I have a spreadsheet with 1,789 rows. Every post I made between February 2025 and February 2026. Every impression, every like, every bookmark, every follow, every unfollow. 365 days of account-level data across 101 active growth days.

I didn't build this spreadsheet because I'm obsessive (though I am). I built it because I kept reading advice that felt wrong. "Post more." "Tutorials build authority." "Threads are the growth format." Every creator in the AI art space was repeating the same playbook.

My data says the playbook is backwards. Almost all of it.

Here's what 1,789 posts actually taught me about what converts followers on X when you're an AI creator.

Of my 1,789 posts, 962 were replies. 827 were original content.

That means 53.8% of my total output was conversations with other people. Not broadcasts. Not content. Conversations.

Here's what that split produced:

Replies had the higher engagement rate (15.2% vs 10.1%). If you stopped reading there, you'd think replies were the better content type.

But replies generated just 14 follows across 962 posts. Original content generated 155 follows from 827 posts. That's an 13x difference in follow conversion per post. For bookmarks, original content produced 26x more per post than replies.

So why did I spend 53.8% of my time on replies?

Because replies are infrastructure. They're the daily community engagement that keeps the algorithm serving your original content to the right people. They're the 50+ conversations per day that make your account visible within your niche. The replies don't convert directly. They create the conditions for everything else to convert.

If you're an AI creator posting images without engaging in the community around them, your images are landing in an empty room. The replies fill the room. The original content works the room.

Here's the content category breakdown. Every post manually categorized. Read it slowly because it contradicts almost every piece of growth advice you've heard.

I need you to sit with that top line for a second. Shoutouts converted at 1.08 follows per post. The next best category was Stor-AI at 0.35. Shoutouts converted 3x better than creative portfolio content and 54x better than tutorials.

54x.

If you are an AI creator and you are spending your time writing tutorials about how to use Firefly or how to structure prompts, and you are not regularly shouting out other creators in your space, you are choosing the least efficient growth strategy available to you.

It makes no sense on the surface. A shoutout is a post saying "Go follow this person, they make great work." Average impressions: 331. That's tiny. No viral potential. No bookmark-worthy content. No algorithmic rocket fuel.

But 1.08 follows per post means that for every shoutout I gave, someone followed me in return. Not the person I shouted out (though they often did too). Other people. People who saw the shoutout and thought: "This person curates. This person knows who's good. I should follow them so I find more creators like this."

For AI creators specifically, this hits different. The AI art community on X is built on mutual discovery. New tools launch constantly. New techniques spread through the community. Nobody can keep up alone. A creator who regularly surfaces other good creators becomes a node in the network. You become the person people follow to find the rest of the community.

25 shoutouts generated 27 follows. My 60 tutorial posts generated 1 follow. I could have written 60 shoutouts in the time it took to write those tutorials and generated an estimated 65 follows instead of 1.

My Stor-AI Time series (AI-adapted folktales as storybook videos) had the highest average impressions of any category at 1,112 and the highest average bookmarks at 4.0 per post. It was my best content by almost any quality metric.

But it only converted at 0.35 follows per post.

This is the portfolio paradox that every AI creator needs to understand. Your best creative work builds your reputation, earns bookmarks, and proves you can produce at a high level. But it doesn't directly drive follows at the rate you'd expect.

Stor-AI was the content that got me the Adobe Firefly Ambassador program. 462 bookmarks on the workflow article. 60 bookmarks on the Mighty Monster Afang launch. This is the content that built credibility. But the shoutouts built the audience.

If you're only posting creative output and wondering why your follower count isn't matching your impression count, this is the reason. Portfolio content and community content serve different functions. You need both.

267 threads. 0.04 follows per post. That's 10 total follows from 267 attempts.

I say this as someone who genuinely believes in thread content. Threads are excellent for education, for SEO-style discovery on X, for establishing expertise. But they don't convert follows efficiently. At least not for me as an AI creator.

Why? Because a thread is complete within itself. You read it, you learn something, you move on. Like a beautiful image, a good thread answers its own question. There's no open loop driving you to the profile.

Compare that to a shoutout (1.08 follows/post) or a luthier/personal story (0.30 follows/post). The shoutout creates reciprocal community energy. The personal story creates human curiosity. Both leave the reader wanting to know more about the person behind the content.

Threads build bookmarks (1.2 average, third highest). They build authority. They build portfolio. But if raw follower conversion is what you're measuring, threads are 27x less efficient than shoutouts.

I calculated the statistical correlation between every daily metric and net follower growth across 101 active days. Here's the ranking:

The bottom line is the one that matters most. Post volume barely correlates with growth at all (r=0.093). You don't grow by posting more. You grow by generating impressions, bookmarks, and engagements, all of which are quality signals, not quantity signals.

For AI creators, this is liberating. You don't need to post 10 images a day. You need to post content that generates bookmarks and impressions. One documented workflow article earned more bookmarks than 267 threads combined. One well-crafted Stor-AI launch outperformed 60 tutorials.

Quality beats volume. My data is unambiguous about this.

