January 14, 2026. I posted a thread called "The algorithm is reading your behavior, not your follower count." It got 9,875 impressions. 55 people visited my profile. Solid numbers.

Two people followed.

Not two hundred. Not two thousand. Two.

At the time, I didn't think much of it. Bad days happen. I'd been growing at 15+ followers per day for weeks, riding a strong recovery out of the holiday crash. One off day wasn't going to rattle me.

But it wasn't one day. It was the start of an 18-day slide that nearly erased everything I'd built.

Some context. I'm an AI creator. I make photorealistic surrealist images in Adobe Firefly. The weird stuff. Giant mushrooms growing out of library shelves, clocks melting into coffee cups, that kind of thing. I'd been growing steadily since November, using a combination of educational AI-art content, community shoutouts, and daily engagement.

The first two weeks of January were strong. Jan 5 through 13, I averaged around +16 net followers per day. My best day in that stretch was January 11: 52 new follows from just 2,726 impressions. That's 19.1 follows per 1,000 impressions, the most efficient conversion day of my entire year. Low reach, but the people who saw it were exactly the right people.

Then I changed what I was posting about.

I'd been reading X's open-sourced algorithm code. Fascinating stuff. So I started writing threads about it.

"X's algorithm has a built-in mechanism for post quality." "I dug through X's open-sourced For You algorithm code." "The algorithm doesn't just predict positive engagement. It runs four separate models."

These threads were good. I spent real time on them, pulled actual data, cited real code. They were educational, well-structured, and genuinely useful for anyone trying to understand how the platform works.

The algorithm rewarded them accordingly. Impressions stayed strong. During the trap period (Jan 14 through 31), my average daily impressions were 6,410. That was actually higher than the 4,639 I averaged during the strong recovery stretch that came before.

More people were seeing my content than before. So naturally, more people would follow.

Right?

Here's what the weekly data looked like, verified against my official X analytics:

Read that trajectory. 15.6 to 14.9 to 8.6 to 5.0. Then an explosion to 27.6.

The first week looks fine. The second week looks fine. But the bleed had already started. By the third week, growth dropped 45%. By the fourth week, I was at 5.0/day. For an account that had been consistently hitting 15+, that's a collapse.

And the impressions tell the real story. Jan 12 through 18 had more impressions (50,475) than Jan 5 through 11 (31,869). The content was reaching more people. It just wasn't converting any of them.

I went back through every post I made during the trap period. All 319 of them. I found 31 posts that were specifically algorithm education content: threads about the open-sourced code, engagement mechanics, posting strategies.

Here's what those 31 algorithm posts produced:

Compare that to January 11, where targeted AI-art community content generated 19.1 follows per 1,000 impressions. That's a 45x difference in conversion efficiency.

The algorithm education threads had bookmarks. They had engagement. People found them useful enough to save. But the people saving them were growth-hackers, marketers, and social media strategists who had zero interest in following an AI art account.

Think about it from their perspective. They see a thread about how the X algorithm works. Useful. Bookmarked. Then they tap my profile and see surrealist rubber ducks in Victorian teacups. Not what they came for. They leave.

Same topic, same quality, same effort. But the audience it attracted had absolutely no overlap with the audience that would actually follow.

What made this especially hard to diagnose was that nothing looked "broken" on the surface. The metrics most creators watch, impressions, likes, engagement rate, all held steady or went up.

January 14, the day it started: 9,875 impressions with 773 engagements. That's a 7.8% engagement rate. Well above average.

But dig into the conversion layer and it falls apart. 55 profile visits from those 9,875 impressions gives a curiosity rate of 0.56%. Of those 55 who visited, only 6 followed and 4 unfollowed. Net: +2.

The people seeing the content weren't curious about who made it. They got what they needed from the thread itself and moved on. An algorithm explainer is a self-contained utility. You read it, you save it, you don't need the person who wrote it.

AI art works differently. Someone sees a surrealist photograph of a clock dissolving into coffee and they have an unanswered question: how? That question drives them to the profile. The profile answers it with 40+ documented techniques, prompt engineering breakdowns, and educational series. That's where the follow happens.

