How to Write Faster With AI Without Sounding Like Everyone Else

How to write faster is the wrong question, and I asked it for roughly fifteen years before I understood why.

Here is the moment it became obvious. Two weeks after the layoff that ended 17 years at one company, I sat down to write a 150-word LinkedIn post. I stared at a blank screen for 45 minutes.

I had managed a team of 30 people. I had presented to C-suite executives without breaking a sweat. And I could not produce 150 words.

Writing speed is not typing speed. It is the total time between having a thought and having a publishable draft, and most of that time is spent not writing at all. It is spent stalling, researching things you already know, and rewriting your first paragraph nine times.

AI collapses that dead time, but only if you point it at the stalls instead of at the sentences.

That distinction is the whole post.

Why You Are Not Actually a Slow Writer

Direct answer: most people are not slow at writing, they are slow at starting and slow at deciding. I timed my own blog drafting across ten posts in 2026.

Roughly 70% of the elapsed time went to staring, searching and reorganising. Only about 30% went to producing sentences.

Speeding up the 30% barely moves the total. Removing most of the 70% cuts it in half.

Think about your last blog post. You probably knew what you wanted to say within the first four minutes, then spent the next hour and a half circling it.

That circling has two causes, and they are worth naming separately.

The blank page stall. You know the topic, you have no entry point, so you write a throat-clearing opening paragraph, hate it, delete it, and repeat. I have burned entire evenings on the first 80 words of a 2,000-word post.

The research stall. You pause mid-sentence to check a statistic, open a browser tab, and surface twenty minutes later having read four articles and written nothing. This one is sneaky because it feels like productive work.

Neither of these is a writing problem. Both are a preparation problem, which is very good news, because preparation is exactly what AI is good at.

How to Write Faster: The Four-Pass Method

Direct answer: the fastest reliable way to write a 2,500-word post is four separate passes with hard time limits. AI does the work in passes one and two. You draft in pass three, then strip the AI residue in pass four.

Total target time is 95 minutes, against the four to six hours the same post used to take me.

The passes are non-negotiable in order. The reason is simple. Every time you try to research and draft simultaneously, you lose the thread and pay a restart cost.

Here is the full method with the time targets I actually hold myself to.

Pass 1: Capture the Raw Thinking (10 minutes)

Before AI touches anything, you dump what you already know.

This is the step almost everyone skips, and skipping it is precisely why AI-assisted writing comes out generic. If you start with a prompt instead of your own material, the model has nothing of yours to work with. So it gives you the average of everything it has read.

I do this by talking, not typing. I speak for eight to ten minutes about the topic with no structure at all. That includes the half-formed opinions and the story I am not sure fits.

Why talking beats typing here: you think faster than you type, and self-editing is what kills the raw material. Spoken capture keeps the messy, specific, opinionated parts that make writing sound like a person.

I use VoiceLab for this. It turns that unstructured ramble into a structured set of notes I can hand to the drafting step. I wrote about the full capture-to-calendar workflow in the AI voice note system that fills my content calendar.

If you would rather type, type. The rule is that pass one contains zero AI and zero structure.

Pass 2: Build the Skeleton (15 minutes)

Now AI earns its place.

You hand over your raw capture plus the keyword and the audience, and you ask for structure only. Section headings, the order they should run in, what each section needs to prove, and which of your own points belongs where.

You are not asking for prose. You are asking for a load-bearing outline.

The specific instruction that matters: tell it to use only the points from your capture and flag gaps rather than inventing filler. Otherwise it will happily manufacture three sections of generic advice you never said and do not believe.

By the end of pass two you should have eight to twelve headings and a one-line note under each. If you cannot see the whole post from the outline, the outline is not done, and drafting now will cost you an hour later.

Pass 3: Draft Against the Skeleton (45 minutes)

This is where you write.

Not AI. You. Forty-five minutes, no research, no browser, no editing what you just wrote.

If you hit a gap where you need a statistic or a link, type a bracket like [CHECK] and keep moving. The single biggest speed leak in drafting is stopping to verify something, because a 30-second lookup reliably becomes a 20-minute detour.

