The Best Prompt to Humanize AI Writing

Last month a client forwarded me a LinkedIn post and asked one question: "Does this sound like AI to you?"

It did. I knew within four seconds. Not because of any detector tool, and not because the information was wrong. The information was fine. It was the music of the thing. Every sentence was the same length. Every idea came in threes. There was an em dash in almost every paragraph, and the post ended with "The future belongs to those who adapt." Nobody talks like that. Nobody has ever talked like that.

Here's the uncomfortable part: your readers have developed the same radar. Two years ago, AI-flavored text slid past most people. Today, the average LinkedIn scroller has read thousands of ChatGPT-generated posts, and their pattern recognition is sharp. When your email or your report intro trips that radar, the reader doesn't consciously think "this is AI." They think something worse: "this person didn't bother." The content might be 100% accurate and the reader still discounts it, because it feels mass-produced.

I use AI for almost everything I write. Drafts, outlines, rewrites, subject lines, the works. But nothing leaves my desk sounding like a model wrote it, and that's not because I rewrite everything by hand. It's because I run every draft through a specific humanizing prompt, and I've built that prompt into a repeatable workflow so I never have to think about it.

This post gives you the whole system:

1. The exact tells that make text read as AI-generated (with fixes)
2. Why models write this way in the first place, in plain English
3. The full master humanizer prompt, ready to copy
4. Three real before/after rewrites: a LinkedIn post, a cold email, a report intro
5. A lighter quick version for everyday use
6. How to make this permanent so you stop pasting prompts manually
7. When you should not humanize

No detector-dodging tricks, no "undetectable AI" nonsense. Just writing that sounds like a person, because that's what actually earns replies, comments, and trust.

The tells: how readers spot AI text in seconds

Before you can fix robotic writing, you need to know exactly what triggers the radar. These aren't vague vibes. They're specific, countable patterns, and once you see them you can't unsee them.

Here's my working list, collected from editing hundreds of AI drafts for my own site and for clients:

Tell	What it looks like	Why it screams AI	The fix
Em-dash abuse	"AI is powerful — but risky — and that's the point."	Models love em dashes as a universal connector; humans use them occasionally, models use them constantly	Max one em dash per ~150 words. Replace the rest with a period, a comma, or a rewrite
The banned-word cluster	"delve," "leverage," "landscape," "realm," "tapestry," "unlock," "harness," "elevate," "robust," "seamless," "game-changer"	These words are statistically overrepresented in model output; readers now flinch at them	Swap for the plain version: dig into, use, market, area, mix, get, use, improve, solid, smooth, useful
Rule of three, everywhere	"It's faster, cheaper, and smarter." "Plan, execute, and iterate."	Triads feel balanced, so models default to them in almost every list and sentence	Break the pattern on purpose: use two items, or four, or one blunt statement
Hedging and throat-clearing	"It's worth noting that…" "It's important to remember…" "While X, it's also true that Y"	Models are trained to be balanced and cautious, so they qualify everything	Delete the hedge. Say the thing. If a caveat matters, give it its own short sentence
Uniform sentence length	Every sentence lands between 15 and 22 words, forever	Models optimize for "safe" medium-length sentences; humans naturally vary	Mix it up. Follow a long sentence with a three-word one. On purpose
Hollow transitions	"Moreover," "Furthermore," "Additionally," "In today's fast-paced world"	Filler connectors that add zero information; a human editor would cut them instantly	Start the sentence with the actual point. Most transitions can just be deleted
The grand empty closer	"In conclusion, the journey is just beginning." "The future belongs to those who embrace change."	Models end essays like a high-school valedictorian; it's a summary reflex with no new content	End on a concrete detail, a specific next step, or a direct question
Symmetrical structure fetish	Every section has exactly three paragraphs; every paragraph has exactly three sentences	Models mirror structure because it's "safe"; real writing is lumpy	Let sections be uneven. One-line paragraphs are allowed. Encouraged, even
Adjective stacking	"a comprehensive, actionable, step-by-step guide"	Models pad nouns with two or three modifiers by default	One adjective, chosen well. Often zero
"Not just X, but Y"	"It's not just a tool, it's a mindset."	An extremely common model construction that sounds profound and says little	State Y directly, or cut the sentence entirely
Emoji-header syndrome	🚀 Every. 💡 Section. ✅ Gets. An emoji	Common in AI-generated LinkedIn content specifically	Zero or one emoji per post, if any. Never as bullet decoration
Perfect grammar, zero personality	No contractions, no fragments, no asides, nothing risked	Models write "correctly" by default; humans bend grammar for rhythm and emphasis	Use contractions. Allow fragments. Add one opinion the reader could disagree with

Print this table. Seriously. Half of humanizing is just knowing what to hunt for, and this table is the hunting list. The master prompt below encodes every row of it, but you'll also use the table directly when you're doing a final pass by eye.

