Home · Insights · How to write social posts that sound like you

How to write social posts that sound like you, not a chatbot.

3 MIN READ · 568 words

By Spencer Kunzlandgren · June 4, 2026 · 11 min read

Every small business owner has the same problem with AI-generated social posts: the output is fine, but it does not sound like them. It sounds like a marketing intern who has never met the business. Customers can tell. Engagement drops. Brand voice flattens.

The fix is not a better prompt. The fix is changing the input the AI is reading from. Here is the working framework, built from running content for restaurants, dental practices, HVAC contractors, and solo realtors over the past 18 months.

Step 1. Build a voice corpus before you build a single post.

A voice corpus is a set of writing samples that are unambiguously yours. The AI reads from this set to understand sentence rhythm, word choice, what you do and do not say, and how you handle a tough topic. Without it, the AI writes from the average of every business that has ever written about your industry, and the output reverts to that average.

What goes in the corpus:

Five service-related emails you have written to customers in the past 60 days.
Three social posts you wrote yourself and were proud of.
One transcript of you describing your business out loud (a recorded voice memo works).
Three written reviews you have replied to in your real voice.
One paragraph describing what makes you different from your competitor down the street.

That is enough to anchor a voice model. Less than 30 minutes of work, one time.

Step 2. The 80/15/5 frame.

Every shipped post should hit three things in this rough ratio:

80 percent real pain or news. Specific to your industry, your city, your guests, your patients, your clients. Named, dated, sourced if possible.
15 percent analytical takeaway. Make a claim and defend it. State an opinion. Be willing to be wrong.
5 percent subtle solution mention. Never lead with the brand. Never use the word "solution". Never close more aggressively than "if this resonates, send it to someone who needs it".

This is the structure most professional editorial publications use. It works because the reader gets value before being sold to. It also works because it forces the writer to actually know something about the industry, which is the part that compounds.

Step 3. Specific rules that fix the chatbot problem fast.

No em-dashes, no en-dashes, no slashes as stylistic separators.

The em-dash is the single biggest tell that a post was written by AI in 2026. Real people use periods.

Name a specific person, place, or number in the first sentence.

If the post could be about any business in your industry, it is not your post. Anchor it to something specific within the first ten words.

Generic — could be any restaurant Quality matters. Our team is committed to bringing you the best experience every single visit, made with love and locally sourced ingredients.

Specific — could only be one restaurant Marie spent four hours yesterday morning trimming the brisket for tonight. By 6pm it will be gone. The math on slow-and-low is that you cannot rush a recovery. The same math applies to the kitchen team. We are hiring a second pitmaster this month.

End with a question, a stat, or an observation. Never a generic CTA.

Banned closes: "DM me", "reach out", "let me know", "happy to connect", "limited time", "book now". Real operators close with something interesting, not a directive.

Never invent a number.

If you cite a statistic, it must be defensible. If it is not, rewrite the sentence. AI tools will fabricate numbers happily. You will not survive the first customer who fact-checks you.

Step 4. The closed loop.

Every post you ship gets measured 24 hours later. The ones that performed get more of that hook formula, that post structure, that topic. The ones that did not, get retired. Three losses on the same template, the template is gone.

This is the single biggest difference between operators who compound and operators who stay flat. Most small businesses post and forget. The ones who track their own performance and adapt within 24 hours pull ahead within a quarter.

The voice corpus is the input. The 80/15/5 frame is the structure. The closed loop is the operating system. All three are required. Pick one and the output reverts to the median.

Step 5. What to do when the AI output still sounds generic.

If you are following all four steps above and the output still sounds like a chatbot, the problem is almost always one of two things:

Your voice corpus is too small or too professional. Add three drafts you wrote at 11pm when you were tired and honest. Those are the ones the AI needs.
You are over-editing. If you rewrite more than 30 percent of every draft, the voice model never gets the feedback it needs. Approve the ones that are close-enough, decline the ones that are not, and let the model learn.

What we built.

One Voice was built around this exact framework. The voice corpus is the first thing the brain reads. The 80/15/5 frame is enforced by the voice guard. The closed loop runs every 24 hours. The four-role bench writes for every channel without losing your voice across any of them. More on the product here.

See your visibility score first.

60 seconds. No credit card. We map your real presence on Instagram, Facebook, LinkedIn, Google, and AI search.

See your visibility score →

How to write social posts that sound like you, not a chatbot.

Step 1. Build a voice corpus before you build a single post.

Step 2. The 80/15/5 frame.

Step 3. Specific rules that fix the chatbot problem fast.

No em-dashes, no en-dashes, no slashes as stylistic separators.

Name a specific person, place, or number in the first sentence.

End with a question, a stat, or an observation. Never a generic CTA.

Never invent a number.

Step 4. The closed loop.

Step 5. What to do when the AI output still sounds generic.

What we built.

See your visibility score first.

What to read next.