How to sound less formal in AI writing

You paste your AI draft into a doc. The facts are right. The structure is clean. But it reads like a corporate press release.

Too many formal words. Too many transition phrases. Every sentence lands with the same weight. Nothing breathes.

This is the default AI voice. It's not broken. It's just boring.

The good news: you can fix it without rewriting the whole thing. Small, targeted edits to rhythm, word choice, and structure make an AI draft read like it came from a human. Here are seven techniques that actually work.

Why AI writing sounds formal (and why it matters)

AI models are trained to be correct. Correct grammar. Correct structure. Correct tone. The problem is that human writing is rarely correct in the textbook sense. Real people write in fragments. They repeat themselves. They trail off and pick back up.

Wikipedia's editors catalogued this in their Signs of AI Writing guide. They found that AI text follows predictable patterns: uniform sentence length, overuse of academic-sounding connectors, and a tendency to sound like it's building toward a TED Talk reveal. Every single time.

Readers notice this. Even if they can't name what's wrong, they feel it. The text scans as "off" and they disengage.

For content creators, founders, and anyone publishing under their name, this matters. If your writing sounds like everyone else's AI output, readers stop trusting it. They stop reading entirely.

Break up uniform sentence rhythm

AI writing tends to produce sentences of roughly equal length. Ten to fifteen words. Same cadence. Same structure. Subject, verb, object. Repeat.

Real writing does not move like that. It stretches. It contracts. Short sentences hit harder after long ones. A one-word sentence lands like a punch.

Here is a simple test: scan your draft and look at the right margin. If every line ends at roughly the same place, your rhythm is too uniform. Break it up.

Try combining two short sentences with a comma. Then follow with a one-line paragraph that stands alone. The variation in pacing tells the reader's brain: a person wrote this.

Read your paragraph aloud. If you're out of breath halfway through, split it. If every sentence takes the same number of syllables to finish, break one in half or triple its length. The goal is variety, not symmetry.

Replace formal transition words with natural ones

AI loves academic connectors. Words that end in "more" or "less." Multi-syllable transitions that would feel at home in a legal brief. These markers scream "I was written by a language model trained on research papers."

Real people do not talk like that. They say "also," "but," "so," and "and." The simplest transitions are usually the best ones.

Here is a quick swap list. Replace these formal transitions with their natural alternatives:

"In addition" becomes "Plus" or "And." "However" becomes "But" at the start of a sentence. Yes, you can start a sentence with "but." "Therefore" becomes "So." Academic transition words with three or more syllables become one or two syllable alternatives.

Write like you talk (without the filler)

Imagine explaining your topic to a friend over coffee. You would not say "the implementation of this methodology yields significant improvements in workflow efficiency." You would say "this approach saves you about three hours a week."

This does not mean writing exactly like you talk. Spoken language has filler words, false starts, and grammatical chaos. Written language needs more structure. But the gap between "how you talk" and "how you write" should be small, not a canyon.

If you want a deeper read on making your AI output sound more natural, check out our guide on how to make AI writing sound natural.

Add concrete details instead of abstractions

AI writing defaults to the abstract. "Many companies struggle with content quality." Which companies? What kind of content? What does "struggle" look like?

Abstract language is safe because it commits to nothing. But safe writing is forgettable writing. Replace vague claims with specific examples:

"Teams save time" becomes "Our content team cut editing time from 4 hours per post to 45 minutes." "Many tools exist" becomes "We tested seven AI detection tools and three of them flagged Shakespeare as machine-generated."

Specific details do two things. First, they prove you actually know what you're talking about instead of remixing surface-level knowledge. Second, they give the reader something concrete to remember. Nobody remembers "improve your workflow." They remember "we cut 3 hours off editing time."

Use contractions and short words

One of the fastest ways to spot AI text is the absence of contractions. "It is" instead of "it's." "Do not" instead of "don't." "We are" instead of "we're."

