I write a lot of long posts, and for a while every draft Claude handed me had the same problem. The grammar was fine. The structure was fine. But it read like an instruction manual that had been to a corporate retreat. Every section opened with “It’s important to note.” Every list came in threes. Every conclusion started with “In conclusion.”
The fix wasn’t a better tool. It was a better prompt. Once I started telling Claude exactly how I talk and exactly which words to throw away, the drafts stopped sounding like a machine and started sounding like me.
Here are the three prompts I keep coming back to. You can use them as they are, swap in your own topic, and tweak the banned word lists to match whatever phrases you personally hate.
Prompt 1: The Voice-First Prompt
This one works because it teaches Claude your rhythm before it writes a single sentence. Most robotic writing comes from flat, even pacing where every sentence is the same length. This prompt kills that on the first pass.
I want you to write a [word count]-word blog post on [topic]. Before you write a single word, internalize how I write, because matching my voice matters more than covering every angle of the subject.
My style: short paragraphs, usually two to four sentences each. First-person voice throughout. I write like I’m talking to a reader who already knows the basics, so I skip the long throat-clearing introductions and get to the point. I use contractions everywhere. I vary sentence length on purpose so the writing has rhythm. Some sentences are short. Others stretch out when an idea genuinely needs the room to breathe and develop. Never let every sentence land at the same length, because that flat, even rhythm is the fastest way to sound mechanical.
Open with a real observation, a small moment, or a specific detail, never a dictionary definition or a sweeping statement about the world. End sections with a thought or a turn, not a neat little summary that repeats what I just said.
Words and phrases to avoid completely: explore, delve, dive into, navigate, unlock, unleash, harness, leverage, revolutionize, transform, elevate, empower, sophisticated, seamless, robust, cutting-edge, game-changer, paradigm, synergy, holistic, curated, bespoke, myriad, plethora, tapestry, realm, landscape, ecosystem, journey, embark, foster, facilitate, utilize, and any sentence that begins with “In today’s fast-paced world,” “In the ever-evolving,” “When it comes to,” or “Whether you’re a beginner or an expert.”
Also avoid these filler transitions: “It’s important to note that,” “It’s worth mentioning,” “Needless to say,” “At the end of the day,” “That being said,” and “Last but not least.” Write the post one section at a time, and stop after each section so I can react before you continue.
The part that matters most here is the line about sentence length. That single instruction does more heavy lifting than the entire banned word list, because uniform pacing is the deepest AI tell of all.
Prompt 2: The Write Like You Talk Prompt
The first prompt fixes rhythm. This one fixes tone. It pushes Claude toward how people actually speak, which means contractions, uneven lists, and the occasional sentence that starts with “But.”
Draft a long-form blog post of around [word count] words about [topic]. The whole goal here is to sound like a person, not a content mill. Imagine you’re explaining this to me across a kitchen table, not optimizing a page for Google.
Use contractions naturally. Ask the occasional rhetorical question, but don’t overdo it, because a question in every paragraph reads as a tic. Drop in personal asides in parentheses where they fit. Where you’d normally reach for a tidy list of three items, sometimes name just two and move on, because actual speech is uneven and three-item lists are the single biggest tell that a machine wrote something.
Sentence structure rules: never start two consecutive sentences with the same word or the same structure. Don’t make every paragraph the same length. Mix one-line paragraphs with longer ones. Begin some sentences with “And,” “But,” or “So,” the way people actually speak. Let a thought trail into the next paragraph instead of wrapping every point up cleanly.
Banned words and phrases, because they instantly signal AI writing: delve, explore, unlock, unleash, harness, leverage, navigate, seamless, robust, elevate, empower, transform, revolutionize, game-changer, cutting-edge, state-of-the-art, sophisticated, comprehensive, versatile, dynamic, innovative, streamline, optimize, supercharge, myriad, plethora, tapestry, symphony, realm, landscape, journey, embark, foster, facilitate, curate, and the whole family of “Imagine a world where,” “Picture this,” “In a world of,” and “Gone are the days.”
Kill these connective phrases on sight: “It’s important to note,” “It’s worth noting,” “Keep in mind that,” “Rest assured,” “Look no further,” “The bottom line is,” “When it comes to,” “In conclusion,” “To sum up,” and “All in all.” Skip emoji unless I ask. Write it section by section.
The “kitchen table” framing is the trick. The second Claude pictures a real listener instead of a search engine, the writing loosens up. The banned list then catches whatever slips through.
Prompt 3: The Anti-Pattern Prompt
The first two prompts shape the draft as it’s written. This one is a cleanup pass. You run it after Claude has produced something, and it forces a second read with one job: find the AI patterns and rip them out.
Write a [word count]-word blog post on [topic]. After you finish the draft, reread the whole thing through one filter: does any of this sound like AI wrote it? Then fix what does, before you show it to me.
Here’s what to strip out. Symmetrical structure, where every section has the same shape, the same intro-body-wrap rhythm, and lists that always come in groups of three. Hedging phrases like “can be a great way to,” “is a powerful tool for,” “plays a crucial role in,” “serves as a,” and “is key to.” Hollow conclusions that start with “In conclusion,” “To wrap up,” “In summary,” “All things considered,” or “At the end of the day.” Throat-clearing openers like “It’s important to note,” “It’s worth mentioning,” “Keep in mind,” and “Needless to say.”
Replace every generic claim with a specific one. If a sentence could appear in any article on any topic, like “This feature offers a range of benefits,” rewrite it so it could only appear in this exact post, with a real number, a named example, or a concrete detail. Vague intensifiers have to go too: “incredibly,” “truly,” “remarkably,” “undeniably,” “significantly,” “vast,” “wide range of,” “numerous,” and “a variety of.”
Cut these words entirely: delve, explore, unlock, unleash, harness, leverage, navigate, elevate, empower, transform, revolutionize, foster, facilitate, utilize, streamline, optimize, seamless, robust, sophisticated, comprehensive, versatile, innovative, cutting-edge, state-of-the-art, game-changer, paradigm shift, synergy, holistic, bespoke, curated, myriad, plethora, tapestry, realm, landscape, ecosystem, and any phrase shaped like “more than just a,” “not only X but also Y,” or “the perfect blend of.”
One more rule: after the rewrite, cut 15 percent of the total words. Tightening alone removes most of what makes writing feel padded and robotic. Show me the post in sections, not all at once.
That last line, the 15 percent cut, is the one I never skip. Most AI padding hides in qualifiers and warm-up phrases, and forcing a word reduction squeezes them out without you having to hunt for each one.
If you want the biggest single upgrade, paste 300 to 500 words of your own published writing into the prompt and tell Claude to match that cadence. The banned word lists handle the obvious tells, but a real sample teaches Claude your actual rhythm, which is the part no list can capture.