For platforms, agencies and developers

AI humanizer API with the score before and after in every response

One endpoint, one key, plain JSON. Send text with a tone and a strength to POST /api/v1/humanize and get back the rewrite, our AI-pattern score before and after, and what is left on your plan.

The API is part of the Scale plan. Same engine, same tones and same limits as the box on this page, so what you test here is what your code gets.

Read the API reference See the Scale plan

Bearer key · JSON in, JSON out · 60 requests a minute

The short answer

Undetected.ai includes an API on the Scale plan: create a key at /account/api, call POST /api/v1/humanize, and every call is one run on your plan with the same 1,500-word ceiling as the web box. Elsewhere in the category, Undetectable.ai includes an API on every paid tier, which is the cheapest self-serve route in, and WriteHuman publishes a documented REST API priced by word. Walter Writes does not document a REST API but does ship an MCP server, as does WriteHuman. QuillBot has no public paraphrase or humanize API at all.

Competitor API and MCP availability was read off each vendor's own documentation on July 31, 2026.

API reference

How the Undetected.ai API works

Two endpoints, authenticated with a Bearer key you create in your account panel. Everything below is what the endpoints accept and return.

Authentication

On the Scale plan, open /account/api and create a key. Send it on every request as Authorization: Bearer <key>. An account holds up to 5 live keys, and you can revoke any of them at any time. A missing or revoked key gets a 401, and a key whose account is not on Scale gets a 403.

Limits

Each call to /humanize is one run on your plan, with the same limits as the web box: up to 1,500 words per call, the daily run cap, and the monthly words and runs of Scale (1,000,000 words across 2,000 runs a month). Requests are throttled at 60 a minute per account. A refused call returns the reason and the usage left, and nothing is counted.

POST /api/v1/humanize

curl https://undetected.ai/api/v1/humanize \
  -H "Authorization: Bearer YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text": "Your AI-drafted paragraph goes here.", "tone": "Standard", "strength": "Balanced"}'

Field	What it is
text (request)	Required. The text to rewrite, up to 1,500 words.
tone (request)	Optional. Standard, Academic, Casual or SEO. Standard if left out.
strength (request)	Optional. Light, Balanced or Strong. Balanced if left out.
output	The rewritten text.
score_before, score_after	Our AI-pattern score of your text and of the rewrite.
words	The word count charged to your plan for this call.
usage	What is left after this call: runs today, words this month and runs this month.

GET /api/v1/usage

Returns your plan, its limits (monthly words, monthly runs, daily runs and words per run), what is left of each, and when the monthly allowance resets. It does not use up a run, so call it before a large job to see what fits.

What Undetected.ai is

What you get, described accurately

So there is no confusion about what you would be buying, including what we do not offer.

Question	Answer
Programmatic access	Yes, on Scale. POST /api/v1/humanize and GET /api/v1/usage with a Bearer key. Starter and Pro do not include it.
Not offered	No SDKs, no webhooks, no MCP server, no SSO, no white label and no custom or volume pricing. The API is plain HTTPS with JSON.
Input	Text, plus the tone you want (Standard, Academic, Casual or SEO) and a Light, Balanced or Strong strength. The same three inputs as the box on this page.
Output	Humanized prose that preserves the facts, argument and structure of the original.
The score	Our own AI-pattern score, measured on sentence rhythm, template phrases, vocabulary variety and passive voice, returned before and after. It is not a GPTZero, Turnitin, Originality.ai, Copyleaks or ZeroGPT reading, and we have no access to those services.
Volume	Up to 1,500 words a run, on the web or through the API. Plans carry a monthly allowance of words and runs: 100,000 words across 300 runs a month at $24 a month billed yearly, up to 1,000,000 words across 2,000 runs a month at $99 a month billed yearly, which is the plan with the API.
Data handling	Your text and its rewrite are stored for 90 days so you can get back to them, then deleted; you can delete any of them yourself sooner. We train nothing on them. Processing goes through a third-party AI model provider, and the privacy page states what the provider may do with text on the tier we run on today.

