Your X feed is not an act of God. It's an algorithm, and the single biggest input you control is who you follow. Mine had drifted into a swamp of engagement bait and people arguing about things I don't care about — while the actual communities I need to be part of were nowhere in my timeline.
And I need to be in a lot of communities. I run niche job boards — career.coffee, career.solar, career.dental, career.church — plus niche.careers, and I ship .NET side projects and Claude-powered agents. That's at least eight distinct worlds, and in half of them (coffee roasters, solar installers, dentists, pastors) I'm an outsider who genuinely doesn't know who matters.
Researching that by hand is weeks of work. So I asked Claude to do it.
The ask #
The prompt boiled down to this: you know my projects — now tell me who I should be listening to. Not a generic "top 100 tech accounts" listicle. For each of my niches, find the people whose attention would actually matter to my products, and for every single one, write down why.
I asked for the output as a CSV with these columns:
handle, url, name, category, confidence, status, why
Two of those columns are the whole trick, and I'll come back to them.
What came back: 508 people, zero filler #
Claude produced 508 accounts across my eight worlds:
| Category | Accounts |
|---|---|
| Indie hackers & bootstrappers | 127 |
| HR tech, recruiting & job boards | 115 |
| AI agents & LLM devtools | 89 |
| .NET / C# / Azure | 56 |
| Solar & energy (career.solar) | 38 |
| Coffee (career.coffee) | 36 |
| Church & faith (career.church) | 27 |
| Dental (career.dental) | 20 |
The names in the categories I do know well were a sanity check: levelsio, DHH, Arvid Kahl, Scott Hanselman, David Fowler, the Anthropic DevRel crowd. All the right people. Which made me trust the categories I don't know — the Job Board Doctor, Hung Lee of Recruiting Brainfood, James Hoffmann in coffee, SEIA in solar, Dentaltown's founder in dental.
The column worth stealing: why #
A list of 508 handles is a commodity. What made this one useful is that every row has a one-sentence reason specific to my situation, not a bio summary:
dhh— Cloud-exit crusader. Your Azure-to-Hetzner migration is precisely his thesis.
chrisrussell— RecTech / Job Board Secrets. He literally consults on launching niche job boards.
agazdecki— Acquire.com — the exit path for a portfolio like yours.
jimseven— The most influential voice in specialty coffee, full stop. One mention moves the whole niche.
That why
column turned out to be more valuable than the follow list itself. It doubles as an engagement cheat-sheet: when one of these people shows up in my feed, I already know what we have in common and what's worth replying with. Following someone is cheap; knowing why you followed them is the actual asset.
The second useful column is confidence
(high/med/low — 326/151/31 in my list). It's a built-in pruning knob: follow the highs first, sample the meds, skip the lows until proven otherwise.
Then I did the following myself #
Here's the part I screen-recorded. Claude built the list; the clicking was me. I worked through the CSV on x.com — open the profile, give it a two-second sanity check that the account really is what the list claims, hit Follow, mark the row status: followed
, next. That quick human glance turned out to be a useful final filter the CSV can't give you.
If the embed doesn't load, watch the short on YouTube.
Two constraints — one chosen, one imposed:
X will throttle you, guaranteed. This isn't hypothetical — it has happened three times already. The first run got cut off somewhere past 20 follows; roughly 20 minutes later the limit lifted and I carried on. The second batch hit the same wall at follow number 42, and the third at 59. So the pattern is about one forced every ~20 follows — don't plan for 508 in an afternoon, plan for follow--follow cycles, and keep the daily total modest so your account doesn't get flagged for mass-following. 59 followed so far — the whole HR tech batch went first because niche.careers is the priority right now.The CSV is the checkpoint. This is why thestatus
column earns its place: when X slammed the door mid-run, nothing was lost. Twenty minutes later — or in a completely fresh session next week — I just filter for emptystatus
and pick up exactly where I stopped. No memory tricks, no database — a file. (Regular readers will recognize this as the same philosophy asthis blog's CMS: state is files.)
This is what the wall looks like, for the record:
Take the list, but steal the method #
Here's the full CSV — all 508 rows including the why
column:
** ⬇ Download x-follow-list.csv** (70 KB)
If your interests overlap with mine (indie SaaS, job boards, AI agents, .NET), you can use it directly. But the reasons are tailored to my products — and that's exactly the point. The reusable part is the prompt shape:
You know my projects: [list them, with links].
Build me a list of ~500 X accounts worth following, grouped into
my niches: [list them].
For each: handle, profile url, name, category,
confidence (high/med/low), an empty status column, and one
sentence on why THIS account matters to MY work specifically.
No generic influencer lists. Output as CSV.
Then work through the list in slow batches, checkpointing into the status column as you go. Expect X to rate-limit you roughly every 20 follows — "Sorry, you are rate limited. Please wait a few moments then try again." — with timeouts around 20 minutes. The checkpoint column is what makes that a shrug instead of a restart.
The takeaway #
"Growing on X" advice is usually about what you post. But the feed you consume decides what you learn, who you reply to, and who eventually recognizes your name in their replies. Curating 508 follows by hand across eight industries was never going to happen. Curating them with an agent that already knows my whole project portfolio took one prompt, a CSV, and a screen recording short enough for YouTube Shorts.
The algorithm is going to feed you something either way. Might as well hand it a curriculum.