Human Voice: Make AI Content Read Like You Wrote It
What it does: Strips the tells that make AI writing recognisable — identical sentence lengths, "In today’s fast-paced world", "In conclusion", and every post coming out in the same shape — and can write every article in your style, learned from your own published posts.
Applies to Antradus AI Publisher v2.7.0 · Last updated 2026-08-26
AI models are trained to produce the most average, most agreeable version of a sentence. That is exactly what makes their writing recognisable: perfect grammar, even rhythm, the same twelve words in every sentence, and a summary paragraph at the end that repeats what you just read.
Human Voice attacks that on five fronts. Four of them are automatic. The fifth is you.
1. Every article is written the human way (automatic)
Once you have Publisher, this is already on. Open Antradus AI → Human Voice to see it.
| Strength | What the AI is told |
|---|---|
| Natural (default) | Vary sentence length hard, open with something specific, active voice, concrete numbers and names instead of vague claims, no summary/conclusion paragraph, and a hard ban on the phrases below. |
| Strong | Everything above plus: fragments and one-word sentences, rhetorical questions, an aside or two, deliberately uneven sections, and almost no bullet lists. |
| Off | No humanizing rules at all. A saved Write Like You profile (below) still applies while its own box is ticked. |
This applies everywhere the plugin writes: single articles, Bulk Publishing, YouTube-to-article, and WooCommerce product descriptions.
The banned phrase list
Around 60 phrases are banned outright — delve, tapestry, in today's fast-paced, moreover, furthermore, it's worth noting, in conclusion, game-changer, seamless, robust, leverage, and the rest of the usual suspects.
The list is a plain text box. Add the words your industry over-uses, delete any you actually want, or click Restore default list to start again. Developers can filter it in code:
add_filter( 'antradus_ai_banned_phrases', function ( $phrases ) {
$phrases[] = 'synergy';
return $phrases;
} );One thing it will not do
The AI is never told to invent personal stories, statistics, studies, or customer quotes to sound more human. That would put false claims out under your name, and a fabricated anecdote is worse than a bland paragraph. Where a real detail would help, it uses a clearly-framed hypothetical instead.
Which is why the last step is still yours — see the part only you can do.
2. A different shape for every article (automatic)
Good sentences are only half of it. The other giveaway is shape: two articles on completely different subjects that open the same way, run the same number of sections, and land the same way. Put them side by side and they read as one template with the nouns swapped — however well each individual sentence is written.
There are three section plans, and by default the plugin cycles through them.
| Blueprint | How the article is built |
|---|---|
| 📄 Classic explainer | Hook, what it is and why it matters, two to four sections on how it works, an honest section on the limits and trade-offs, then what to do next. |
| ❓ Question-led | Answers the main question in the first sentence, then every heading is a real question a reader would type — each answered immediately, including one uncomfortable question a marketing page would skip. |
| 🔢 Ranked rundown | Strongest first, one heading per item, each entry saying what it is, what it does better, who it suits and where it falls short — ending in a verdict. When a topic has no comparable options, it ranks the factors that matter instead. |
Set it under Antradus AI → Human Voice → Article structure:
- 🔄 Rotate automatically (default) — a different shape each article, in turn. No two in a row share a skeleton.
- Or lock one blueprint for the whole site.
Rotation matters most in Bulk Publishing, where thirty keywords used to produce thirty identical layouts. Each job in a batch takes the next shape in the cycle.
To override it for one post, use the ARTICLE STRUCTURE dropdown in the Antradus AI panel. Leave it on Site default and the rotation carries on as normal. After the article is written, the status line names the shape you got.
The blueprint plans the sections only. Your language, style, tone, keyphrase placement, FAQ setting and word count are untouched. WooCommerce product descriptions ignore it completely — they have their own format.
3. Write Like You — a style profile from your own posts
Instead of a generic "human" voice, the AI can copy yours.
- Go to Antradus AI → Human Voice → Write Like You.
- Paste 2–5 of your own published posts into the samples box (about 250 words minimum, the more the better) — or click Use my recent posts instead to pull your 5 latest published posts automatically.
- Click 🧬 Analyse my writing.
- A style profile appears: sentence rhythm, vocabulary, how you address the reader, how strong your opinions are, how you open and close, your signature moves, and what you never do.
- Read it, edit anything that is wrong (it is a normal text box — plain instructions like "never use bullet lists" work fine), and click Save Changes.
From then on that profile goes into every article the AI writes.
⚠️ Paste writing you actually wrote. If you feed it posts this plugin generated, you are teaching the AI to imitate itself. Your samples are used once to build the profile and are never stored.
4. Humanize something already written
Two ways, depending on how much you want to change.
The whole post — in the Antradus AI panel, click 🧍 Humanize this post. The AI rewrites the body for rhythm and wording while keeping every fact, number, link, image, heading and section exactly where it was. Best on posts under ~2,000 words; longer ones are refused rather than silently cut short.
One paragraph — highlight it, open the AI ✨ menu in the toolbar, and pick 🧍 Humanize (stop sounding like AI). Works in both the Classic and Block editors, on posts of any length.
Ctrl+Z brings the original back either way.
5. Human Check — see the tells before you publish
Click 🔍 Check how human this reads. This one is instant and costs nothing — it does not call the AI at all, it reads the post right there in your browser session.
You get a score out of 100 and a list of what stands out, worst first:
- Sentence rhythm — the strongest signal. Uniform sentence lengths are the clearest sign of machine writing.
- Paragraph shape — every paragraph the same size is a tell.
- AI-tell phrases — each one found, with how many times.
- Formulaic connectors — sentences starting with Moreover, Furthermore, Additionally.
- Hedging — may, might, generally, typically piling up.
- Passive voice — "the product was purchased by the customer".
- Em dash overload and "Conclusion"-style headings.
| Score | What it means |
|---|---|
| 80–100 | Reads human |
| 60–79 | A few AI tells |
| 40–59 | Sounds AI-written in places |
| 0–39 | Reads like AI |
The phrase, hedging and passive-voice checks are English-based. On a post in Arabic, Chinese, Russian or any other non-Latin script, the check tells you so and scores only the parts that work in every language (rhythm, paragraph shape, punctuation).
The part only you can do
A model can vary its sentence lengths. It cannot tell your reader about the client who called at 11pm, the price you actually paid, or the thing you got wrong for two years.
The workable habit: let the AI build the structure, then spend five minutes adding what only you know.
- Swap one abstraction for a real detail. "Financial challenges" → "my income that month was $500 and rent was due."
- Break the smoothest paragraph. Find the most polished one and put a four-word sentence in it.
- Add one anecdote as a bridge before a list or a new section — two sentences is plenty.
- Make the ending yours. Formal sign-offs read like a brochure; write it the way you’d say it.
- Run Human Check again and see the score move.
Troubleshooting
The score barely moved after Humanize. Some models follow style instructions loosely. Switch to a stronger model in Antradus AI → AI Providers, or set Strength to Strong.
My article lost a section. Humanize is not allowed to remove sections — if you see this, the model over-edited. Ctrl+Z, then humanize one section at a time with the AI ✨ menu.
A word I use constantly is flagged. Delete it from the banned list in the settings — it is your list.
Human Check says a good post is robotic. Look at the findings, not the number. If the only finding is rhythm and you are happy with how it reads, publish it.