One more meta-tell that doesn't fit in a table: AI text is frictionless. It never surprises you, never takes a small risk, never includes the oddly specific detail that only someone who was actually there would mention. Human writing has texture. "I tested this on a Tuesday because my Monday was destroyed by a deployment issue" is a sentence no model volunteers, and it's exactly the kind of sentence that makes readers trust the rest.

Why models write this way (plain English, no math)

You don't need machine learning theory to fix AI writing, but a rough mental model helps, because it tells you why the prompt below is written the way it is.

Three forces produce the robotic voice.

Force one: models predict the most likely next word. A language model is, at its core, a very sophisticated autocomplete. At every step it asks "given everything so far, what word most plausibly comes next?" and it leans toward high-probability choices. But here's the thing about human writing: the most likely next word is, by definition, the most average one. Real writers earn attention precisely by sometimes picking the less likely word, the sharper verb, the weird-but-perfect metaphor. A model left to its defaults regresses toward the statistical middle of everything it has read. That middle is smooth, competent, and completely without fingerprints. "Delve" and "landscape" are overrepresented for exactly this reason: they're the averaged residue of millions of blog posts, press releases, and consultant decks.

Force two: models were trained to please graders. After the initial training, models go through a phase where human reviewers rate their answers, and the model learns to produce answers people rate highly. Sounds good, right? Except raters, reviewing hundreds of responses quickly, reward things that look thorough and safe at a glance: balanced structure, hedged claims, complete sentences, tidy lists of three, a nice summarizing conclusion. Nobody ever got a low rating for writing "it's important to note that." So the model learned that inoffensive, symmetrical, over-qualified prose is the winning move. That's why AI text feels like it was written by a very diligent employee who is terrified of being wrong. In a sense, it was.

Force three: the model doesn't know you. Your writing voice is the accumulated residue of your specific life: the industry jargon you absorbed, the sentence rhythms of writers you loved, the phrases your first boss beat out of you, your tolerance for bluntness. The model has none of that. Given no voice specification, it falls back to its default, and its default is the averaged voice from force one plus the please-everyone caution from force two.

Put the three together and you get the signature AI sound: statistically average word choices, arranged in grader-pleasing symmetrical structures, in nobody's particular voice.

And this is the genuinely good news, because none of these forces are permanent. They're defaults, and defaults can be overridden with instructions. The model that writes "leverage the evolving landscape" will happily write like a blunt human being instead. You just have to tell it exactly how, with enough specificity that the instructions beat the defaults. Vague requests like "make it sound human" don't work, because "human" isn't specific enough to override anything. A list of banned words, hard rhythm rules, and concrete voice constraints does work, because now the high-probability choices are blocked and the model has to route around them.

That's what the master prompt is: a systematic override for every default.

The master humanizer prompt

Here it is. This is the full version I use for anything that matters: published posts, client-facing emails, reports that go up the chain. Paste your AI draft below it, or paste the prompt after a draft the model already produced in the same chat.

You are my editor. Rewrite the text below so it reads like a sharp, busy
human wrote it. Keep every fact, claim, number, name, and commitment
exactly intact. Change the music, not the meaning.

VOICE
- First person, plain English, confident and direct.
- Write like you're explaining this to a smart colleague over coffee,
  not presenting to a board.
- Contractions are required (I'm, don't, it's, that's).
- Take a position. If the draft hedges, pick the stronger reading and
  commit to it. Cut "arguably," "potentially," "it could be said."
- Keep any dry humor or personality already in the draft. Don't add jokes.

BANNED WORDS AND PHRASES (never use these, or close variants)
delve, leverage, landscape, realm, tapestry, unlock, unleash, harness,
elevate, empower, robust, seamless, streamline, supercharge, game-changer,
cutting-edge, revolutionize, navigate (metaphorical), journey (metaphorical),
"in today's fast-paced world," "it's worth noting," "it's important to
remember," "at the end of the day," "moreover," "furthermore,"
"additionally," "in conclusion," "dive into," "let's explore,"
"not just X, but Y," "the future of," "embrace change," "take it to
the next level," "look no further."

