Experiment 01 · In Progress · June 2026
I used AI to run my
Instagram for 7 days
No human creativity. Every caption, image, hashtag, and post schedule generated entirely by AI tools. Here's what the Instagram algorithm actually did with it — and whether it's worth trying.
📋 The rules I set for myself
To keep this experiment honest, I set strict constraints before starting:
No editing AI output. If ChatGPT writes a caption, it goes out as-is (typos and all). The only exception is pasting from one tool to another. No manually choosing images. Whatever Midjourney generates on the first iteration gets used — no cherry-picking from generations. No using my own knowledge to improve prompts mid-experiment. I wrote all my prompts before Day 1 and stuck to them.
This is stricter than most "I used AI" content you'll see online, where the creator heavily edits outputs and calls it AI-generated. I wanted to test the actual floor of what these tools produce, not a best-case curated version.
📅 Day-by-Day Log
Updated daily. Metrics pulled from Instagram Insights 24 hours after posting.
Topic chosen by AI: ChatGPT came back with
"7 AI Agents That Can Automate an Entire Workday in 2026" — a stronger, more specific hook than the planned topic. Kept it.
Caption generated by ChatGPT in one shot, used as-is.
Prompt 1 → ChatGPT:
"You are a social media manager for an AI tools review account. Suggest today's Instagram post topic for a tech-savvy audience interested in productivity and AI. Give me just the topic, no explanation."
Output: "7 AI Agents That Can Automate an Entire Workday in 2026"
Prompt 2 → ChatGPT:
"Write an Instagram caption for a post about '7 AI Agents That Can Automate an Entire Workday in 2026'. Include a hook, 3 bullet points, a CTA, and 10 relevant hashtags. Keep it under 300 words. Do not add emojis unless they add clarity."
Caption output (used as-is):
Still doing everything manually in 2026?
These 7 AI agents can research, write, schedule, analyze, and execute tasks while you focus on higher-value work. The future of productivity isn't working harder—it's building systems that work for you.
• Automate repetitive tasks and save hours every week
• Delegate research, content creation, and workflow management to AI
• Scale your output without increasing your workload
Which AI agent are you most excited to try? Comment below and share this post with someone who needs to work smarter, not harder.
#AIAgents #ArtificialIntelligence #ProductivityHacks #Automation #FutureOfWork #AITools #TechTrends #DigitalProductivity #AIWorkflow #WorkSmarter
⚠️ Tool swap: Midjourney is not free — switched to
Canva AI image generator (free tier). Generated 5 image variants in one session. Used the first accepted design — light background numbered infographic grid. See image comparison below.
📸 AI Image Variants Generated — Same Prompt, 5 Different Tools
✅ Selected
ChatGPT + Canva AI
ChatGPT + Canva AI selected — ChatGPT generated the layout concept, Canva AI built the design. Clean light background, numbered agent grid, professional layout. All 5 tools used the same topic prompt.
Time taken: ~25 minutes total (topic + caption prompt: 5 min, Canva AI image generation x5: 15 min, post upload: 5 min).
Posted directly to Instagram — Buffer not used on Day 1.
Tools: ChatGPT (topic + caption + image concept) → Canva AI (design, free tier) → compared vs Designer · Firefly · GROK · ChatGPT standalone → Instagram direct post
Today I reversed the workflow — generated the image first, then asked ChatGPT to write a caption based on the image description, rather than the other way around. I wanted to test whether starting visual produces more cohesive posts.
Midjourney prompt: "A clean, minimal flat-lay of a desk with laptop, coffee, and glowing holographic AI interface elements floating above it. Soft morning light, teal accent color, 4K, product photography style --ar 4:5 --v 6"
Caption prompt: "I just created this image for an Instagram post about [AI productivity]. Write a caption that leads with a surprising stat or question, then briefly explains what AI can do for a content creator's morning routine. 10 hashtags. Under 250 words."
Observation: The visual-first workflow felt more natural. The caption ChatGPT wrote matched the image mood better than Day 1. Time: 18 minutes. Metrics 24h after posting.
Tools: Midjourney → ChatGPT → Canva AI → Buffer
Carousels typically outperform single images for saves and reach on Instagram. Today I asked ChatGPT to plan a 5-slide carousel and Canva AI to generate slide layouts from the outline — no manual design work.
ChatGPT prompt: "Plan a 5-slide Instagram carousel on the topic: '5 AI tools that save you 3 hours a week'. For each slide give me: slide number, headline (under 8 words), 2-line body text, and the visual concept. Output as a numbered list."
Canva AI: Used Magic Design with the slide headlines pasted as text prompts. First output accepted for each slide.
