Why AI-Written Blog Posts Sound Generic (and How to Fix It)

Generic AI writing has a recognisable shape: broad claims with no specifics, hedging where an opinion should be, transitions like "when it comes to" and "at the end of the day," and sentences that would fit any company in any industry. It reads that way because the model was handed a bare topic and left to write toward the average of everything it has read. Give it real research, a point of view, your own voice, and a structure built around the reader's question, and most of the flatness goes. Each of those is worth taking on its own.

You gave the AI almost nothing to work with

A blank instruction like "write a blog post about email onboarding" hands the model no facts of its own. It writes toward the average of everything it has read on the subject, and that average is bland by definition.

One of the simplest fixes is to start with a voice note. Open a dictation app (there are plenty, from free open-source ones to paid tools like Superwhisper and Wispr Flow) and talk through your thoughts on the topic without editing yourself.

Hand that transcript to the model along with the title as your prompt. It gives the draft your point of view and the thrust of your argument from the first line, so the piece comes from you.

Now even if the opinion is not wholly unique or original, it is yours. And that is usually enough to lift it out of the generic middle.

The analog for businesses is to include real data: your own numbers, call transcripts (after anonymising), a customer quote, a finding from a recent report, the questions your sales team fields every week. Given something concrete to build on, the model writes concretely.

It has no point of view

Left to itself, a model hedges. It sets out both sides, settles on "it depends," and finishes with a summary no one could argue with and no one will remember. Useful writing commits to something. Tell the model what you believe about the topic and ask it to make that case. "Argue that most onboarding emails go out too late" will beat "write about onboarding emails" every time.

Every model has a house style

Every model has verbal habits it reaches for when nothing stops it: the throat-clearing opener about a fast-changing industry, the neat list of three, the closing paragraph that restates what came above it. Read a few hundred AI drafts and the same skeleton shows through all of them. Telling it to "write casually" does almost nothing to shift this.

Rules the model can follow do more: short sentences, no exclamation marks, British or American spelling, address the reader as "you," never open on a rhetorical question.

I mentioned some of the classic LLM tropes to avoid in your writing in this tongue-in-cheek LinkedIn post titled How to stop sounding like ChatGPT's poor cousin on LinkedIn.

Having clear rules can help a lot, but what's even more powerful than rules is to provide specific examples of good and bad writing. "Show, don't tell" surprisingly does a lot to improve the model's output.

Abstractions instead of detail

Generic writing stays up at altitude, where everything is abstract. It reaches for "improve engagement" over "40% more replies," "various tools" over three named ones, "significant results" over the figure itself. Ask for the names, the numbers, and the examples in every section, and tell the model to cut any sentence that would still be true for a completely different company. Most of the gap between writing that sounds authoritative and writing that reads like a press release is concrete detail.

It builds from a template, not your reader's question

Ask for "a blog post" and you get the default blog shape: an introduction, a run of evenly sized headings, a listicle wherever there is an excuse for one, and a summary at the end. That gives you a page organised around the format instead of around the reader's question.

A stronger draft starts from the exact question someone typed. A reader searching "why are my onboarding emails getting ignored" should meet the most likely answer first and work down from there, with no warm-up paragraph defining what onboarding is.

What the fix looks like on the page

Say the topic is email onboarding. A plain prompt gives you something close to this:

Email onboarding is an important part of any successful business strategy. There are many factors to consider when setting up your sequence. By following best practices, you can improve engagement and drive better results for your company.

Put in some research, a clear opinion, and one real example first, and the same model produces something closer to this:

Most onboarding emails arrive too late to matter. One SaaS team found that users who received their first email within an hour of signing up were about twice as likely to come back the next day as those who waited for the standard "day one" send. If your sequence only starts the morning after signup, you are writing to people who have already forgotten why they joined.

Even though the model was identical in both cases, the material that went into the prompt beforehand made a noticeable difference.

The tricky part: doing this on every post

Each of these fixes is simple enough on its own. Doing all of them on every post starts to become a lot of work: pulling the research together, writing out the voice rules, supplying the specifics, shaping the structure, then reading the draft and sending it back when it slides into filler.

For a single article that is a good afternoon. Kept up at a few articles a week for a business blog, it becomes a full-time job.

This is exactly what Evatype can help you with. It reads your site to build a brand voice guide, runs live research so every draft is grounded in real facts and sources, and scores each piece against a quality bar before it revises and publishes straight to your CMS. You still own the prompts and the feedback rules, and the repetitive quality work runs on every article, not only the one you had the energy for.

If you have been fighting generic drafts by hand, try Evatype for free today.