SENTENCE RHYTHM (this matters most)
- Vary sentence length hard. If one sentence runs long, the next one
  should be short. Some sentences should be under six words.
- Maximum one em dash per 150 words. Prefer periods and commas.
- Break the rule of three: when the draft lists three parallel items,
  make it two, or four, or restructure entirely. At most one triad
  in the whole piece.
- Sentence fragments are allowed for emphasis. Like this.
- No sentence may start with "Moreover," "Furthermore," "Additionally,"
  or "However," — restructure instead.
- Paragraphs must be uneven lengths. One-sentence paragraphs are good.

STRUCTURE
- Cut the throat-clearing intro. Start where the interesting part starts.
- Delete any sentence that only summarizes the previous sentence.
- End on something concrete: a specific detail, a next step, or a
  direct question. Never end on an inspirational abstraction.
- If a transition adds no information, delete it and let the sentences
  sit next to each other.

INTEGRITY RULES
- Keep facts, change music. Do not add claims, statistics, examples,
  or anecdotes that aren't in the original.
- Do not change the meaning of any hedge that is load-bearing (legal,
  medical, financial caveats stay).
- Keep roughly the same length, within 20% either way, unless cutting
  filler naturally shortens it.
- Match the format of the original (email stays an email, post stays
  a post).

Before you output the rewrite, silently check your own draft against the
banned list and the rhythm rules, and fix any violations. Output only
the final rewritten text. No preamble, no explanation of changes.

TEXT TO REWRITE:
[paste your draft here]

A few notes on why it's built this way, because understanding the design lets you adapt it.

"Keep facts, change music" is the spine of the whole prompt. The single biggest failure mode of humanizer prompts is that the model treats "rewrite" as permission to improvise. It invents a statistic, adds an anecdote you never lived, "improves" a commitment you made in an email. The integrity rules exist to make the boundary explicit: content is frozen, only delivery changes. I've found that stating this both at the top ("keep every fact...intact") and in a dedicated section is worth the repetition. Models follow instructions better when the important ones appear twice.

The banned list has to be literal. "Avoid corporate buzzwords" doesn't work; the model's idea of a buzzword differs from yours. A concrete list of thirty specific words and phrases works extremely well, because now those exact tokens are blocked. Add your own pet peeves over time. Mine grew from about ten entries to thirty as I kept catching new offenders.

The rhythm section does the heaviest lifting. Word choice is what people talk about, but sentence rhythm is what people feel. A draft can contain zero banned words and still read as pure AI if every sentence is eighteen words long. The rules "vary length hard," "one em dash per 150 words," and "at most one triad" attack the three strongest rhythm tells directly, with numbers the model can actually check.

The self-check line matters. "Before you output, silently check against the banned list" catches maybe a third of violations that would otherwise slip through. It's a cheap instruction with a real payoff. If you want to go further, my post on prompt engineering fundamentals covers self-review patterns in depth.

"Output only the final text" saves you a step. Without it, models love to append "I made the following changes..." which you then have to delete every single time.

Before and after: three real rewrites

Rules are abstract until you see them applied. Here are three formats I use AI for constantly, each shown as a typical raw AI draft and then the output after the master prompt. I wrote both versions of each pair, and the "before" versions are painfully faithful to what models actually produce when you prompt them naively.

Example 1: the LinkedIn post

Before (raw AI draft):

🚀 Excited to share some thoughts on AI in the workplace!

In today's fast-paced digital landscape, artificial intelligence is revolutionizing the way we work. It's not just a tool — it's a game-changer that empowers professionals to unlock new levels of productivity, creativity, and efficiency.

Here are three key takeaways from my journey:

✅ Embrace the technology — those who adapt will thrive ✅ Focus on upskilling — continuous learning is essential ✅ Stay human — AI augments us, it doesn't replace us

The future belongs to those who leverage these powerful tools while staying true to their authentic selves.

What's your take on AI in the workplace? Let me know in the comments! 👇

#AI #FutureOfWork #Innovation #Productivity

Count the tells: emoji headers, "fast-paced digital landscape," "not just a tool," "game-changer," "empowers," "unlock," a productivity-creativity-efficiency triad, three checkmark bullets, "leverage," "the future belongs to," and an engagement-bait closer. This post could have been generated by anyone about anything. That's the problem.

