AntiGPT

Turn AI drafts into publishable blog posts and scripts

By Vitalik Hakim · Published 20 September 2026 · Updated 18 October 2026

Freelance writers and content teams draft with AI because it is fast. Publishing the raw draft is where it goes wrong: readers bounce from prose that has no author, and platforms increasingly weigh whether content shows real experience and a point of view. AntiGPT sits between the draft and the publish button: it gives the post a voice and lets you check how it reads to a classifier before a client or editor does. Video and podcast scripts, written for the ear rather than the page, have their own sections below.

Why AI blog drafts underperform

A model's blog post is built from the average of every blog post: a tidy introduction that restates the title, subheadings that promise more than the paragraphs deliver, and a conclusion beginning "In conclusion". The sentences are the same length. Nothing in it could only have been written by someone who has done the thing. Readers feel this within a paragraph, and search engines increasingly reward the opposite: specificity, experience, and a recognisable author.

Humanizing addresses the surface of that problem (rhythm, register, stock phrases) and gives you a much better base to add the parts only you can add: the example from a real project, the opinion, the number you actually measured.

The freelancer's workflow

Draft the post with your assistant from a brief that includes the client's audience and one thing you personally know about the topic. Paste it into AntiGPT section by section in Balanced with Journalist readability, which favours clear, varied sentences over academic ones. Read each rewritten section and insert one specific detail of your own before moving on: a screenshot, a quote from a customer, a mistake you made.

Run the detector on the finished post. A lower estimated-AI score is useful as a signal that the text has stopped sounding generic; a client who runs a detector will see the same thing. Then do the final edit yourself: cut the introduction by half, make the subheadings say something, and end when you are done rather than with a summary.

Working at volume

If you publish daily, the Pro plan's 35,000 words a month and 3,000 words per request cover roughly one long post a day with room to re-humanize; Max doubles the per-request cap and triples the allowance for agencies. Long-text chunking means a 2,500-word post goes through in one action and comes back stitched. Words are counted on what you paste, never on the output, and failed requests are free.

Keep a house style: the same mode and readability setting for the same client keeps their posts consistent across weeks.

Disclosure and ethics

Follow your client's and the platform's rules on AI-assisted content; some require a disclosure line. Humanizing does not change whether assistance was used. It changes whether the piece is worth reading, and that is the part that keeps clients.

Draft fast, but with a seed

Give the assistant more than a title. Feed it the audience, the angle, and one thing you personally know about the topic: a number from your own work, a mistake you made, a customer you talked to. A draft built around a real seed is far easier to make human than one generated from a bare prompt, because there's something true in it to build on.

Accept that the draft is a scaffold, not a post. Its job is to get the structure and the obvious points down so you can spend your effort on the parts that matter.

The three edits that matter most

First, cut the introduction in half. AI intros restate the title and delay the point; readers and search engines both reward getting to it. Second, make every subheading a claim, not a label ('The three edits that matter most', not 'Editing'). Third, end when you're done, not with an 'In conclusion' that summarises what the reader just read.

Do those three things, add your specifics, humanize for rhythm, and you have a post that reads like a person wrote it because, in every way that matters, one did.

Scripts: writing for the ear, not the eye

Spoken language is shorter, looser, and more repetitive than written language, and it varies its rhythm constantly: a three-word sentence after a long one is what keeps a listener with you. AI scripts are the opposite: uniform, formal, and full of connectives nobody says aloud. Balanced breaks the rhythm and swaps the written transitions for spoken ones, which is most of what makes a script deliverable.

The real test is reading it aloud. Humanize a section, then say it. Where you stumble, the sentence is still too written, so re-humanize that part or cut it. A script you can say in one breath per line is a script that sounds natural on camera.

Scripts: keep the structure and the facts

Humanize the spoken prose, but protect the things that must stay exact: names, statistics, product features, legal disclaimers, and any sponsor read that has approved wording. Keep those out of the rewrite and reinsert them. If your script has on-screen cues or timestamps, strip them before humanizing and add them back after.

Work scene by scene or beat by beat. A hook, the setup, the payoff, the call to action: each has a different job and a different rhythm, and humanizing them separately lets you keep the tight hook the algorithm needs while loosening the body so it sounds human.

Scripts: your delivery is the last edit

No rewrite replaces saying it in your own voice. Treat the humanized script as a strong draft to perform, not a teleprompter to obey: the small changes you make as you speak are what finish the job of making it yours.

A worked example

Before

In today's fast-paced digital landscape, content marketing has become an essential strategy for businesses looking to engage their audience. By creating valuable and relevant content, companies can build trust, establish authority, and ultimately drive conversions.

After

Content marketing works when someone reads to the end. Most of what we published in our first year did not get that far. I know because we finally looked at the scroll data. The posts that did get finished had one thing in common: they answered a question a real customer had emailed us.

The opener becomes a claim, the writer appears ("I know because"), and a specific practice replaces the abstract benefits. This is a post a reader can trust.

More worked examples

A video script

Before

In today's video, we will be discussing three essential tips that will help you improve your productivity. Furthermore, we will explore how these techniques can be implemented into your daily routine. Without further ado, let us begin.

After

Okay. Three things that actually made me faster this year. Not apps. Not a new planner. The first one you'll hate, so let's start there.

The written intro becomes spoken: a false start ('Okay.'), a promise, a bit of tension ('you'll hate'), and no 'without further ado'. It sounds like the first ten seconds of a real video.

Try it on your own text

Paste a paragraph you drafted with AI help. You will see how a detector scores it, then you can rewrite it in your own voice. Free: 300 words a week, sign in with Google; your text is kept while you do.

0 words

Detector scores are estimates from third-party classifiers and change as those tools update. No text can be guaranteed to pass any detector. Use this to improve your writing, and follow your institution's or employer's rules on AI use.

Recommended strength for this use: Balanced. It is preset when you continue.

Questions

Will humanized posts rank?

Ranking depends on usefulness and experience, not on any tool. Humanizing removes the generic register; your specifics and expertise are what earn the ranking.

Can I humanize a 2,000-word post at once?

On Pro and Max, yes: it is chunked at paragraph boundaries. Free and Starter work best a section at a time.

Do you store my drafts?

History is on by default so you can return to a rewrite; switch it off in Settings to stop saving.

How do I keep sponsor reads and stats exact?

Keep approved wording, names and numbers out of the rewrite and reinsert them; strip on-screen cues before humanizing.

What's the real test?

Read it aloud. Where you stumble, the line is still too written. Re-humanize it or cut it.

All use cases