What an API is for

Three things teams build with a humanizer API

Each of these works with our endpoint, and each is a fair test of any vendor you compare us with.

A humanize step in a product

If your app generates drafts, ads, listings or descriptions with a model, the humanize call becomes the last stage of the generation pipeline. Your users get output that reads like a person wrote it, and you never build the rewriting layer yourself.

A pass over a backlog

Agencies and publishers with hundreds of AI-assisted drafts want them run by a script rather than one by one. Split each draft into pieces of up to 1,500 words, send them in turn, and read usage from each response so the job stops before the allowance does.

A pre-publish gate in a CMS

Score every draft on the way to publish and hold back the ones that read as machine-written. This needs a score in the response, not just a rewrite, which is why ours returns score_before and score_after: several vendors return text alone.

The landscape

Which AI humanizers actually ship an API

Each vendor's own docs were read on July 31, 2026. If a row says the pricing is not published, that is because the vendor does not publish it, not because we did not look.

Tool	API	Agent callable (MCP)	How access works
Undetectable.ai	Yes, and unusually it is included on every paid tier rather than gated behind an enterprise plan	Not published	Self-serve. Buy any plan from about $5 a month billed yearly and the API comes with it
WriteHuman	Yes, a documented REST API with JavaScript, Python and cURL examples, priced by word	Yes, it publishes an MCP server	Self-serve, word-priced separately from the browser plans
Walter Writes AI	Not documented publicly	Yes, it publishes an MCP server	Browser plans from $8 a month billed yearly; the MCP server is the programmatic route
QuillBot	No publicly documented paraphrase or humanize API	No	Browser and extension only, at $8.33 a month billed yearly
Undetected.ai (us)	Yes, on Scale. One endpoint that returns the rewrite and the score before and after	No	Self-serve. Subscribe to Scale at $99 a month billed yearly and create a key at /account/api
The detectors	Copyleaks, Originality.ai and GPTZero all sell detection APIs, and none of them sells a humanizer	Varies	Useful as the scoring half of a pipeline, never the rewriting half

Two things in that table are worth acting on. First, if you want the cheapest possible programmatic access, Undetectable.ai bundling the API into every paid tier is genuinely hard to beat; we include ours on Scale only, so on price alone they win. Second, MCP is the interesting shift: two vendors now ship servers that let an AI agent call the humanize step as a tool, which means the rewrite can happen inside an agent workflow rather than inside your application code. If you are building agents rather than apps, ask about MCP before you ask about REST.

We keep every vendor claim on this site dated and re-checked. The full method, and every claim in this category you can verify for nothing, is in the AI humanizer that actually works.

Before you integrate anyone

What matters when the rewrite happens inside your own product

When the humanize step is behind your API, your users blame you for a bad one. These are the three questions worth answering before you write the integration.

Does it return a score you can act on?

A humanizer that returns only text forces you to trust it. One that returns a reading lets you gate on a threshold, retry, or show the user something. Ask what the number actually measures, too: an in-house heuristic and a third-party detector result are very different things to build a gate on.

Does the meaning survive?

Synonym-swapping humanizers score well and read badly, and inside an API that failure is invisible until a customer complains. Run your own worst paragraph through every candidate before you compare prices, and read the output rather than the score.

What happens to your customers' text?

Content sent through an API usually belongs to your customers, not to you. Get the retention period, the training policy and the name of the upstream model provider in writing, because "we do not train on it" says nothing about who you passed it to.

Good questions

Questions about the AI humanizer API

Does Undetected.ai have an AI humanizer API?

Yes, on the Scale plan. You create a key in your account panel at /account/api and send POST /api/v1/humanize with that key as a Bearer token and a JSON body holding the text, the tone and the strength. The response carries the rewrite, the AI-pattern score before and after, the word count and what is left on your plan. GET /api/v1/usage returns your limits and what remains.

Which plan includes the API?

