How AI Writes Blog Posts That Sound Like You, Not a Robot

Give the AI Your Voice Before You Give It a Topic

Most people skip this step entirely, and it's the one that matters most.

Before you type a single topic into a prompt, feed the model something to work with. Paste in a recent email you wrote, a paragraph from a proposal, even a WhatsApp message where you were explaining something you know well. Tell it to study how you phrase things, what you tend to leave out, whether you write in long winding sentences or short declarative ones. The more raw, unpolished material you give it, the better. A cleaned-up, formal writing sample will push the output back toward bland assistant-speak, because the AI has nothing to distinguish your voice from a thousand other professional writers. The rough stuff, the way you'd explain something to a client on the phone, is where your register lives. When you're working out how to frame your own words for the web, that same instinct applies here too.

Think of it as calibration before content. A model that has read three paragraphs of how you write will produce a first draft that sounds like a rough version of you, not a polished version of nobody. That draft still needs editing, but you're correcting tone and detail rather than tearing out generic filler from the ground up. The difference in time is significant.

Start With a Brief, Not a Blank Prompt

The most common mistake people make with AI writing tools is treating them like a search engine. They type one sentence, something like "write a blog post about social media marketing", and then wonder why what comes back sounds like it could belong to any of a thousand other websites. That output isn't wrong, exactly. It just has no point of view, no specific audience in mind, and nothing to say that hasn't already been said. The AI fills the gap with generics because that's all you gave it to work with. Feed it a structured brief instead, and the whole character of the response shifts.

A brief doesn't need to be long. Three things make the biggest difference, who you're writing for, what position you actually hold on the topic, and one concrete example or claim you want the piece to anchor around. That last one matters more than most people expect. Once you give the AI something specific to build from, a real number, a named frustration your customers voice, a scenario you've seen play out, it stops producing average and starts producing particular.

If you're working on your site's copy more broadly, the same thinking applies. There's a useful overlap with writing your own website copy without overthinking it, where a clear point of view does more work than polished sentences alone.

Tell It What to Avoid

Most prompts tell the AI what to produce. The better move is telling it what to stop doing first.

Left to its defaults, the model will reach for the same verbal scaffolding every time: "delve into", "it's worth noting", "in today's fast-paced world", and the ever-present em-dash connecting thoughts that would be cleaner as two sentences. These aren't random choices. They're the statistical centre of gravity for text trained on the whole internet. If you want output that sounds like a specific person, you have to push it off that centre deliberately. Build a short banned list into every prompt you use and keep it there permanently. Something like, no em-dashes, no colons in body prose, no "leverage" or "robust" or "seamlessly", no opening a section with a question, no wrapping up with "in conclusion". The specifics can vary, but having a list at all makes a real difference.

The same logic applies to structure. If you notice the AI keeps opening with a contrast ("Some people do X, others do Y, the truth is somewhere in between"), add that to the banned list too.

Banned lists cover rhythm as well as vocabulary

One pattern that trips people up is thinking the banned list only needs to cover individual words. It also needs to cover rhythm and shape. Forbid the three-bullet-point summary at the end of every section. Forbid the "not only... but also" construction. Forbid consecutive paragraphs that are all the same length. If you're working on blog copy specifically, our post on writing copy without overthinking it covers how to build that kind of structural instinct from scratch. The banned list grows over time as you spot new tics, and each addition makes the next draft a little less generic.

Keep Your Opinions in the Draft

AI hedges by default. Ask it to write about whether businesses should invest in long-form content and it will give you a balanced, measured, neither-here-nor-there answer that offends no one and persuades no one. That is not writing. That is committee output.

The fix is to go back into the draft and add the thing AI consistently removes, a clear verdict. Not "there are several factors to consider" but "short posts rarely rank for anything competitive, and if you are writing 400 words hoping for organic traffic, you are probably wasting your time." That kind of bluntness is what a reader remembers. It also signals that a person with real experience wrote the piece, not a tool trying to avoid upsetting anyone. If a draft reads polished but bland, the issue is almost always that your actual opinion never made it in. Scroll through and ask, at each point, what would I tell a friend? Then write that sentence.

The same applies to disagreements and downsides. If a tactic has a real catch, say so plainly and say it where it matters in the argument, not tucked away at the end. A post on common SEO myths lands harder when it names which popular advice is actively wrong, not when it gently notes there are "differing schools of thought."

Push past the hedge. That is where your voice lives.

Edit for Rhythm, Not Just Accuracy

Read the draft aloud. If you get three sentences in and they all land with the same beat, the AI wrote it.

Uniform sentence length is the clearest tell. A typical AI paragraph runs every line to roughly the same word count, which creates a flat, metronomic quality that no human writer produces. The fix is surgical, not wholesale. Find a long sentence and break it in two. Then take two short ones and run them together with a comma and a connector. That alone shifts the rhythm enough that the paragraph stops feeling machine-made.

A useful test is to check whether your ear expects the next sentence before it arrives. If you can predict the shape of it, the writing has no tension left, and tension is what keeps someone reading past the first screen.

Accuracy edits catch wrong facts. Rhythm edits catch wrong feeling. Both matter, but most people only do the first one. If you want to go deeper on the actual words before you reach the editing stage, the guidance on writing your own copy without overthinking it covers how to feed the AI better raw material, which means less fixing at this stage. Fewer flat paragraphs to rescue in the first place.

Build a Repeatable System

The single thing that separates people who get consistent results from AI and those who spend an hour fixing every draft is whether they've saved their inputs somewhere reusable.

Your voice brief, your banned-word list, your preferred prompt structure, these aren't things you should be reconstructing from memory each time you sit down to write. Save them in a shared doc, a Notion page, anywhere you'll actually open it, and paste them at the start of every session. Over time you'll notice the drafts arriving closer to publishable, not because the AI got smarter, but because you stopped giving it vague instructions.

Treat it like a brief you'd hand a freelancer

The more specific you are about tone, length, examples you like, and phrases you hate, the less back-and-forth you end up doing. A banned-word list alone, even just fifteen or twenty terms that don't sound like you, will cut your editing time considerably on each post.

Where this pays off most clearly is inside a WordPress workflow. When your site already has an SEO layer baked into the publishing process, you're not bolting on optimisation as an afterthought. The prompt structure handles voice, the SEO layer handles structure and signals, and the whole thing becomes a pipeline rather than a series of one-off scrambles. Set it up properly once and you'll feel the compound effect within a few posts.