Why you draft and not the AI. Sentence-level rhythm is where voice lives. The word choices, the fragments, the abrupt one-line paragraph, the interjection that would never survive a model's instinct toward smoothness. Hand that over and you get something readable and completely anonymous.

Does that mean AI should never write a sentence? Not really. It can draft the boring connective sections and the summary of something factual.

It should not draft your opinions, your stories, or your opening.

I covered where that line sits in what to automate and what to leave alone. The boundary has held up well over months of use.

Pass 4: The De-AI Edit (25 minutes)

Pass four is the one nobody talks about, and it is the reason most AI-assisted content reads identically.

Even when you write the sentences yourself, working alongside a model pulls your phrasing toward its register. You start writing "it is important to note" and "this allows you to" without noticing. The de-AI pass catches that drift before it reaches the reader.

I run a fixed checklist, in this order:

1. Delete every banned phrase. Mine includes game-changer, delve, seamless, robust, comprehensive, leverage as a verb, and roughly forty others. Search and destroy, no exceptions.
2. Kill the em dashes. Replace with a comma, colon, or full stop. Heavy em dash usage is currently one of the loudest tells.
3. Check paragraph length. Anything over three sentences gets split. Uniform four-sentence paragraphs are a machine signature.
4. Test every paragraph for one specific detail. A number, a name, a date, an outcome. Paragraphs with no specifics are usually the ones the model wrote.
5. Read the opening line alone. If it works as a setup sentence for any article on any topic, it is generic. Replace it with something only you could write.
6. Read the last line alone. It should be the strongest line in the piece, not a summary of the piece.

That is 25 minutes of unglamorous work. It is also the difference between content that sounds like you and content that sounds like the internet average.

Total: 95 minutes. Ten, fifteen, forty-five, twenty-five.

How to Write Faster Without Losing Your Voice

Direct answer: voice survives speed when the model is constrained by your material rather than asked to generate from nothing. The practical mechanism is a permanent context container holding your writing samples, your rules and your banned list. Every draft then starts from your constraints instead of the model's defaults.

I keep four documents that never change between sessions.

A profile of who I am and what I have done. A voice document with real samples of my writing. A rules document listing everything banned, and a FAQ covering the questions I get repeatedly.

Setting that up once took me about 45 minutes. It has saved hundreds of hours since. I walked through the exact setup in how to use Claude Projects to keep your content sounding like you.

Here is the thing about voice drift. It does not happen in one draft. It happens across thirty drafts, gradually, until you read something you published and cannot recognise yourself in it.

The fix is structural, not vigilance-based. You cannot out-discipline a default. You can only replace the default.

This is where the Content Creator's Claude Skill Stack does the heavy lifting for people who do not want to build the scaffolding themselves. It is 18 pre-built Claude skills covering the actual content workflow, including the anti-AI writing pass I run on everything I publish.

Everything is set up in plain English with no coding involved. Maybe you have read the posts about Claude skills, tried the documentation, and found it written for developers. That gap is exactly what this was built to close.

The skills are already written. That is the entire point of it.

What to Hand the AI and What to Keep

Direct answer: give AI everything that is retrieval, structure or mechanical checking. Keep everything that is judgement, opinion or personal experience. When people say AI writing sounds hollow, they have almost always crossed that line in one specific direction.

I keep this table taped to my process, figuratively speaking, because the temptation to hand over one more thing is constant.

The pattern is clearer once you see it laid out. Anything with a right answer goes to the machine. Anything with your answer stays with you.

Here is the part people get backwards. They hand over the opening and the opinions because those are the hardest parts, and keep the research because it feels like real work.

That is exactly inverted.

The research is the part that costs you 40 minutes and produces nothing anyone can tell you wrote. The opening is the part that costs you eight minutes and determines whether anybody reads the rest.

A quick test for any task you are unsure about. Ask whether two competent writers given the same brief would produce roughly the same output. If yes, hand it over. If no, that difference is your voice and it is the only thing on the page nobody can replicate.