After (through the master prompt):

I spent six months telling my team to "use AI more."

Nothing happened.

Then I changed one thing: instead of encouraging AI use, I asked everyone to bring one workflow they'd automated to our Friday call. Just one. Show the prompt, show the output, show the time saved.

First week, two people showed up with something. By week eight it was the whole team, and our proposal turnaround dropped from four days to one.

The lesson wasn't about AI at all. People don't adopt tools because you tell them the tools are important. They adopt tools when there's a specific moment where they have to show their work.

If AI adoption is stalling on your team, stop selling the technology. Create the Friday call.

Same topic, completely different animal. It has a specific story with a specific mechanism, uneven paragraphs including two one-liners, zero banned words, and it ends on a concrete instruction rather than a comment-bait question. Note that the "after" works because the underlying draft had a real experience in it. The prompt can't invent your Friday call for you; give the model real raw material and it will arrange it honestly. If you post regularly, pair this prompt with a real content strategy so the raw material keeps coming.

Example 2: the cold email

Before (raw AI draft):

Subject: Unlocking New Opportunities for Meridian Consulting

Dear Ms. Chen,

I hope this email finds you well. My name is Alex, and I'm reaching out because I believe there may be a valuable opportunity for collaboration between our organizations.

In today's competitive landscape, companies like Meridian Consulting are constantly seeking innovative ways to streamline their content operations, enhance their digital presence, and drive meaningful engagement. Our comprehensive suite of AI-powered workflow solutions has empowered numerous organizations to achieve remarkable results.

I would love to schedule a brief 15-minute call at your earliest convenience to explore how we might leverage these capabilities to support your goals.

Looking forward to hearing from you!

Best regards, Alex

Every recruiter, founder, and manager has deleted five hundred of these. "I hope this email finds you well" plus "competitive landscape" plus a streamline-enhance-drive triad plus "leverage" in the ask. The subject line alone gets it archived.

After (through the master prompt):

Subject: your Q2 content backlog

Hi Ms. Chen,

I saw Meridian is hiring two content marketers, which usually means the backlog got bigger than the team.

I build AI writing workflows for consulting firms. The short version: your existing team ships roughly 3x the drafts, and the drafts don't read like ChatGPT wrote them. One firm about your size went from 4 published pieces a month to 14 without a new hire.

Worth a 15-minute look? If the backlog isn't actually a problem right now, tell me and I'll leave you alone.

Alex

Shorter, specific, and it opens with an observed fact about them instead of a claim about us. One number-backed proof point instead of "remarkable results." And the closer gives her a genuinely easy out, which paradoxically raises reply rates because it reads like a person who respects her time. Again: the prompt handled the voice, but the observation about the job postings had to come from me. Research is your job; music is the prompt's job.

Example 3: the report intro

Before (raw AI draft):

Executive Summary

In an era of rapid technological advancement, organizations must continuously evaluate their operational strategies to remain competitive. This report delves into the current state of our customer onboarding process, examining key metrics, identifying pain points, and proposing actionable recommendations. It is important to note that while significant progress has been made, there remain substantial opportunities for improvement. By leveraging data-driven insights, we can navigate the challenges ahead and unlock enhanced efficiency across the customer journey.

Ninety words, zero information. You could staple this paragraph to any report in any company in any industry. "Delves," "leveraging," "navigate," "unlock," a metrics-painpoints-recommendations triad, and the classic "it is important to note" hedge. An executive reads this and learns nothing except that the author owns a chatbot subscription.

After (through the master prompt):

Executive Summary

Our onboarding is losing customers at one specific step. 38% of new signups stall at data import, and stalled customers churn within 60 days at three times the base rate. Everything else in the funnel is healthy.

This report shows where the drop happens, why (the import tool fails silently on files over 10MB), and what it costs us (an estimated $340K in annual recurring revenue). We recommend two fixes: a file-size warning that ships in under a week, and a guided import flow for Q3. The first is cheap and recovers most of the loss.

Every sentence carries a number or a decision. The hedging is gone because the data supports a direct claim, and the "substantial opportunities for improvement" abstraction became "$340K in annual recurring revenue," which is the kind of phrase that gets a report actually read. One caveat, and it's important: those specifics existed in the body of the original report. The prompt's integrity rules pulled the intro toward concrete material that was already there. If your draft has no specifics anywhere, humanizing can't conjure them, and that's a research problem, not a prompt problem.

