The problem with AI-generated content isn't that it shows. It's why it shows: because it doesn't say anything only you could say. A correct, well-written text, completely interchangeable with any competitor's.
AI does solve the real bottleneck of social media, which is volume. What it doesn't solve — and won't solve — is having something to say.
How can you tell a text was written by AI?
If your content has three or more of these, your audience has already noticed:
- Filler opener. "In today's world, more and more businesses…" Nobody talks like that. Start on the third sentence.
- Suspicious symmetry. Three points, each two lines long, all the same length. Reality isn't that tidy.
- A close that doesn't close. "In conclusion, digital transformation is key to success." That's not a close, it's filling space.
- Zero friction. Everything's an upside. Never a cost, never a "this isn't for you if…"
- Adjectives with no numbers. "Significant," "robust," "innovative." If you can swap the adjective for a data point, do it.
- Stock imagery. People smiling in offices that don't exist, hands with extra fingers, warped text on the signs in the background.
What should you delegate to AI and what not?
The raw material comes from your business
The core mistake is asking AI to invent content. What works is using it to recycle what already exists and today goes to waste:
- The questions people ask you every day. Every repeated customer question is a post you know someone cares about.
- A finished job. Before and after, what got complicated, how long it took.
- A call with a customer. Record it with permission, transcribe it, and pull three real topics out of it.
- A mistake you made. It's the content that gets shared the most and almost nobody publishes it.
- Your own numbers. How many projects, how long they take, what people ask for most.
One hour recording what you already know produces more material than a week asking a model for ideas. AI is an excellent amplifier and a terrible source.
A weekly workflow
This is enough to keep an account active without anyone spending their whole week on social media:
| When | Who | What | Time |
|---|---|---|---|
| Monday | Human | Pick 2 topics from what happened during the week | 20 min |
| Monday | AI | Drafts, headline variants, and format adaptation | 15 min |
| Tuesday | Human | Rewrite so it sounds like you and add the real data point | 40 min |
| Tuesday | AI | Graphics on the brand template | 20 min |
| Wednesday | Human | Final review and scheduling | 20 min |
Under two hours a week. The key is that the human comes in at the start (choosing) and at the end (approving), not in the middle producing.
How much does it cost to produce content with AI?
Three costs worth separating:
- The tools. Between $400 and $1,500 a month for a generative AI subscription for text and images. It's the cheap part.
- The visual line. This is a one-time cost: templates, palette, typography, and rules. Without it, every piece looks like it's from a different family, no matter how much AI you use.
- The human time. The two weekly hours above. If you do them yourself, they cost your hourly rate; if an agency does them, they're built into the retainer.
Three expensive mistakes
- Publishing without reading. Models invent facts with total confidence. A false number in your own post is your credibility, not the model's.
- Automating replies to comments. Produce content at scale if you want, but when someone comments, a person answers. That's where sales close.
- Measuring likes. Reach doesn't pay the bills. Measure how many conversations and how many customers came out of it — it's what we do from the Marketing service.
If you want to hand off the whole process, that's exactly what our Design service does: the visual line gets defined once and monthly production runs on AI plus human review. And if you'd rather try it small first, the AI Buffet includes the social content package at a fixed price.
