Clone what works: the EdTech twin that did 8.2M views
8.2 million views. 41,400 followers. 200x view-to-follower ratio. Same persona, rendered twice, sitting symmetrically. $150,000 in influencer fees, replaced by 4 agents.
TL;DR
@sofieestudies. 108 posts. 107 days. 8.2M views. Account run by 4 AI agents, not a human creator. The persona is rendered twice as "identical twins" sitting symmetrically. The visual hook the algorithm already proved out on the source account. Five moves: scrape the format, mirror the persona, list the proven beats, ship on cadence, queue the next clone. The new playbook: don't guess what works. Clone what already did.
Move 01
Scrape the format the algorithm already paid for
The discovery agent doesn't pick formats by taste. It scrapes the top-performing accounts in the niche, ranks each format by engagement curve and posting cadence, and surfaces the ones the algorithm has already validated. Taste-by-committee dies here. The agent ships only what the data already proved.
Most operators waste their first 90 days inventing formats that flop. This agent collapses that window to a single afternoon. Pull the top 50 accounts, score them, pick the winner. The bet you make is on a format that has already paid out for someone else.
Apply this: Before you generate a single asset, scrape the top 50 accounts in your niche. Score by view-to-follower ratio. The winner is your template.
Move 02
Mirror the persona: symmetry is the hook
The persona agent renders the "twin" duo. Same face, same lighting, mirrored body language. Symmetry is the visual hook. The viewer's eye locks onto the impossible-twin frame for the first second, which is the only second that matters in the algorithm.
The trick is the rendering pipeline doesn't need two characters. It needs one character, rendered twice, sitting symmetrically. Half the cost. Twice the visual interest. The asymmetry of standard talking-head content is what the format is leveraging against.
Apply this: Pick one visual gimmick the niche hasn't seen yet. Mirror it. Duplicate it. The eye-catch is the conversion.
Move 03
Clone the script structure that already converted
The clone agent writes a list-style script matching the source format's pacing. "Science → Khan Academy → Chemistry → Quizlet...". No invention. Just pattern-match to what already worked, swap the niche's nouns, ship.
Most brands try to differentiate at the script level. That's where they lose. The script is the part the algorithm already approved on the source account. Keeping the structure intact and only swapping the brand mentions is what makes the clone recognisable enough to surf the same wave without being flagged as a duplicate.
Apply this: Don't rewrite the structure. Copy the pacing, the beat count, the list length. Swap the nouns. Ship.
Move 04
Ship 3x a week for 100+ days: cadence is the moat
The publish agent schedules 3 posts a week for 100+ days, reads engagement after each drop, and queues the next clone before the previous one stops converting. 108 posts in 107 days. That cadence is what most operators can't sustain manually. And that's why most operators don't get to 200x view-to-follower ratios.
The format works because it doesn't stop. The algorithm rewards consistency over creativity at this stage. Three posts a week, every week, for 100 days, with each iteration informed by the last one's engagement data. That compounding is the moat. Not the persona. Not the script. The shipping cadence.
Apply this: Commit to 100 days of 3x/week before evaluating. Anyone who quits at day 30 is paying for the operators who didn't.
Move 05
Queue the next clone, kill the one that flopped
The agent doesn't ship one clone. It ships a portfolio. Each account in the portfolio is a separate test of the same format with one variable changed. Different demographic, different niche, different micro-aesthetic. The winners get the next 30 posts. The losers get archived.
Operators running 100 of these accounts are not competing for attention. They are competing for the 6-figure budget the brand on the other side is still spending on one influencer contract. The portfolio's worst account out-performs the brand's best paid placement.
Apply this: One account is a hobby. A portfolio of 50 is a business. Build the second account before the first one ships.
- "[Brand] saved $[X] in [creator/agency] fees. And replaced it with [AI format] that did [Y]M views in [Z] days"
- "The [niche] [persona/format] you're watching does not have a creator. It runs on a [N]-agent AI playbook"
- "Don't guess what works. Clone what already did"
- "Per-[clone/render/clip] cost falling from [6-figure contracts] to cents per [render]"
- "The brands still booking [real creators/influencers/UGC] are paying to invent formats. The ones running this playbook are paying nothing to copy them"
What’s actually running underneath
- Discovery agent Scrapes top-performing accounts in the niche via TikTok/IG public-facing data. Ranks formats by engagement curve, posting cadence, view-to-follower ratio. Surfaces the format the algorithm has already paid for. Not the one your creative director thinks should work.
- Persona agent (Seedance 2.0 / kie.ai) Renders the "twin" duo. Same face, mirrored body language, identical lighting signature. One persona, rendered twice. Half the cost of two distinct characters, twice the visual interest. Runs on kie.ai's Seedance pipeline (cheaper at scale than fal.ai).
- Clone agent (Claude script writer) Writes the list-style script matching the source format's pacing. Same beat count, same list length, swapped nouns. The script the algorithm already approved, with the brand's products inserted where the source account had theirs. Same Claude agents we run across the network.
- Publish agent Schedules 3x/week for 100+ days. Reads engagement after each post. Queues the next clone before the previous one stops converting. The compounding cadence is the moat. Not the persona, not the script. 108 posts in 107 days.
A mid-range EdTech influencer run for 108 posts costs $150,000+. This portfolio rendered for cents per clone. Same outcome, same algorithm reward, fractionally the cost.
The brands still booking real influencers are paying to invent formats. The brands running this playbook are paying nothing to copy them. That’s the whole arbitrage.
Want this shipped on your product in 2-3 days?
Book a 15-min call. I’ll show you exactly how the pipeline adapts.
Book a call →