The quick version

The master prompt is what I use for anything with my name on it. But sometimes you're rewriting a two-line Slack message or an internal email, and a 400-word prompt is overkill. Here's the compressed version I keep as a text expansion snippet:

Rewrite this so it sounds like a busy human wrote it. Keep all facts and
commitments exactly as they are. Rules: use contractions; vary sentence
length (some under six words); max one em dash total; no "delve, leverage,
landscape, seamless, robust, unlock, empower, streamline, moreover,
furthermore, it's worth noting, in today's world"; kill any list of three
parallel items; cut the intro throat-clearing; end on something concrete.
Output only the rewrite.

[paste draft]

That's the 80/20. It hits the banned-word cluster, the rhythm rules, and the integrity constraint in about seventy words. For everyday messages it produces output nearly as good as the full prompt, because everyday messages are short enough that structure problems barely have room to appear.

My rule of thumb for choosing between them: if the text is under 150 words or will only be read once, use the quick version. If it's going to be published, sent to someone who matters, or read by more than a handful of people, use the master prompt. The full version's structural rules (uneven paragraphs, concrete endings, transition deletion) only pay off on longer pieces, but on longer pieces they pay off a lot.

Making it permanent: from prompt to workflow

Here's where most people stop, and it's exactly the wrong place to stop. A great prompt you have to remember to paste is a prompt you'll use for two weeks and then forget. The difference between people who get consistent value from AI and people who don't is almost never prompt quality. It's whether the prompt is wired into a system that runs without willpower. Repeatable beats brilliant.

There are three layers to wiring this in, and you want all three.

Layer 1: custom instructions (the always-on floor)

Every major AI tool lets you set standing instructions that apply to every conversation: Custom Instructions in ChatGPT, personal preferences and project instructions in Claude, a CLAUDE.md file if you work in Claude Code. Put a condensed version of the voice rules there:

When writing or rewriting any prose for me: plain English, contractions,
varied sentence length, max one em dash per 150 words, at most one list
of three per piece. Never use: delve, leverage, landscape, seamless,
robust, unlock, empower, elevate, streamline, moreover, furthermore,
"it's worth noting," "in today's," "not just X but Y." Start where the
interesting part starts. End on something concrete, never an
inspirational abstraction.

This raises the floor. Your first drafts come out 70% humanized before you've asked for anything, which means the dedicated humanizing pass has less to fix. Don't put the entire master prompt in custom instructions, though. Standing instructions compete with everything else in the conversation and long ones get diluted. Keep the always-on layer short and punchy, and save the full prompt for the dedicated pass.

Layer 2: a reusable skill (the dedicated pass)

For the full treatment, package the master prompt as a named, reusable asset instead of a note you paste around. In Claude this means a skill or project; in ChatGPT, a custom GPT or saved prompt; in Claude Code, a skill file you invoke with a slash command. The mechanics differ by tool, but the principle is identical: the prompt gets a name, and invoking the name runs the whole procedure.

This matters more than it sounds. Once the prompt is a named thing, three good side effects kick in. You stop maintaining seven slightly different copies in various notes apps. You can improve it in one place, and every future use gets the improvement, so when you catch a new AI-ism ("in the ever-evolving world of...") you add it to the banned list once and it's blocked forever. And you can hand it to other people: my humanizer went from a personal snippet to something my whole client team uses, purely because it became a shareable file instead of a paste-around. I've packaged my own as a downloadable skill at /skills/humanize-writing if you'd rather start from a working version, and the general pattern of turning prompts into named, reusable assets is the core of my Claude workflow template.

The workflow, end to end, looks like this:

1. Draft with AI as usual (or paste your own rough human draft, which works just as well)
2. Run the humanizer skill on the draft
3. Read the output aloud once and fix what trips your tongue
4. Ship

Step 3 is not optional and takes two minutes. Reading aloud is the single best AI-detector ever built, because rhythm problems that are invisible on the screen are unmissable in the mouth.

Layer 3: the editing checklist (the human backstop)

The final layer is a 60-second manual check before anything important goes out. Mine lives on a sticky note:

• Read it aloud. Stumble anywhere? Fix that sentence.
• Ctrl+F the top offenders. Search for "leverage," "delve," "seamless," and "—". Takes fifteen seconds.
• Check the first sentence. Does it say something, or does it warm up? Cut warmups.
• Check the last sentence. Concrete detail or next step? Good. Inspirational fog? Rewrite it.
• Count the triads. More than one list of three? Break one.
• Find one fingerprint. Is there at least one sentence only you could have written: a specific number, a real event, an opinion with teeth? If not, add one. This is the step no prompt can do for you.