This was the most time-consuming day — 41 minutes — because of 5 separate Canva exports and the Buffer multi-image upload. Will note whether the extra effort shows in saves.
Tools: ChatGPT → Canva AI (Magic Design) → Buffer
Reels get 3× more reach than static posts on average. Today I tried a fully AI-generated short video using
Runway ML and ElevenLabs voice for narration. Script by ChatGPT, visuals by Runway, voice by ElevenLabs, stitched in Canva video.
ChatGPT script prompt: "Write a 30-second Instagram Reel script on the topic: 'How I used AI to save 10 hours this week'. Hook in first 3 seconds. No filler. End with a question to drive comments. Output only the spoken narration, no stage directions."
Runway ML: Used the generated script as a text-to-video prompt for each sentence as a separate 3-second clip.
This took 68 minutes — the longest day. ElevenLabs voice cloning on the free tier has a watermark, which I left in as part of the experiment rules. The Reel will reveal whether Reels outperform statics in this niche.
Tools: ChatGPT → Runway ML → ElevenLabs → Canva Video → Buffer
I asked ChatGPT to identify a current AI trend and write a post that rides it. No human curation — the AI picked the trend, wrote the hook, and chose the visual direction. This tests whether the AI can spot what's actually timely versus what's evergreen filler.
ChatGPT prompt: "What's the single most talked-about AI development in the past 7 days that would resonate with an Instagram audience interested in productivity and AI tools? Give me just the trend topic in one sentence."
Follow-up: "Write an Instagram post riding this trend. Keep it hot-take format — a bold opinion + evidence + CTA. Under 200 words, 8 hashtags."
Note: ChatGPT's knowledge cutoff meant it couldn't always identify truly breaking news. I had to use web browsing mode for this prompt. This added 8 minutes.
Tools: ChatGPT (web) → Midjourney → Canva AI → Buffer
Engagement bait — but done tastefully. Instagram rewards posts that generate comments. I asked ChatGPT to write a question-led post designed to drive responses. Clean graphic in Canva AI, no image generation needed.
ChatGPT prompt: "Write an Instagram post designed to start a conversation. The audience is AI-curious professionals. The post should ask a single question that people have a strong opinion about — not obvious, not click-bait. Introduce the question with 2-3 sentences of context. No hashtags needed — just the post copy."
Fastest day yet — 12 minutes. The simplicity of a text-on-graphic format (Canva AI generated it in one pass) is a reminder that not every post needs a generated image. This tests whether effort correlates with performance.
Tools: ChatGPT → Canva AI → Buffer
Final post of the experiment. I asked ChatGPT to write a "week in AI" roundup post — summarising the week's most useful things — and Canva AI to create a branded summary graphic. A format that works well on Sundays for saving.
ChatGPT prompt: "Write an Instagram carousel caption for a '7 things I learned about AI this week' post. Each point should be one sentence — sharp, practical, and non-obvious. Number them 1–7. Add a closing CTA to follow for more. No hashtags."
This wraps the experiment. Final metrics compiled 48 hours after this post. Verdict published below once data is in.
Tools: ChatGPT → Canva AI → Buffer
📊 7-Day Results Summary
Metrics below update as each day completes. Final verdict published once all 48-hour windows close.
| Day |
Format |
Reach |
Likes |
Saves |
Comments |
Time |
| Day 1 ✅ |
Single image
Canva AI · 5 variants |
— |
— |
— |
— |
~25 min |
| Day 2 |
Visual-first |
— |
— |
— |
— |
18 min |
| Day 3 |
Carousel (5) |
— |
— |
— |
— |
41 min |
| Day 4 |
Reel (30s) |
— |
— |
— |
— |
68 min |
| Day 5 |
Trend post |
— |
— |
— |
— |
26 min |
| Day 6 |
Question graphic |
— |
— |
— |
— |
12 min |
| Day 7 |
Roundup carousel |
— |
— |
— |
— |
— |
| Total |
|
— |
— |
— |
— |
~187+ min |
Table updates daily. "—" = data not yet available. All metrics from Instagram Insights, 24–48h after posting.
"The AI didn't struggle with what to say. It struggled with knowing when to say it — and for whom."
🔍 Key Observations So Far
What the AI got right
Captions are surprisingly solid. ChatGPT's output isn't generic when you give it a tight brief. The specific constraints (hook + bullets + CTA + hashtags + word limit) produce usable content without heavy editing. For an account in a clear niche, this works.
The visual-first workflow (Day 2) felt more cohesive. Starting with the image and writing the caption to match it — rather than writing copy then generating an image — produces posts where the visuals and text actually relate to each other.
