I want to be honest about something upfront: I went into this experiment skeptical of my own idea.
The premise — handing my content creation entirely to AI tools for a full month — felt either brilliant or embarrassing depending on who I told. My friend who works in marketing said "that's the future." My partner said "you're going to lose followers." My gut said both of them were probably right.
So I ran the experiment anyway. Thirty days. Four platforms. Every post scripted, captioned, timed, and in some cases entirely generated by AI. Here's what actually happened.
The Setup
I started with a clear problem: I was posting inconsistently because content creation was taking too long. I'd spend 90 minutes on a single Instagram carousel and another 45 minutes on the caption. By the time I finished, I'd lost the will to promote it. The content was sitting at 200-400 views per post, which felt proportional to how much I hated making it.
My platforms: Instagram, LinkedIn, Twitter/X, and TikTok. My niche: productivity and AI tools for solo business owners.
The AI tools I used:
- Claude (Anthropic) for writing: all captions, hooks, email newsletters, and long-form posts
- ChatGPT-4o for ideation: generating topic angles and responding to trending questions in my niche
- Canva AI for visuals: the Magic Write feature for carousel text, background removal, and image generation for some graphics
- ElevenLabs for TikTok voiceovers on two test videos
My role: approving or rejecting content, making light edits, hitting publish, and responding to comments manually (I drew the line at automating replies — felt wrong).
What Worked
Speed. Genuine, absurd speed.
My content production time dropped from roughly 8-10 hours per week to about 2. Those 2 hours included reviewing drafts, making edits, and scheduling. That's it.
The quality difference between my AI-assisted captions and my handwritten ones was negligible after editing. Actually, some of the AI-assisted captions were better — specifically on LinkedIn, where I tend to write stiff and formal when I draft myself. Claude has a looser, more human tone by default. I kept it.
The LinkedIn posts performed better than anything I'd ever written manually.
This was the biggest surprise of the experiment. Over the 30-day period, my LinkedIn impressions increased by 340% compared to the prior month. Three posts hit over 5,000 impressions — more than my previous five months combined.
I think I know why. LinkedIn rewards consistency and frequency. I was posting 4-5 times per week (previously: 1-2 times, erratically). The AI made daily posting sustainable. I wasn't tired. I wasn't running out of things to say. I was reviewing and approving instead of creating from scratch, which is a fundamentally different cognitive task.
My newsletter open rate went up 8 points.
Newsletter is the channel I care most about. The open rate going from 31% to 39% over the month surprised me. I think it happened because the AI was better than I am at writing subject lines. I'd give it the email topic, it would generate 10 subject lines, I'd pick the best one, and it was almost always more clickable than what I would have written alone.
What Flopped
TikTok was a disaster.
I tried using ElevenLabs-generated voiceovers for two TikToks. Both flopped spectacularly — under 200 views each, in an account that averages 1,500-3,000.
I think TikTok audiences are more sensitive to authenticity than other platforms. Something about the synthetic voice was detectable even if you couldn't consciously identify it. The videos felt off in a way that was hard to articulate but clearly communicated itself in the metrics.
The videos I made myself — scripted by AI but with my own voice — performed normally. The lesson: AI can write your script. Your voice still has to deliver it.
Instagram Reels were mixed.
Some hit. Some didn't. I couldn't find a consistent pattern. The Reels with AI-generated text overlays (using Canva AI) performed worse than those without. The Reels I scripted using AI but filmed naturally performed comparably to my manual content.
My hypothesis: Instagram's Reels algorithm rewards the kind of natural hesitation, authentic camera presence, and imperfect energy that doesn't exist in AI-scripted, tightly edited content. There's something in the human messiness that the algorithm or the audience prefers.
I lost some comments.
My comment engagement on Instagram dropped during the 30 days, not dramatically but noticeably. My best posts typically get 20-40 genuine comments. During the AI month, most posts were getting 5-12.
The posts that got the most comments were the ones where I added a personal paragraph at the end that wasn't in the AI draft. The posts that got fewest comments were the ones I published mostly unedited.
This told me something important: my audience follows me partly for my perspective, not just for useful information. The AI can give them the information. Only I can give them me.
What Changed in My Process After the Experiment
I didn't go back to doing everything manually. But I also didn't stay at 100% AI.
Here's where I landed after the 30 days:
I kept AI for:
- First drafts of all LinkedIn posts (then personalized heavily)
- Newsletter subject lines (test 5, pick the best)
- Research: pulling together angles, counterpoints, statistics on a topic
- Carousel copy (AI draft, I write the visual prompts and headline myself)
- Repurposing: turning one blog post into 6 different social formats
I went back to manual for:
- TikTok and Reels scripts (the voice has to be mine from the start, not adapted)
- The opening paragraph of every LinkedIn post (first 2-3 lines determine everything)
- Replies to every comment (non-negotiable — relationships live in comments)
- Anything opinion-based or personal (AI has no opinions, so I have to bring mine)
The number that matters most:
Before the experiment: ~8 hours per week on content creation.
After building the AI-hybrid system: ~3.5 hours per week on content creation.
Content output: essentially unchanged.
LinkedIn performance: stayed elevated even after the experiment ended.
Honest Takeaway
If you've been thinking about using AI for social media, do it. The fear that your audience will somehow detect you're using AI assistance is, in most cases, unfounded. Most of what makes content good — clarity, specificity, relevance to the audience — is exactly what AI assists with.
The fear you should actually have is the opposite one: that you'll remove yourself so thoroughly that the content becomes technically good but emotionally empty. That happened in pockets of my experiment. The posts with no personal element got fewer comments, fewer saves, and fewer shares.
AI is a co-writer, not a replacement. The experiment proved that to me faster than a year of theorizing would have.
The 30 days aren't a flex. They're a proof of concept. I found where the line is between AI's strengths and mine, and now I know how to work on both sides of it.
I write about AI tools, productivity systems, and building solo businesses. If you found this useful, the clap button exists — and costs nothing.
Top comments (0)