I created 5000 AI bots to completely decode the X algorithm.
Here's what I found :
The X algo has 4 main components:
• Home Mixer (orchestrates everything)
• Thunder (stores posts from people you follow)
• Phoenix (the AI that scores and ranks everything)
• Candidate Pipeline (the framework connecting it all)
Here's how they work together:
Every time someone opens X, the algo builds a pool of ~1,500 posts to even consider showing them.
500 million posts created daily.
Narrowed to 1,500 per user in under 200ms.
If your post doesn't make the candidate pool, it gets a score of zero.
Nobody sees it. Ever.
Your follower count does absolutely nothing.
Buried in the codebase:
author_followers_count is pulled for display only.
Not fed into any scorer. Anywhere.
100K followers = 1K followers in Phoenix's eyes.
What travels with your post is a PREDICTED engagement score based on your recent history.
This is the most important thing I found.
It's called the Prediction Trap.
Before your post reaches a single person, Phoenix scores it across 19 signals and assigns reach based on that prediction.
Phoenix predicts low engagement
→ shows it to fewer people
→ fewer people = fewer chances for engagement
→ prediction validates itself
→ post gets buried
Phoenix predicts high engagement
→ bigger test group
→ more engagement
→ prediction validates itself
→ post gets pushed further
The algorithm creates the outcome it predicted.
The 19 signals Phoenix predicts (ranked by importance):
1. Reply rate — 10/10 (hardest to fake, weighted heaviest)
2. Dwell time — 9/10 (how long they pause, two separate signals)
3. Profile clicks — 8/10 (they want to know more about you)
4. Bookmarks — 8/10 (saved = genuinely valuable)
5. Likes — 3/10 (cheapest signal, barely moves the needle)
One block = -3.0
One like = +0.5
A single block wipes out 6 likes worth of positive signal.
The scoring formula:
Final Score = Σ (weight × probability of each action)
Positive actions → positive weights
Negative actions → negative weights
Nobody outside X knows the actual weights.
But the architecture is confirmed from the public codebase.
Your recent posting history sets the baseline prediction for every future post.
This is the part that destroys most accounts.
If your last 10 posts averaged 0.3% reply rate, Phoenix predicts your next post will also perform poorly and distributes accordingly.
Bad posts don't just die.
They make your next post die too.
The AuthorDiversityScorer applies exponential decay every time you appear in the same follower's feed session.
Post 1 in their session → full score
Post 2 → decayed
Post 3 → decayed further
More posts ≠ more reach
The sweet spot is 2-3 posts per day with 4-6 hour gaps.
Posts die after 48 hours. Hard stop.
Thunder (X's in-memory post store) auto-trims every 2 minutes.
Retention window: 48 hours.
After that your post is gone from the candidate pool entirely.
The idea that consistent posting lets older content keep circulating is wrong at the architecture level.
You start from zero every 48 hours.
Reposts of other people's work = 90% impression deduction on that post.
Self-reposts = completely fine.
The Bloom filter resets each session.
A follower who hasn't opened X since this morning hasn't seen your earlier post in their current session.
Self-repost reaches them fresh.
Out-of-network reach is a mathematical process, not luck.
Phoenix uses a Two-Tower Model:
Tower 1 → encodes your engagement history into a taste fingerprint
Tower 2 → encodes every post into a content fingerprint
Phoenix then runs similarity search across millions of users.
If your post's fingerprint matches someone's taste fingerprint, it enters their candidate pool even if they don't follow you.
This is the viral mechanism.
Niche consistency compounds this.
An account that posts in one lane for 6 months builds a tight content vector Phoenix can reliably match to the right audiences.
What to do:
→ Engineer reply triggers into every post
→ Post scene-setting stories with a question at the end
→ Space posts 4-6 hours apart
→ Self-repost your best content 12-24 hours later to hit new sessions
→ Recycle proven hooks — 48hr reset means new audiences each time
→ Track reply rate, new follows, bookmark rate per post — nothing else matters
What to stop immediately:
→ Posting standalone links (near-zero reply rate poisons your Phoenix baseline)
→ Public "sent to your DMs" confirmation replies (each one is a dead post in your history)
→ Telegram/channel promotion posts as standalone feed posts
→ Posting vanity metrics with no story or question attached
→ Optimizing for impressions (vanity) or likes (weakest signal)
→ Posting 5+ times per day (you're decaying your own posts)
The platform changed the rules completely.
