is version drift. Anime cuts arcs, light novels get adapted loosely, films contradict the series they came from, and filler sits at odds with the chapter it was written around. Every link in a chain needs the same question asked of it: which chapter, which episode, which adaptation. A chain does not have to be wrong to be unusable. It has to be checkable, and version drift is the most common reason it cannot be.
Statements are clues, not verdicts
Long-running shonen are built on boastful characters, so a large share of the available evidence is dialogue written to make the speaker impressive in the current scene. The community splits on how to treat this, and that split is the source of most flame wars with nothing to do with the feats.
One position is that in-verse statements are canon and should be taken at face value, on the grounds that the story presents them as true. One Piece is the obvious example, with Luffy described repeatedly as the man who will become King of the Pirates and a late reveal reframing his inherited ability in terms the story had withheld for most of the run. The other position is that hyperbole is a narrative device, that no shonen character is a reliable narrator about their own power, and that statements belong in a separate category from demonstrated capability. The documentation lists Hyperbole and Statements as their own topics for exactly this reason.
The workable position is narrower than either camp. A statement is strong evidence about how the story wants a character perceived and how consistently it treats them, which makes it useful for consistency checks and for bounding a ceiling. It is weak evidence about raw magnitude, because the same scene usually hands you a character boasting while outmatched. Calling yourself the strongest in the room is a fact about the scene. What you can actually do is a fact about the fight.
One category carries more weight, and it is worth isolating. When a character with direct knowledge of a feat describes it to another character, and the description matches what we later see, you have something closer to evidence. Credibility and consistency do the work; unqualified bragging does not. Neither can ever stand in for a feat, though. A statement can justify a ceiling, but it cannot manufacture a floor.
Why careful scalers lowball on purpose
Lowballing means rating a character only as high as what has actually been shown on the page, and it drives people who like those characters genuinely mad. It is also the correct default, and the method’s documentation explains why in unusually blunt terms: without scaling, characters the story consistently portrays on par with planet-level fighters end up rated wall-level, and well-known fighters get placed below characters who are plainly their inferiors in-story. Going strictly by feats is not a neutral choice between accuracy and modesty. It is a specific and wrong answer that happens to flatter nobody in particular.
The case for lowballing is about distrust. Serialized fiction has a structural incentive to overstate, and a character revealed to be stronger is a story beat. The story is not obliged to tell you in advance which beats are planned and which are improvised. Take every momentary maximum as the new baseline and you end up rating characters by their most convenient appearance. Rating low and noting a possible ceiling keeps both errors visible instead of hiding one inside your numbers.
The cost is real. Lowballed matchups are dull, and a fan who has watched a character fight for two hundred episodes reasonably resents being told the character is weaker than someone they visibly outclassed. That is where equality statements earn their place. When a story frames two characters as equals, or presents a fight as genuinely even rather than as a protagonist overcoming an obstacle, that is evidence about parity and can cap both in the same bracket. The mistake is reading a parity claim as a number. It only says close, and close is what lowballing cannot see on its own.
Running your own comparison with Otaku AI
Doing this by hand means keeping a document, and most people abandon it around the third character because the bookkeeping is dull and the arguments are fun. Otaku AI handles the boring half: its power mode works through canon feats, statements and established scaling, naming the specific moments it relies on rather than returning a bare verdict, so a number arrives attached to the scene that produced it.
The quality of the answer depends almost entirely on how you ask. Name both characters, specify the version and the point in the timeline, and state the win condition, because those details decide more matchups than any statistic. “Gojo versus Sukuna” is a different question from “Gojo at full strength after Shibuya versus Sukuna with Mahito merged in, Gojo losing if he cannot hold his concentration.”
Worth knowing before you start: how Otaku AI works across every mode is the short version of what the tool can and cannot do. When you want the reasoning rather than the result, the matchup walkthrough on comparing anime characters properly covers versus conditions in detail. For checking that you are arguing about the right thing, series metadata and episode listings through the AniList API documentation settle version questions faster than any search.
Then do the part the tool cannot. Open the chapter, watch the scene, check whether the character was holding back. A confident answer from a language model is a starting position rather than a verdict. Enjoy the confidence, then open the source.
If you want to test yourself before trusting anyone’s numbers, the collection of anime character quiz ideas makes a decent warm-up for a fan group, and if the argument is really about who you would rather be than who would win, working out what anime character you are most like is the more useful detour. When it is time to build your own, creating an anime character from scratch beats copying a stat sheet every time.
