Score your post before X does.
Paste your draft. Get a 0 to 100 score and the exact fixes that raise it, before you post.
Try:
“Finally launched my SaaS after 6 months of building! It finds Reddit threads where people ask for a tool like yours. …”
Needs work. Four fixes take it to 68.
How it scores
- Hook 25%50
- Payoff 20%70
- Specifics 15%96
- Reply pull 15%69
- Shareability 10%33
- Readability 15%56
Fix these first
- Link in the postX shows posts with outside links to fewer people. Post it without the link, then add the link as the first reply.+8
- Reads as an adPitches get muted and marked "not interested", which cost more reach than likes earn. Lead with the lesson or the result. Mention the product once, at the end.+8
- First line plays it safeSafe openers blend into the feed. Bold or specific ones stop the scroll. Make line one a claim someone could disagree with, or a number that surprises.+5
- No clear payoffPeople reward posts that give them something: a tactic, a number, a laugh. Give the reader one thing to take away: a tactic, a tool, a number, or a punchline.+5
- Easy to read in one pass.
- Concrete numbers and details.
- Gives people a reason to reply.
Announcement posts: Lead with the result or the problem it solves. 'We launched' is the least interesting part.
201 / 280 characters · 1 line · 3 hashtags · 1 link
The manual way.
- Write the post, reread it ten times, still not sure the first line works.
- Paste it into a chatbot and get "Great tweet!" with no reasons.
- Post it, watch it stall at 200 views, and never learn why.
One question.
- Paste the draft. Get a 0 to 100 score across the six things X rewards.
- Every fix names the problem and the points it adds, biggest first.
- Reach flags catch what X quietly punishes: outside links, hashtag piles, engagement bait.
A draft post in. A 0 to 100 score out.
Three steps, one request. You never call the tool by name — your agent does.
- 1
Judge the draft
A judge model answers two dozen yes-or-no questions about your post. Does line one make you want line two? Is there a real number? Would anyone reply?
- 2
Weigh it for the format
The answers roll up into six scores: hook, payoff, specifics, reply pull, shareability and readability. A joke is not marked down for having no numbers.
- 3
Flag and fix
Plain checks catch what X penalises, like outside links and hashtag piles. Every weak spot comes back as a fix with the points it would add.
Same tool, one prompt away.
Tweet Score runs inside Claude Code and any MCP client. Ask in plain words; your agent calls score_x_post and comes back with the score and the fixes.
10 requests a day, free. No card.
Write 5 versions of this launch tweet, score them, and give me the best one with its fixes applied.
score_x_post5 drafts, for indie hackers best: 81score_x_postdraft 3, fixes applied 89
X Draft 3 wins at 81. The first line is a specific result and the closing question pulls replies.- X The link costs 8 points of reach. It moves to the first reply.
- X With both fixes applied it scores 89. Ready to post.
Read-only. Nothing posted, nothing stored.
What comes back, and how to read it.
- You ask
- “Write 5 versions of this launch tweet, score them, and give me the best one with its fixes applied.”
- You get
- Five drafts scored in one request. Draft 3 wins at 81: a specific result in the first line and a question that pulls replies. Its link costs 8 points of reach, so it moves to the first reply. With two fixes applied it scores 89.
Read the fixes, not the number
A 72 is a checklist, not a forecast. The top fix is often worth more than the next three put together.
Replies beat likes
X ranks a reply far above a like. That is why reply pull carries real weight, and why a flat statement with nothing to answer scores lower than you expect.
Compare drafts, not accounts
The rubric is fixed and versioned, so scores compare fairly across drafts. A rewrite that scores higher is closer to what works. It does not predict views.
Honest limits.
- It scores the writing, not your account. Followers, timing and luck still decide most of the reach.
- It reads text only. Images, video and quoted posts are not scored yet.
- For a thread, score the first post. That is the one the feed shows.
- English drafts score best. Other languages work, with less nuance.
Asked before the first run.
Will it post for me?
No. Reddibee is read-only. Your draft is scored and dropped. We do not store it.
What does a good score look like?
85 and up is ready to post. 70 to 84 is strong with a fix or two left. Under 50 usually means the first line is not doing its job.
How is this different from asking ChatGPT?
A chatbot tells you it looks great. This runs the same fixed rubric every time and shows the points behind every fix, so you can tell if a rewrite actually got better.
Can my agent use it?
Yes. In Claude Code or any MCP client, ask it to write five versions and keep the best. It scores all five in one call, which counts as one request.
Five more questions, answered.
- Where's My AudiencePaste your URL. Get the communities your users already gather in, ranked and rule-checked.Try it free
- Post ScorePick a subreddit, paste your draft. Get a title score and a body score with the exact fixes, before you hit submit.Try it free
- Demand CheckDescribe the idea in one line. Get a verdict, an ask count, and the threads behind it.Try it free
- Switching MapName a product. See where people go when they leave it — ranked, cited, week by week.Try it free
- Idea AtlasWe read thousands of posts across builder communities. Pick a niche and see what people keep asking for.Try it free
Try Tweet Score now.
Free in the browser. 10 agent requests a day, no card.