Count choices, not visual complexity
Password Length and Randomness: Read the Numbers
Twenty characters can describe a random password, a repeated pattern, or part of a familiar quotation. Their visible lengths may match while their selection processes differ completely. Useful password math begins with how the value was chosen, not how complicated it appears on the screen.
Use the password generator to create new values, not to score an existing password. Character mode supports 12 to 128 characters with a default of 20. Passphrase mode supports four to ten independent word draws with a default of six. The settings describe different distributions, not interchangeable units of security.
Ask where every unpredictable choice comes from
Imagine two procedures. One chooses each character independently from an allowed pool. The other writes a familiar word and repeats it until the requested length is reached. Both can produce long output, but only the first procedure makes a fresh unpredictable choice at every position. Length multiplies possibilities only when the underlying generation model supports that multiplication. A predictable extension, such as another copy of the same pattern, is not equivalent to another independent draw.
The browser provides a cryptographically secure pseudorandom source through Web Crypto. That is different from a person trying to type randomly or selecting words that feel unrelated. It is also not physical dice. The useful assumption for the calculations is that the generator maps its random source into the permitted choices uniformly and performs the specified independent draws. A theoretical figure describes that model; it is not a laboratory measurement of fresh physical entropy inside your particular machine.
Once a result has been generated, avoid personal improvements such as adding your birth year, changing the first character to an initial, or substituting a favorite word. Those changes create a different distribution. There is no general rule that a secret becomes stronger because it looks more individualized. Set the constraints you actually need, generate within them, and use the resulting value unchanged rather than attaching an interpretation to its appearance.
Use the simple character formula only when it applies
For an unconstrained uniform character generator, let A be the number of allowed characters and L the length. The number of possible strings is A raised to L, and theoretical entropy is L multiplied by log2(A). This describes equally likely independent choices, not arbitrary text with the same alphabet. A poem does not receive that calculation simply because all its letters happen to belong to an allowed character pool.
For example, 20 independently selected lowercase English letters from a pool of 26 give about 94.0 theoretical bits. Twenty independent draws from 62 letters and digits give about 119.1. These examples assume lookalike exclusion is off and there is no requirement to include every selected class. They illustrate how the model works; they are not universal minimums or guarantees. Adding one character in either example multiplies the number of possible strings by that pool size.
Changing the pool and changing the length are different operations. More allowed characters add possibilities per position; more positions repeat the choice more times. If a destination disallows symbols, additional randomly selected letters and digits can still add possibilities within its maximum length. The sensible setting is a random value that fits the real form and intended workflow, not a pursuit of the largest number regardless of whether you can enter and retain the result accurately.
Account for class requirements and excluded characters
Character mode lets you select lowercase, uppercase, digits, and symbols. Include every selected character type is on by default and requires at least one character from each selected class. This restricts the set of permitted strings. Consequently, the unrestricted A-to-the-L count is not the exact count for that constrained setting. If strings are sampled uniformly over the valid set, entropy is the base-two logarithm of that set size; if a construction samples unevenly, its distribution must also be considered.
Excluding lookalikes removes the exact characters Il1O0o from whichever selected classes contain them. It can make manual entry less ambiguous, but it reduces the available pool. Do not keep using a full-pool calculation after making that change. Likewise, requiring all selected types is a compatibility choice, not proof of superior account security. NIST does not mandate a mixture of every character class. An individual service may still enforce its own form rules, which you need to check directly.
Use the right model
Choose characters or independent words
Set the length and compatibility options before generating a new result you will keep unchanged.
Count passphrases by word draws, not spelling length
Passphrase mode draws uniformly and independently from the 7,776-entry EFF long wordlist, allowing repeated words. For W words there are 7776 raised to W possible ordered sequences. The theoretical calculation is W multiplied by log2(7776). That is about 12.925 bits per word: roughly 51.7 for four words, 77.5 for six, and 129.2 for ten. The browser source is pseudorandom, but the intended selection model matches independent fair draws from this list.
A word with nine letters is still one list choice, just as a shorter word is one list choice. Counting every letter as an independent alphabet selection would overstate the possibilities because letters within a selected word are not independent draws. The fixed hyphen, space, or period separator contributes no additional entropy either. It changes the representation and total character length, which matters for form compatibility, without adding another random choice at each boundary.
A repeated word remains an allowed outcome of independent sampling. Removing duplicates, choosing only short words, or rearranging the sequence changes the process being analyzed. The same caution applies when composing your own sentence: its entropy cannot be inferred from the number of words or characters alone. Remember generated words with a private cue if helpful, but do not rewrite them into grammatical prose and continue quoting the original generated-distribution figure.
Separate search-space arithmetic from a real attack
An entropy figure has no clock built into it. Converting possibilities into a duration requires assumptions about guessing speed, the attack method, and what the attacker can access. Attempts against a live login can be limited by the service. Guesses against stolen password hashes depend on storage and computation costs. A single years-to-crack label would hide those distinctions. This generator therefore does not turn theoretical entropy into a prediction for your account or certify a password as unbreakable.
Some attacks bypass guessing altogether. A copied credential exposed to another application, an intercepted keystroke, or a deceptive sign-in page can reveal the exact value. Adding more random characters does not repair that exposure. Use a separate credential for each purpose and follow the account owner security requirements, including additional authentication where available. A generation tool is one part of credential handling; it cannot establish the trustworthiness of the destination, its recovery process, or your device.
Make a setting decision, then complete the handoff
First check the actual destination maximum, accepted symbols, and treatment of spaces. Start from the default 20 characters or six words, then adjust for those constraints and your entry needs. Generate returns a masked result; Reveal and Hide change visibility, Copy transfers it explicitly, and Clear removes it. Copy before changing tabs because hidden state, pagehide, pageshow, and freeze clear the result. Earlier clipboard copies are not erased by those events, so finish their handling separately.
There is no existing-password input, checker, breach lookup, network request for generation, saved history, storage, PDF export, or vault here. An opaque-origin sandboxed HTTP iframe isolates the generator from the surrounding page, but it does not secure HTTP transport or prevent a hostile parent from replacing the interface. Browser extensions and operating-system compromise remain outside that isolation. Treat the numbers as theoretical properties of newly generated distributions and the result as a secret requiring careful transfer, not as a real-account guarantee.
Questions about Length and Randomness
Does doubling password length double theoretical entropy?
For independent uniform draws from the same unconstrained pool, yes: L multiplied by log2(A) is linear in L. The number of possible strings grows exponentially. This statement does not apply automatically to repeated patterns, human-written text, or a setting with additional constraints.
Can I use the displayed theory to rate a password I already have?
No. The calculation concerns the specified new-generation process, not an existing string of unknown origin. This tool has no password input, strength checker, or breach lookup. Length and visible character types alone cannot establish the distribution that produced an old password.
Why is there no guaranteed crack time?
Time requires an attack model and a guessing rate, neither of which is determined by the generated value alone. Account controls, password storage, hardware, and direct credential theft can change the outcome. Theoretical entropy is not a promise about a real account.
Continue with another password guide
Make a fresh result in the Strong Password & Passphrase Generator or choose a related task below. Explore more practical utilities in all free tools.
Turn the calculation into a usable choice
Generate within the real form limits
Choose an accepted format, copy while the page is visible, and handle the clipboard separately.