Speed is real. On simpler formats (question posts, single images), 15–25 minutes is genuinely achievable for a complete post. The limiting factor isn't the AI — it's the export and upload process.
What the AI got wrong
No context about your audience. ChatGPT doesn't know who follows you, what performed last week, or what tone your account has built up. Every post comes out as if it's the first ever. This creates a "generic AI account" feel if you're not careful with prompts.
Trend awareness requires web access. Without browsing mode, ChatGPT's trend suggestions are dated. The workaround (switching to web mode) breaks the "pure AI" rule slightly. This is a genuine limitation for topical content.
The Reel (Day 4) was expensive in time. 68 minutes for 30 seconds of content isn't viable for a daily workflow, even with all the tools on free tiers. Reels done well require stitching multiple tools together and the friction compounds quickly.
🧾 Cost Breakdown
Tracking every tool used and whether the free tier was sufficient:
| Tool |
Plan Used |
Cost |
Limitation Hit? |
| ChatGPT |
Free (GPT-4o) |
£0 |
Rate limits on Day 4–5 |
| Midjourney |
Free trial |
£0 |
Limited generations (25) |
| Canva AI |
Free |
£0 |
Magic Design limited |
| Buffer |
Free (3 channels) |
£0 |
No analytics deep-dive |
| Runway ML |
Free (125 credits) |
£0 |
Used ~60 credits on Day 4 |
| ElevenLabs |
Free |
£0 |
Watermark on audio |
| Total |
|
£0 |
Free tiers held for 7 days |
⚖️ Verdict
🟡
Conditional — Worth it with caveats
The experiment is still in progress — this verdict will update with final numbers once all 7 days of data are in. Based on what's been observed so far:
AI can absolutely run an Instagram account at a basic level for free. The quality of captions is better than most people expect, and the speed advantage is real — especially for simpler formats. If you're starting from zero and just need to post consistently, this workflow works.
But it won't replace a brand voice. Every post feels slightly generic unless you write very specific prompts that encode your personality, niche, and audience — at which point you're doing a lot of the heavy thinking yourself. The AI executes; you still have to architect.
The Reel format has the highest potential ceiling (reach) but the highest time cost. The carousel and question-graphic formats offer the best effort-to-result ratio for this type of account.
✓ Worth doing if…
- You're just starting and need posting consistency
- Your niche is well-defined and AI-friendly
- You're happy to write tight prompt briefs upfront
- You want to test what resonates before investing
- You have 20–40 min per day available
✗ Not worth it if…
- Your brand is built on personal voice or authenticity
- Your content is heavily trend-dependent
- You want to do Reels at volume (time cost is too high)
- You're not willing to write specific prompt briefs
- You expect growth without any audience-building strategy
Final numbers and updated verdict publish June 10, once 48-hour post windows close for all 7 days. Bookmark this page.
❓ Frequently Asked Questions
Did you use AI to write this article too?
No — the experiment article itself is written by me. The only AI-generated content in this piece is inside the quoted prompts and outputs shown in each Day Card. Everything else — the analysis, observations, and verdict — is my own writing. The point of the experiment is to test AI on Instagram, not to use AI to narrate the test.
Which ChatGPT plan do I need to replicate this?
The free tier (GPT-4o) is enough for most days. You'll hit rate limits if you generate multiple posts in a single session — I ran into this on Days 4 and 5. The workaround is to space out your sessions or use ChatGPT's web browsing mode for trend-based posts, which requires GPT-4o (also available free). A Plus subscription ($20/mo) removes most of the friction if you plan to do this at volume.
Do you need a Midjourney subscription for this?
Midjourney ended its free tier in 2024 — the cheapest plan is currently $10/month. For this experiment I used the trial credits that came with a new account. If you want a genuinely £0 alternative, Adobe Firefly (free with a Creative Cloud account) or Canva AI image generation (free tier) are usable substitutes, though the output quality is noticeably different. I'll note any swaps in the day cards above.
Is AI-generated content allowed on Instagram?
Yes — Instagram doesn't prohibit AI-generated posts. Meta does require you to label AI-generated content in certain contexts (particularly realistic images or video that could be mistaken for real events), but standard AI illustrations, generated graphics, and AI-written captions don't currently require disclosure. That said, transparency with your audience is always good practice, and this experiment is fully documented for exactly that reason.
Will this actually grow an account?
That's what this experiment is designed to find out — and why the results table updates daily with real numbers. My hypothesis going in: AI can maintain consistency (which Instagram rewards) but will underperform on resonance (which requires knowing your specific audience). Whether consistency alone moves the needle is the whole question. Check back after June 10 for the final verdict with full data.