The habits that built audiences under the old algo are actively working against you under the new one.
Consistency without reply-rate optimization is just feeding Phoenix bad data points.
One post with 1.5% reply rate does more for your reach than twenty posts with 0.1%.
Understand the system.
Then work with it.
This is how I consistently do millions of views across all X accounts daily. Not only views but more importantly , target high quality US audience so I can easily do $2K-15K sale with digital products everyday. (quality of views is more important than # of views)
Want the full breakdown of every component including the GitHub repo secretly released by xAI?
Reply "ALGO" and I'll send it to you.
**must be following + retweet to receive it.
I was smoking hookah with a guy who did $65M+ selling AI to Fortune 500 companies in West Hollywood
we did a little mastermind and he asked me how I did $3.5M+ selling ebooks organically.
So I begin explaining and show him my 60 modules blueprints and videos explaining how it all works
He looks at me, straight in the eye, and asks me why I'm not selling that for $10K at least???
I replied:
"Oh I don't feel like it. I'm busy with other stuff"
He said :
BRO AS SOON AS YOU GO HOME SELL IT TO YOUR FOLLOWERS FOR $10K PER PERSON THAT IS THE MOST VALUABLE PRODUCT YOU HAVE YOU COULD MAKE $100K TONIGHT JUST FROM DOING THAT
so now im giving it out to you guys.
but for free.
Comment "BLUEPRINT" and I'll send it to you via DM
**must be following + retweet
iman and jason are running one of the best webinar funnels I’ve ever seen
they prolly made millions in a day
i’ve funnel hacked his entire strategy including his exact ads
if you want me to send that over to you for free
repost and comment the word “iman”
Digital Products is the #1 BEST business in 2026
- AI does all the work for you
- create 5 ebooks , 200 pages each within 20 minutes with AI , sell each for $500 ( other businesses take days at minimum )
- 99% margins and instant access & delivery to customer
- ZERO headaches
- takes 1 hour of work per day .
- $40 investment per month necessary to get started
Easily do $3000+ profit days
I started 2 years ago and now make $400K per month at 22 years old
And now, I'm teaching how, by giving you for free:
1) full ebook 100 step-by-step blueprint
2) YouTube tutorial ( 40 minutes )
3) Miro board + 15 templates + tools to get started
comment "HOW" if you want ALL this :)
** DELETING ALL THIS IN 24 HOURS
** Retweet and follow required to get
i found a way to make AI videos that don't sound like AI...
example below....
sora 2, veo 3.1, kling 2.6 are decent but they all have that robotic AI accent that everyone can detect immediately
and when people try replacing the voices with elevenlabs, it always sounds way too professional
like it's recorded on a professional mic in a studio
but if i'm making a UGC video of someone in their room, i don't want studio quality
it should sound like it's actually in that room
the workflow i use completely removes that fake AI voice sound
why this matters:
when your AI voices sound too polished or robotic, people notice. their guard goes up. they disengage.
but when your voices sound like they're coming from a real person in a real room, that's when they work the best
i created a guide breaking down exactly how:
> to create voiceovers that sound completely real
> to replace fake-sounding voices in AI videos (not elevenlabs)
> to make voices sound like they're in the actual environment, not a studio
RT + reply "VOICE" and i'll send you the FULL guide (must follow so i can DM)
most people create digital products backwards and wonder why nobody buys
they make the product first, then try to find customers
I use the "reverse engineering" method instead
here's the full 25-step system:
**PHASE 1: MINING YOUR AUDIENCE (Steps 1-7)**
1. Post about your niche for 14 days straight without selling anything. just value.
2. Screenshot every single DM you receive asking "how do I do X?" - this is gold.
3. Save every comment that says "can you explain this more?" - these are product features.
4. Check which tweets got the most bookmarks. bookmarks = "I need to save this for later" = people want to go deeper.
5. Look at which tweets got the most replies asking questions. questions = confusion = opportunity.
6. Create a google doc and dump all of this into categories. you'll see patterns.
7. The category with the most entries? that's your product. the market just told you what to build.
**PHASE 2: VALIDATION BEFORE CREATION (Steps 8-12)**
8. Post a tweet: "thinking about making a guide on [topic]. would anyone actually want this?"
9. If less than 20 people respond with interest, pick a different topic. the market spoke.
10. If 50+ people respond, you've got a winner. screenshot every response for your sales page later.
11. DM 10 of the most engaged responders. ask them: "what would need to be in this for you to buy it?"
12. Write down every answer. these aren't suggestions. they're the actual modules you'll create.
**PHASE 3: THE PRE-SALE (Steps 13-17)**
13. Create a simple landing page describing the product BEFORE you build it. use Carrd ($19/year).
14. List exactly what's included based on the DM responses from step 12.
15. Price it at 50% off "founding member" rate. scarcity is real because you're only taking early buyers.
16. Post: "building this guide in public. first 20 buyers get it at half price and can shape what goes in it."
17. If you get 10+ pre-sales, build it. if you don't, you just saved yourself weeks of wasted work.
**PHASE 4: BUILDING WITH BUYERS (Steps 18-22)**
18. Create a group chat with your pre-sale buyers. discord, telegram, whatever.
19. Build the product module by module. share each one as you finish it.
20. Ask for feedback after each module. what's confusing? what's missing? what do they want more of?
21. Your buyers are now co-creating the product with you. they're invested. they'll become your best testimonials.
22. By the time you "launch" you have a product shaped by real customers, not your assumptions.
**PHASE 5: THE TESTIMONIAL ENGINE (Steps 23-25)**
23. As founding members get results, screenshot everything. DMs, tweets, comments, all of it.
24. Turn each testimonial into content. "just got this DM from a student who [result]."
25. Each testimonial tweet sells more than any sales page ever could. social proof compounds.
**why this works better than the normal way:**
traditional method:
- spend 3 months building
- launch to crickets
- realize nobody wanted it
- repeat with new product
reverse engineering method:
- validate in 2 weeks
- pre-sell before building
- build exactly what people asked for
- launch with testimonials already collected
you literally cannot fail because you only build what's already been validated.
**the pricing ladder most people miss:**
founding member price: 50% off (rewards early believers)
launch week price: 25% off (creates urgency)
regular price: full price (anchor for future discounts)
premium tier: 2x price with added bonuses (captures people who want more)
I always sell at founding price first. even if only 10 people buy.
those 10 people become case studies that sell 1,000 more copies.
**the content loop that sells forever:**
step 1: post free content about topic
step 2: people ask questions in comments
step 3: answer questions in future content
step 4: mention "I go deeper on this in my guide"
step 5: curious people check your profile and see the offer
step 6: repeat forever
you're not launching once. you're selling every single day through content.
**the math on why this prints:**
10 founding members at $50 = $500 before you build anything
validation that the market wants it = priceless
50 launch week buyers at $75 = $3,750
ongoing sales at $97 = $500-$2,000/month passive
from a product you knew would sell before you made it.
**the biggest mindset shift:**
stop thinking like a creator. start thinking like a researcher.
your job isn't to guess what people want.
your job is to listen until they tell you exactly what to build.
then build that. nothing more. nothing less.
I used this exact system to go from $0 to $600K in digital product sales.
never launched a product that flopped because I never built anything until the market told me to.
put together a complete guide walking through this entire reverse engineering system.
the validation tweets word for word
the DM scripts for customer research
the pre-sale page template
the founding member launch sequence
the testimonial collection system
how to price each tier
45+ modules breaking down every single step.
comment "REVERSE" and I'll send you the link.
must follow + RT to receive.
gone in 48 hours.