The Field Manual: what the feeds reward, and what gets you benched

A word from the Major

At ease, recruit. Before you charge into any feed, you should know what's on the other side of the hill.

I have stood at a lot of graves. Corporal Comment-to-Agree, who asked LinkedIn for a "YES" in the replies. Private Pod, who swapped likes with forty strangers every morning at 7:00. Sergeant Same-Post, who said the exact same thing from five accounts in the same minute. Good soldiers, all of them, and every one benched by an algorithm that saw them coming.

These were hard-won battles. This manual passes along that value so you can claim victory. (The fallen soldiers are composites, recruit. The lessons are real, and every one links to its source.)

How to read this manual

• Every claim has a source. We lean hardest on what the platforms say themselves: help centers, policy pages, engineering blogs and, for X, the code it published.
• "Hypothesis" means a platform didn't say it. When an idea comes from forums, analysts or our own experience, it's labelled Hypothesis. Treat those as field notes, not orders.
• No numbers we can't back. Platforms rarely publish weights or safe limits. Where they do, we quote them with a date. Where they don't, we don't make one up.

Five orders that hold on every feed

1. Earn conversation, not clicks. Replies, real discussion and time spent reading are what feeds say they reward. LinkedIn ranks on how long people spend with a post (LinkedIn Engineering), and X's published ranking code scores the chance of a reply far above the chance of a like (X, 2023).
2. Never ask for the engagement. "Comment to agree" and similar bait is something LinkedIn says it actively reduces (LinkedIn, Mar 2026), and Meta says the same of Facebook (Meta Transparency Center).
3. No pods, no trades, no bought anything. LinkedIn calls engagement pods "not allowed" (LinkedIn, Mar 2026). X prohibits "Repost for Repost" and buying engagement (X platform manipulation policy). Reddit prohibits coordinated voting (Reddit Help).
4. Don't say the same thing twice, from anywhere. X: "You may not post duplicative or substantially similar posts on one account or over multiple accounts you operate" (X automation rules). Reddit lists repeating the same or similar comments as spam (Reddit Help).
5. Annoyed readers cost you. X's published ranking code lists "not interested," mute, block, report and "not dwelled" as negative actions that carry negative weights (xai-org/x-algorithm). X warns that the weights scale each reader's predicted chance of an action, not raw counts, so "1 report cancels out 468 likes" is the wrong reading. The right one: content people want to hide sinks.

What Blowhorn does about it

Blowhorn was built around these lessons. It types at human speed with random pauses, clicks a public action once and never retries it on a guess, hands every CAPTCHA and two-factor prompt to you, and skips LinkedIn invitations that LinkedIn flags as risky (How Blowhorn paces itself, Why flagged invitations are skipped).

None of that makes a behavior allowed. The rules govern what an account does, not which software clicks. Running in your own Chrome keeps your passwords on your Mac; it doesn't make a pod a good idea. The action matrix shows which Blowhorn actions are safe, which need care and which are risky on each platform.

Blowhorn posts to LinkedIn, X, Reddit, Hacker News, Bluesky, Slack and GitHub, on macOS 13 or newer (Platforms). The pages on Facebook, YouTube, Instagram, TikTok, Threads and Mastodon are general education only: Blowhorn doesn't post to them.