Scale only. Starter and Pro do not include it. Scale is 1,000,000 words across 2,000 runs a month at $99 a month billed yearly, or $199 billed monthly, and the same plan also carries a team of 10 people, batch upload of .docx and .txt files and CSV export of the run history.

What are the API limits?

Each call is one run on your plan, with the same limits as the box on the website: up to 1,500 words per call, the daily run cap, and the monthly words and runs of the plan. Requests are throttled at 60 a minute per account, and an account can hold up to 5 live keys at once. A refused call says why in the response and counts nothing.

Which AI humanizers have an API?

Fewer than the search results suggest. Read off each vendor's own documentation on July 31, 2026: Undetectable.ai includes an API on every paid tier, and WriteHuman publishes a documented REST API with JavaScript, Python and cURL examples priced by word. Walter Writes does not document a REST API publicly. QuillBot has no public humanize or paraphrase API at all. Undetected.ai includes its API on the Scale plan only.

What is the best AI humanizer API?

It depends which constraint binds. For the cheapest self-serve access, Undetectable.ai bundles an API into every paid tier from about $5 a month billed yearly, which is cheaper than our Scale plan. For transparent per-word pricing and copy-paste code samples, WriteHuman publishes clear documentation. Ours is the one to pick if you want the AI-pattern score before and after in the same response as the rewrite, and the same engine you can test in the box on this page before you write any code.

How do I get an AI humanizer API key?

On Undetected.ai you subscribe to Scale, open /account/api and create a key; it is shown once, so copy it then. You can revoke a key at any time and create another, with up to 5 live at once. With other self-serve vendors the key also appears in your dashboard after you buy a plan. Either way, get a test call working before you write any integration code, because the shape of the response is what your code has to live with.

Is there a free AI humanizer API?

Not in any usable sense. Free browser tiers in this category run from 125 words per request to 750 words a month, and none of them extends to programmatic access, because an API with a free allowance would be scraped within a day. Our free trial is the box on this page, not the API. Budget for paid access from the start if you are putting this in production.

Can an AI agent call a humanizer as a tool?

Yes, and this is the newest change in the category. WriteHuman and Walter Writes both publish MCP servers, which let an AI agent invoke the humanize step as a tool. We do not publish an MCP server, webhooks or SDKs; our API is plain HTTPS with JSON, which any agent framework that can make an HTTP request can call.

Is there an open-source AI humanizer?

There are open-source projects on GitHub that paraphrase text with a local model, and they are a reasonable starting point if you want to own the whole stack. What they do not give you is a score in the same response, ongoing adjustment as detectors change, or anyone to call when output quality drops. Detection models get updated constantly, so an unmaintained rewriter degrades quietly rather than breaking loudly.

What would I build with a humanizer API?

The three common builds are a humanize step inside a content platform, a pipeline that cleans a backlog of AI-assisted drafts before publishing, and a pre-publish check inside a CMS. Because our response returns the score before and after, the third one works without a second service: gate on score_after and retry or hold the draft.

Is my text stored or used for training?

In the web app, yes it is stored: your text and its rewrite are saved for 90 days so you can get back to them, then deleted automatically, and you can delete any of them yourself at any time. We train nothing on them. Processing happens through a third-party AI model provider, and our privacy page states plainly what the provider may do with text on the tier we currently run on. Read that page before you send anything that belongs to a client.

Can I use Undetected.ai for a lot of text without the API?

Yes. A run handles up to 1,500 words, and the paid plans carry a monthly allowance of both words and runs: 1,000,000 words across 2000 runs on the largest. On Pro and Scale you can also upload up to 20 .docx or .txt files at once and get one ZIP back, each file split into runs of up to 1,500 words, every run counted like any other.

Explore more

AI humanizer for content teams

How agencies and marketing teams handle AI content at volume, with shared allowances, batch upload and the API.

Learn more

Best AI humanizer

An honest side-by-side of the leading humanizers, including which of them offer an API.

Compare tools

Humanize AI text for SEO

Publishing AI-assisted content that reads human and holds up in search.