I ran that test across my own pipeline in 2026. It found four steps I had been doing manually for no reason, and two I had automated that I should never have touched. Net result was about 90 minutes saved a week.

Batching Capture So the Blank Page Never Happens

Direct answer: separate capture from writing by several days. By the time you sit down to draft, the raw material already exists. This is the single change that cut my drafting time more than any tool did.

Capture is a five to ten minute task. Drafting is a 45 minute task. Doing them in the same session forces you to switch between two completely different mental modes, and the switch is where the hour goes.

So I split them across the week.

1. Monday to Thursday, capture only. Two or three short bursts, usually while walking, one topic each. No structure, no editing, no thinking about whether it is good.
2. Thursday evening, review the captures. Ten minutes. Which of these has enough substance to become a post, and which is really just a social post.
3. Friday morning, outline all three. Fifteen minutes each, using the captures as the only source material.
4. Friday, draft all three. Three 45-minute blocks with breaks between them.
5. Friday afternoon, edit pass on all three. The de-AI checklist runs faster in a batch because your eye is already tuned to the patterns.

Batching the edit pass alone saves roughly 20% of the edit time. You catch your own repeated tics across three posts instead of hunting for them one at a time.

One warning from experience. Do not let captures pile up beyond about ten days.

A voice note from three weeks ago has lost the context that made the idea interesting. You end up writing a competent post about a thought you no longer find compelling.

Captured this week, drafted this week. That is the whole rule.

The Mistake That Made Me Rewrite Eleven Posts

I want to name a failure of my own here, because I made it for about two months in early 2026.

I got fast. Too fast.

I was producing three posts a week and the pass-three time kept shrinking. I let the model draft more and more of the connective tissue, then the examples, then the opinions.

Then a reader emailed me. She had been reading since 2005 and said one line I still think about: "This doesn't sound like you anymore."

She was right. I went back through eleven posts and found the same tells in every one. Perfectly balanced paragraphs, no fragments, no interjections, every section the same length, and not one sentence that took a risk.

Why it happened: speed became the metric, and voice is invisible on a stopwatch. When you optimise for a number, everything that is not that number degrades without warning you.

How I fixed it: I put a hard floor on pass three. Forty-five minutes of human drafting is a minimum, not a target to beat. If a post takes fewer than 45 minutes of my own writing, I have delegated something I should not have.

That floor has held for months now, and nobody has emailed me that line again.

Typing Faster Versus Thinking Faster

Since we are talking about writing speed, it is worth separating two solutions that look similar and solve different problems.

Dictation solves the typing bottleneck. If you produce 45 words a minute typed and 130 spoken, dictation roughly triples raw output speed.

I tested this over 30 days with Wispr Flow and the speed gain was real. Edit time claws some of it back on technical terms.

Structured capture solves the thinking bottleneck. It does not make you produce words faster. It makes sure you never sit down to write without material. That is a bigger win, because the blank page costs more than slow typing ever did.

How to Write Faster Based on Where Your Time Goes

Choose the fix that matches your actual bottleneck, not the one that sounds most appealing.

• If you write for two hours and produce 1,500 words, your bottleneck is thinking, not typing. Fix pass one.
• If you write for two hours and produce 600 words while knowing exactly what you want to say, dictation will help immediately.
• If you write for two hours and produce a post you dislike, your bottleneck is the outline. Fix pass two.

Most people I talk to assume they have the second problem. Almost all of them have the first.

What This Actually Looks Like in a Week

Let me make it concrete, because a method without a schedule is just a theory.

I publish three blog posts a week on this site. Monday, Wednesday, Friday, at 8 AM IST.

The captures happen throughout the week in five to ten minute bursts, usually while walking. By Friday I have raw material for all three posts sitting in structured notes. Friday is drafting day, three posts, roughly five hours total including all four passes.

That is the same volume that used to consume my entire weekend. The slow way cost me four to six hours per post.