Humans use contractions constantly in speech and informal writing. AI defaults to the expanded form because it's been trained to be precise. But precision at the cost of naturalness is a bad trade.

Go through your draft and expand nothing. Contract everything. Drop "it is" for "it's." Swap "that is" for "that's." Trade "you will" for "you'll." The cumulative effect is significant. The text immediately sheds about 30% of its stiffness.

Same logic applies to word choice. "Utilize" becomes "use." "Implement" becomes "do" or "try." "Facilitate" becomes "help." Short Anglo-Saxon words land harder than Latinate ones. Hemingway knew this. So should you.

Read it out loud before you ship

This is the oldest editing trick in the book and it works especially well for AI drafts. Reading aloud exposes stiffness that your eyes skip over on the screen.

When you hit a sentence that's hard to say, it's hard to read. Mark it. Shorten it. Split it. When a paragraph runs out of air before you do, it's too long. When you stumble on a word, swap it for something simpler.

Most people skip this step because it feels silly. Do it anyway. Three minutes of reading aloud catches more stiffness than twenty minutes of silent editing.

If you want to add more personality beyond just de-formalizing, our guide on adding personality to AI writing covers voice, anecdotes, and the specific edits that make text feel like yours.

The 2026 tone test: what readers notice first

Detector research and reader surveys keep converging on the same finding: tone is the loudest tell. A 2026 analysis of reader reactions found that formal register was the single strongest signal that text was machine-written, stronger than word choice or factual accuracy.

The reason is simple. Language models default to the safest, most correct version of every sentence. That produces the corporate press release voice: precise, balanced, and completely forgettable. Readers do not consciously notice the register. They just feel that nobody is home.

The fix has not changed, but the stakes have. Every technique in this guide, breaking rhythm, swapping transitions, contracting, reading aloud, directly attacks the formal register. Run even three of them and the draft immediately sounds like a person wrote it.

One useful 2026 shortcut: write your first sentence before you prompt the AI. A conversational opener anchors the whole piece. The model will match the register you set, so a casual first line prevents the formal drift that usually follows.

If you want to go further than de-formalizing, the guide on how to make AI writing sound human covers the full set of AI tells and how to fix each one.

And for a tool-assisted pass, our comparison of the best AI humanizer tools shows which ones strip formality without mangling your meaning.

Making AI writing sound less formal is not about dumbing things down. It is about removing the layer of polish that signals "a machine wrote this" and replacing it with the natural variation that signals "a person wrote this."

The August 2026 style score: why tone now shows up in detector output

August 2026 brought a shift that matters for this guide. Several major detectors now return a separate style score alongside the usual probability. That score measures exactly what this article covers: sentence rhythm, formal connectors, and the presence of contractions. Tone is no longer an invisible quality. It is a number.

The practical effect: the seven techniques here are not just about sounding better. They move a number that readers and tools both check. A draft that passes on perplexity can still fail on style if every sentence lands at the same length and the verbs are all in the expanded form.

The fastest win stays the same. Contract every verb, break one sentence short after a long one, and cut the academic connectors at paragraph starts. Those three moves take ten minutes and shift the style score more than an hour of surface polish.

One caution from the 2026 data: style scores still flag non-native speakers at higher rates, the same bias the probability scores had. If English is not your first language, do not treat a high style score as proof that you write badly. Use the checklist in this guide, then judge the draft on how it reads.

If your draft also carries the other AI tells, the guide to common AI writing patterns walks through the full set and how to break each one.

And when you want to compare which detectors are actually tracking style in 2026, our detector accuracy comparison has the numbers.

The bottom line has not changed, but the evidence got stronger: formal register is the loudest tell, and it is now measurable. Run the seven techniques, check the style score, and watch both improve.

The 60-second formality audit

Run this before you publish any AI draft. It takes one minute and catches the formal tells that make readers click away.

