ai vs. humans • the social media experiment
humans won. ai held the floor.
Everyone has an opinion on AI and what it means for how we work. We wanted data instead. So we ran a 60-day campaign testing AI-generated content against human-created content, head-to-head, to see which actually performed.
The verdict: human-created content excelled on every metric we tracked. AI cut production time, but that speed didn't translate into performance. AI content saw lower impressions and engagement than the human-made version.
157
TOTAL POSTS PUBLISHED
60
DAY CAMPAIGN WINDOW
72%
FINAL ENGAGEMENT RATE GAP
“The right strategy matches the job to the format. It doesn't pick one approach and stick with it everywhere."
+ Our Takeaway
Lean on AI for volume and a steady thought-leadership baseline: drafting, options at speed, holding the calendar when it's thin. Reserve the wins, the milestones, and the people for a human to write.
Human content led on every normalized measure.
All comparisons use averages per post rather than total volume, to account for the difference in post count between tracks (human content had 48% more posts at midpoint, and nearly double AI's count by the end).
+ Halfway Point • by june 24, 2026
Impressions gap: 101%
Engagements gap: 173%
LinkedIn engagement-rate gap: 81% — the most consistent split in the midpoint data.
+ final results • full campaign, june 1 - july 31
Impressions gap: 141% (Instagram alone: +133%)
Engagements gap: 311%
Engagement-rate gap: 72%, with LinkedIn again the most decisive split at +122%.
Human reach nearly doubled AI's at midpoint (798 vs. 397 impressions/post) and stretched further by the finish (668 vs. 277) — a 141% gap in the final numbers.
Human engagement efficiency widened over the campaign: 173% higher per post at midpoint, 311% higher by the final tally.
LinkedIn was the most decisive platform, with human engagement rate beating AI's by 81% at midpoint and 122% by the end.
AI found real footing in concise thought leadership. Its top post of the full campaign — "AI is useful. That doesn't mean it's strategic." — earned 42 engagements on 1,179 impressions.
The single highest-performing post of the entire experiment was a human-written LinkedIn PR piece: 135 engagements at midpoint, growing to 151 engagements on 353 impressions (a 42.8% engagement rate) by the end.
Two of the human track's highest-performing posts were culture and people content — a category AI cannot enter by nature. Removing those outliers narrows the gap considerably; the fairer head-to-head is thought leadership, where AI held its own on Instagram and LinkedIn.
+JUNE 1
Campaign launched. Human track required 52 hours vs. 8 hours for the AI track.
+JUNE 24
Midpoint established human lead. 25 AI posts and 39 human posts showed human content ahead on impressions, engagements, and engagement rate.
+MID-CAMPAIGN
AI track adapts. After seeing the midpoint gap, AI introduced new content formats, including long-form blog content, to try to close the deficit.
+JULY 31
Campaign closes. 55 AI posts and 102 human posts published in total. The performance gap held to the finish line.
Controlled intent. Real teams on both sides.
The experiment asked whether AI-generated content could perform comparably to human-created content, under identical goals, platforms, and audience.
Window: 60 days: June 1 through July 31, 2026.
Channels: Facebook, Instagram, and LinkedIn — the same platforms and audience for both tracks.
Guardrail: Same business goals for both teams, but the creative approach was completely up to each. Human content additionally covered culture and people posts, a category AI cannot produce by nature.
Measurement: Engagement rate, reach, impressions, saves, and shares, tracked across every content type. Per-post averages used throughout to account for the difference in post volume between tracks.
Strategy Shift: After midpoint numbers came in, the AI track adapted — introducing new formats, including long-form blog content, to try to close the gap.
Efficiency: Human track required 52 planned hours vs. approximately 57.7 completed — an 11% overage. AI ran its track on schedule without limits or fatigue.