The gain did not come from writing faster. It came from never sitting down to a blank page, and never stopping mid-draft to look something up. If you want the SEO structure that sits underneath this, I broke that out separately in how to write SEO blog posts with AI.

Frequently Asked Questions

These are the questions people ask me most often about writing speed and AI, answered on their own terms.

How can I write faster without my writing sounding like AI?

Do the capture step before you touch any AI tool, then draft the sentences yourself and use AI only for structure and research. The generic sound comes from asking a model to generate from a prompt rather than from your own raw material. In my four-pass method, 45 of the 95 minutes are human drafting, and that ratio is what keeps the voice intact.

How long should it take to write a 2,000 word blog post?

With a system, roughly 90 to 120 minutes from raw idea to publishable draft. Without one, most people take four to six hours, which was my own number for years. The difference is almost entirely preparation, because unprepared drafting is where the hours disappear.

What is the fastest way to beat writer's block with AI?

Ask the model to interview you about the topic instead of asking it to write about the topic. Ten questions you have to answer produces material you actually believe, and answering is far easier than generating from nothing. This turns the blank page into a conversation, which is a problem your brain already knows how to solve.

Should I let AI write my first draft?

No, not the parts that carry your opinion, your stories, or your opening. AI first drafts feel fast but create a heavier edit burden, because rewriting someone else's phrasing takes longer than writing your own. Use it for outlines, research summaries, and factual connective sections.

Does dictation actually make you write faster?

Yes for raw output, with a caveat. Speaking runs around 130 words per minute against roughly 45 typed. Edit time recovers some of that gain, especially with technical terms and Indian English place names. Over 30 days of testing, the net gain was meaningful but nowhere near the 3x the raw numbers suggest.

How do I stop AI from changing my writing voice over time?

Build a permanent context container with real samples of your writing, an explicit banned phrase list, and rules about paragraph length and structure. Voice drift happens gradually across many drafts, not in one, so vigilance alone will not catch it. Setting that up once takes about 45 minutes and prevents the slow slide.

What are the biggest tells that content was written by AI?

Heavy em dash usage, uniformly four-sentence paragraphs, and phrases like "it is important to note". Add perfectly balanced section lengths and a complete absence of sentence fragments. Then flag any paragraph containing no number, name, date or specific outcome.

How many blog posts a week can one person realistically produce?

Three is sustainable for one person using a structured system and roughly five hours of drafting time. I publish Monday, Wednesday and Friday and have held that cadence without the weekend disappearing. Beyond three, quality tends to slip unless you are repurposing existing material rather than writing new.

Do I need to know how to code to use AI writing tools well?

No. Everything in this method runs through normal conversation with a model, plus a set of documents you write in plain English. The technical-sounding parts, like skills and projects, are configuration rather than programming.

What should I do if my draft is finished but reads flat?

Read only the first line and only the last line, in isolation. A flat draft almost always has a generic opening that could sit on any article and an ending that summarises instead of landing. Replacing those two lines fixes more perceived flatness than rewriting the middle ever does.

Is it worth writing an outline if I already know the topic?

Yes, and especially then. Knowing the topic is exactly when people skip the outline and end up reorganising for an hour after drafting. Fifteen minutes of structure reliably saves 45 minutes of restructuring.

How do I write faster when I only have an hour a day?

Split the passes across days rather than trying to compress them. Capture on Monday, outline on Tuesday, draft on Wednesday, edit on Thursday. The passes were designed to be separable precisely because most people building on the side of a job do not have a clean 95-minute block.

Back to the Blank Screen

That 45 minutes I lost to a 150-word LinkedIn post was not a writing failure.

It was a preparation failure. I sat down with no captured material and no structure. I expected words to arrive because they always used to when the topic was a quarterly review deck.

Writing faster is mostly about never being in that position again. Have the raw thinking already captured and the structure already decided. Reserve your energy for the only part a machine cannot do, which is sounding like yourself.

You are not slow. You are just starting from nothing.

What is the step in your own process where the time actually disappears?