Step one: read your first paragraph out loud. If you would not say the sentence to a coworker, rewrite it in the way you would actually speak.

Step two: search for the expanded forms. Every it is and do not becomes it's and don't. Contracted verbs are the single fastest formality signal.

Step three: count the sentences that start with a transition word. If more than one in four does, delete the connector and let the idea stand alone.

Step four: check your last paragraph. A formal draft ends with a summary and a bow. A human draft ends with a point or a question.

The audit works on any draft type, but it pays off fastest in email, where the robotic tone guide covers the tells readers notice first.

If the audit flags your drafts, the how to make AI writing sound human guide takes the seven techniques further, and the robotic email tone guide shows where formality kills replies.

That is the whole audit. One minute, four checks, and the draft reads the way you actually talk.

The late August 2026 formality audit update

We ran the 60 second formality audit from this guide on a fresh batch of AI drafts in late August 2026. The formal tells are still the same four: opener fillers, hedged verbs, even sentence lengths, and a vocabulary that is one step too polite.

The good news is the audit still catches all of them in under a minute. The bad news is newer models hide the tells a little better, so you have to look at rhythm as well as word choice.

One late August finding worth stealing: read the draft aloud and mark every sentence longer than 25 words. In our test batch, those long sentences were where the formality hid.

The seven techniques in this guide all held up on the new models. Sentence shortening did the most work, exactly as it did in the original tests.

If you are still seeing press-release phrasing after the audit, our guide on common AI writing patterns breaks down why it happens and how to fix it at the source.

We also confirmed that the voice snapshot trick works better than a long style prompt. A paragraph of your real writing pulls the model toward your rhythm faster than ten instructions.

The other late August change: models now imitate casual tone more convincingly at the sentence level, so the formality shows up in structure, even pacing, and polite hedges, more than in word choice.

That means the audit needs one extra step: check sentence length variance. If every sentence is between 15 and 25 words, the draft is structurally formal even when it sounds relaxed.

And if you want the draft to sound like a person rather than a polished assistant, our guide on adding personality to AI writing shows where to put your own details and opinions.

Informal is not the same as sloppy

The fastest way to fail at this is to overcorrect. Strip too much formality and the draft stops sounding like a person and starts sounding like a text message. Casual and careless are different things, and readers separate them immediately.

What separates the two is precision, not vocabulary. An informal sentence still makes one clear claim. It just stops wrapping that claim in ceremony. "We cut the review step and shipped two days late anyway" is informal and precise. "So yeah we kinda just sent it" is informal and says nothing.

A quick way to check whether you crossed the line:

• Cover the contractions and read the sentence again. If the claim disappears, the informality was carrying the meaning instead of the sentence.
• Count hedges. In formal writing they arrive as "it may be advisable to." In casual writing they arrive as kinda, sort of and I guess. Both weaken the same claim.
• Check the greeting and the sign-off. That is where casual drafts get most informal, and where readers decide how seriously to take the rest.

The same rule carries into email, where the stakes are higher because the reader knows you. Colloquial phrasing lands well in a message to a colleague and badly in a message to a client who has never met you. If your draft is heading somewhere with a relationship attached, run it past a robotic email tone guide keyed to the situation before you hit send.

Worth saying plainly: the goal was never to make writing sound less intelligent. Formal writing is not the enemy, undifferentiated writing is. A page where every sentence carries the same weight and the same length is hard to read whatever register it lands in, which is why the fixes that make writing sound conversational overlap so heavily with the fixes that make it readable.

None of these checks ask you to lower your standards. They ask you to state them out loud instead of dressing them up.

Where this gets subtle is the middle. A sentence can be informal and still hedge, or formal and still be sharp. Register is not the same axis as clarity, and the audit below keeps the two separate.

Run the audit below on a draft you already consider finished. Most people find two or three sentences that were doing ceremony instead of work, and cutting them is usually the whole edit.

Frequently asked questions