Check Every Post Before It Goes Out and Stop Wording That Gets Buried

The writer can't see what makes a post hard to reach

A social media post looks fine to the person who wrote it. But a URL in the middle of the text, four hashtags in a row, "please support us with likes," or text almost identical to last week's can all be reasons a post is less likely to be recommended. X's recommendation system is public, and you can read that it is built to dislike wording that asks for reactions and repetition of similar posts. The details are in How Posts Get Recommended on X.

In a company where one person writes and one person publishes, nobody points these things out. The SNS TOWN pre-post check is there so the screen can play that role. The same yardstick is applied to posts written by people and to AI drafts.

What it looks at

The check looks at the text, attachments and time, and reports the result in two groups: "Why it can't be posted" and "Things to watch." For AI-written drafts on X, the following are treated as reasons it can't be posted.

• A URL or an @mention in the body
• Two or more hashtags
• Requests for likes, reposts, follows or sharing
• Wording that assumes a following post, such as "more in the replies" or "in the thread"
• Text whose sentence sequence is 60% or more the same as any of the last 30 posts
• An empty body

For AI drafts, a post is also stopped if it contains 12 or more characters in a row that are identical to a post from a reference account. This keeps posts that copy another account's text from going out.

Similarity is judged by splitting text into three-character pieces and comparing how much they overlap. An overlap of 40% to 60% goes under "Things to watch." Stock phrases that read as if an AI wrote them, like "Did you enjoy this?", are also "Things to watch."

Timing is checked too. A post scheduled within 1 hour of the one before or after it on the same account is a "Why it can't be posted," and within 3 hours is a "Thing to watch." Posting in quick succession makes the later post eat into the reactions of the earlier one. A body over the character limit is stopped on every network and for any writer. The counting is the same as in the composer.

Posts written by people are shown as cautions

For X posts a person writes by hand, all of the items above are shown only as "Things to watch." That's because an announcement that needs a URL, or a post that mentions a business partner, has a reason to go out as it is. The only thing that stops it is something that will certainly fail on X's side, such as going over the character limit.

On Threads, Instagram, Facebook pages, LinkedIn, YouTube Shorts and TikTok, URLs and hashtags are handled differently from X, so the same items stay as cautions. Beyond that, we do stop posts that go over the character limit, posts within 1 hour of the one before or after, and posts without an image or video on Instagram, YouTube Shorts and TikTok. All of them either fail if published or get seen less when posted in succession. The table summarizes what stops a post.

For AI drafts, the reasons are handed back for a rewrite

When an AI draft is stopped by the check, SNS TOWN gives the reasons to the AI as they are and has it rewrite once. This rewrite is included in the 3 credits for a draft and doesn't use more. If reasons remain after the rewrite, the draft stays under "Awaiting approval" and can't be approved until it is fixed.

It works the same way when the post cycle is in auto mode. A draft that fails the check isn't published as it is; it goes to Awaiting approval. This prevents a post that goes out without a person's eyes from carrying a problem the check found.

One more look right before it goes out

The check runs when a draft is created, when it is saved, and again right before it goes out. Between creating a draft and publishing it, another post may have been placed near the same time, or a similar post may have gone out first. A post stopped by the last-minute check is left unpublished, and the reason is shown on the post's page.

A post's page lists the check results as "Why it can't be posted" and "Things to watch," and shows "All checks passed." when there is nothing. To fix a stopped post, rewrite the text, move the time, or rewrite it with AI. How to handle stopped posts is also covered in When a Post Fails to Go Out.

The yardstick follows public information

The check items and the rules we give the AI when it writes drafts were decided by reading the source code of X's public recommendation system. We also record which version we read, and if the system changes, we revisit the yardstick. The thinking behind how to write posts, including for other networks, is in How to Write Posts People Read.

Frequently asked questions