Why keep a manual layer at all if the prompt is good? Because the prompt fails occasionally, maybe one draft in ten, usually on rhythm. And because the checklist trains you. After a month of running it, you'll catch tells in your own raw writing, in your team's drafts, in the wild. The checklist is how the skill transfers from the tool to your head.

If you produce content at volume, this whole stack slots into a bigger pipeline: research and outline, draft with a long-form SEO writing workflow, humanize, then pressure-test the argument with something like my grill-me skill before publishing. Each stage is a named, repeatable step. That's the whole philosophy of this site in one sentence.

When NOT to over-humanize

A tool this effective needs a warning label, so here's mine. There are real situations where cranking the humanizer to maximum is the wrong move.

Technical documentation and instructions. Docs should be boring. Uniform structure, predictable phrasing, and parallel construction are features when someone is scanning for step 7 at 2 a.m. during an outage. "Vary your sentence rhythm" actively harms a runbook. Humanize the introduction if you like; leave the procedure alone.

Legal, financial, and compliance text. Remember the integrity rule about load-bearing hedges? In these domains, almost every hedge is load-bearing. "This may constitute" and "this constitutes" are different legal claims. If a compliance officer or lawyer wrote the qualifier, the qualifier stays. Run these texts through nothing.

High-stakes messages where a wrong nuance costs you. Salary negotiations, incident apologies, anything HR-adjacent. The humanizer keeps facts, but tone is content in these messages, and you should own every word choice personally. Draft with AI if you want; do the final pass entirely by hand.

When "too human" becomes its own tell. Here's the subtle one. If every post you publish has a folksy anecdote, a strategically placed fragment, and a punchy one-line paragraph, that becomes a pattern too, and patterns get detected. I've started seeing "humanized AI" content that's recognizable precisely because it's so aggressively casual. The fix is the same as ever: variation. Some pieces should be more formal. Some emails should just be three efficient sentences with no personality at all, because that's what a real busy human sends most of the time.

When the writing is the point. A eulogy. A recommendation letter for someone who changed your career. The personal essay you've been meaning to write. Use AI to organize your thoughts if it helps, but write the words yourself. Not because AI can't produce something adequate, but because the reader is receiving effort as much as text, and in these cases outsourced effort is a broken promise.

The honest framing is this: the humanizer prompt removes a distraction. Robotic prose distracts readers from your ideas, and the prompt clears that away. But it cannot supply the ideas, the specifics, the lived detail, or the opinion. Text with nothing to say sounds hollow in any voice. Humanize the delivery, but bring the substance yourself. That division of labor, substance from you, music from the machine, is the entire trick, and it's why the people who get the most from AI writing tools are the ones who were already worth reading.

Start tonight. Take the last thing AI wrote for you, run it through the master prompt, and read both versions aloud. You'll hear the difference in the first paragraph.

Resources & links

• /skills/humanize-writing — the master humanizer prompt from this post, packaged as a downloadable skill you can install and invoke in one command.
• /blog/write-seo-articles-with-ai — my full workflow for drafting long-form SEO articles with AI, the step that comes right before humanizing.
• /blog/six-claude-prompts-ahead-of-99-percent — six Claude prompts that cover most professional writing situations, humanizer included.
• /blog/prompt-engineering-complete-guide-2026 — the fundamentals behind why prompts like this one work: constraints, self-checks, and instruction design.
• /blog/ai-marketing-team-in-claude — how I run an entire marketing content pipeline in Claude, with humanizing built in as a standing stage.
• /blog/claude-workflow-template — the template I use to turn any good prompt into a permanent, named workflow.
• /skills/seo-longform-writer — a skill that drafts 3,000+ word articles with structure and interlinking handled for you.
• /skills/ai-marketing-team — a set of coordinated agents for content, email, and social, all sharing one voice profile.
• /skills/context-engineering — how to give AI the background it needs so drafts start out sounding informed instead of generic.
• /skills/grill-me — a skill that stress-tests your draft's argument with hard questions before you publish.
• /skills/cover-letter-writer — tailored, sub-300-word cover letters that don't read like a template or a chatbot.

If this was useful, I send one practical AI workflow like this every week — no hype, no news roundups. Subscribe at alexandrrich.substack.com.