Two posts out of 1,789 crossed 10,000 impressions. Those 2 posts (0.1% of my content) generated:

9 posts (0.5% of content) accounted for nearly half of all bookmarks. The other 1,780 posts split the rest.

This is the power law and it is merciless. Most of what you post will not move the needle. A tiny fraction of your output will drive most of your results.

For AI creators, the takeaway isn't "only try to make viral posts." You can't predict which post will break through. What you can do is stack the odds. Documentation and creative portfolio content dominate the top of every performance list. My top 10 bookmarked posts were all either process documentation or Stor-AI launches. Not a single thread. Not a single tutorial. Not a single tool showcase.

Give the power law the right inputs and it will eventually produce the right outputs.

This one surprised me. Lower-impression posts had higher engagement rates:

Your low-impression posts aren't failing. They're being seen by your most engaged audience: the people who already follow you and interact with everything. As posts reach wider audiences, ER drops because you're hitting cold traffic.

This means engagement rate alone is a terrible metric for judging content quality. A post with 200 impressions and 20% ER looks great on paper but reached almost nobody new. A post with 31,929 impressions and 7.31% ER is outperforming it in every way that matters for growth.

AI creators: stop optimizing for ER. Optimize for bookmarks and impressions. The ER will take care of itself.

My luthier/personal content had the highest engagement rate of any category at 15.6%. Only 10 posts but people deeply connected with the craftsman backstory.

One post in particular stood out. "A year ago I almost quit three times." It generated the highest curiosity rate of any post: 4.5% of people who saw it visited my profile. But it produced 0 direct follows.

Vulnerability drives curiosity. Expertise closes the follow. People wanted to know who this person was. They clicked through. But when they landed on the profile, they needed to see documented skill and methodology to hit that follow button.

For AI creators, this suggests a specific approach: lead with the human story, back it up with the documented expertise. "I build guitars by hand and create AI art" is more compelling than "AI artist." The personal angle drives the profile click. The portfolio converts the follow.

One last finding for the AI creators who want to optimize timing:

Wednesday outperformed Sunday by nearly 2x in net follows and 3.6x in bookmarks. Save your best content for midweek. The AI art community is more active, more engaged, and more likely to save content Tuesday through Thursday than on weekends.

The Wednesday bookmark number (43.4 average) is inflated by the Feb 4 Ambassador announcement landing on a Wednesday. But even adjusting for that outlier, midweek consistently outperformed weekends across 101 days.

Here's what 1,789 posts taught me, distilled into what you can actually use:

Give shoutouts regularly. 1.08 follows per post. Nothing else comes close. If you're an AI creator, shout out other AI creators. Build the network. Become the person people follow to find the community. Even two shoutouts per week would have generated more followers than all my tutorials combined.

Replies are invisible infrastructure. 53.8% of my posts were replies. They drove almost zero direct follows. But they kept my account visible, built relationships, and created the community context that made everything else work. Aim for 50+ daily conversations.

Document your process, don't just showcase your output. My top bookmarked content was all documentation and creative launches. Tutorials (the step-by-step kind) underperformed by 54x compared to shoutouts. But comprehensive documentation (the article kind) was my most bookmarked content ever. The difference: tutorials tell people what to do. Documentation shows them what you did and how it works.

Stop counting likes. Start counting bookmarks. Bookmarks predict growth at r=0.698. Likes predict at r=0.504. The gap compounds over weeks and months. Create content people save, not just content people appreciate.

Post less, post better. Volume correlates with growth at r=0.093. That's barely above random noise. One great documentation post per week will outperform daily image dumps. Every time.

Save your best work for midweek. Wednesday is the peak for follows, impressions, and bookmarks. Schedule your biggest launches and best content for Tuesday through Thursday.

Accept the power law. 0.1% of my posts drove 27.8% of follows. You can't control which post breaks through. But you can control whether you're creating the kind of content that shows up at the top of every performance list: documentation, creative portfolio launches, and community shoutouts.

1,789 posts. One year. Here's the uncomfortable truth:

Shoutouts beat tutorials. Bookmarks beat likes. Documentation beat threads. Community beat broadcasting. Quality beat volume. And most of what I posted, the vast majority, barely moved the needle at all.

But the infrastructure mattered. The 962 replies that generated 14 follows filled the room. The 267 threads that generated 10 follows built the authority. The 115 tool showcases that generated 6 follows kept me visible in the Firefly community.

None of it was wasted. But if I'd understood these numbers six months earlier, I would have written half the tutorials, double the shoutouts, and spent a lot more time documenting my process for other AI creators to learn from.

The data is there. 1,789 rows. Every number verified against official X analytics exports. The playbook isn't what you've been told. It's what the numbers actually say.

Glenn is an Adobe Firefly Ambassador and AI creator documenting the craft of prompt engineering and creative process at @GlennHasABeard. He publishes The Render newsletter and creates the Stor-AI Time series adapting world folktales through AI-generated video.

This article is part of a series analyzing one full year of X analytics: 365 days of account data and 1,789 individual post records. Every number is from official X analytics exports.