The algorithm threads had no unanswered questions. They were too complete. Too useful. Too generic.

February 1, I course-corrected. Stopped the algorithm education threads. Went back to what built the audience in the first place: AI-art-anchored content. Documented process. Creative work with visible methodology behind it.

February 3: I posted a single link to an article documenting my complete storybook video workflow. Just a URL, nothing fancy.

31,929 impressions. 462 bookmarks. 38 direct follows from that one post.

February 4: Adobe officially accepted me into the Firefly Ambassador program. I posted the announcement. That day: 107 new follows. +105 net. 32,978 impressions. 459 bookmarks. All-time records across every metric.

The week of Feb 2 through 8 produced +193 net followers at 27.6/day. Compare that to the final trap week (Jan 26 through Feb 1) at +35 and 5.0/day.

One week of AI-art-anchored content produced 5.5x the growth of one week of generic algorithm content.

Same account. Same follower count. Same tools. Same platform. Completely different results.

This isn't just a "find your niche" story. The trap is subtler than that.

I didn't leave my niche. I was still talking about growth, strategy, and platform mechanics, topics I'd been covering successfully for months alongside my AI art. The difference was framing.

When I wrote "Here's what I discovered about the algorithm while growing an AI art account," the thread attracted AI artists who wanted to grow. When I wrote "Here's how the X algorithm works," the thread attracted anyone interested in algorithms. One framing attracts your people. The other attracts everyone. And on social media, everyone is worse than someone.

The data backs this up at the platform level. My content category analysis across 1,789 posts showed that "Algorithm" as a category averaged 0.33 follows per post with 357 average impressions. But that average includes both the AI-art-anchored algorithm content (which converted well) and the generic algorithm content (which barely converted at all). Same topic. Different audience filter. 3 to 7x difference in conversion.

Key Principle #4 from my year of tracking: Audience matching beats content quality. Every time.

Excellent content reaching the wrong audience produces worse results than good content reaching the right audience. January proved it beyond any doubt.

If you're an AI creator and your growth has stalled despite "good" metrics, here's what to check:

Are your impressions up but your follows flat? That's the first signal. You're reaching people who aren't your people. Pull your last 30 days and compare impressions trend to follows trend. If they're diverging, you've got an audience matching problem.

Look at your recent bookmarks vs. follows. During my trap, bookmarks per day averaged 9.2 but follows per day dropped to 8.6. People were saving the content but not following the creator. In recovery, bookmarks jumped to 49.3/day and follows hit 18.9/day. When content and audience align, both metrics move together.

Check what you're posting vs. what built your audience. I'm not saying never post about strategy or algorithms. I still do. But I anchor it to my AI art practice. "Here's what I learned about the algorithm while generating 120 hidden object images" converts. "Here's what I learned about the algorithm" doesn't. The difference is one sentence of framing, and it determines who sees it.

Track your follow efficiency. Divide new follows by impressions, multiply by 1,000. My best days hit 19.1 follows per 1K impressions. My worst trap days were under 0.5. If your efficiency suddenly drops while impressions hold, the algorithm is serving your content to the wrong room.

The hardest part of the Algorithm Trap wasn't the lost growth. I made that back in a single week.

The hardest part was accepting that my "best" content, the threads I was proudest of, the ones where I really went deep into the open-sourced code and pulled out genuine insights, were actively hurting my account. Not because they were bad. Because they were too good at attracting the wrong people.

Quality is not the same as relevance. Reach is not the same as resonance.

I spent 12 years as a luthier building guitars by hand. One thing that teaches you is that the most technically perfect joint doesn't matter if it's in the wrong instrument. A flawless dovetail neck joint on a ukulele body isn't craftsmanship. It's a mismatch.

The algorithm threads were my dovetail on a ukulele. Perfect execution, wrong context.

When I came back to AI-art-anchored content, the growth didn't just recover. It accelerated. February 2026 is tracking at 18.9 follows per day with a curiosity rate of 0.70%. Both are the highest of any month in the entire year.

The algorithm didn't change. The audience did.

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 is currently 1,390 followers from his 5,000 goal.

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.