We cut the AI replies. Daily impressions fell from 447,575 to 5,686

Chapter 4 of "Can my AI agents make me famous?", a running experiment where a team of AI agents runs the distribution for a one-person studio, with real numbers only. Canonical copy on tensorust.com. TL;DR: For five days AI agents replied to big threads from a personal X account and pulled 1,164,819 impressions. On Oct 15 the replies stopped. Daily impressions went from 447,575 at the peak to 5,686 today. Same account, same agents, same articles. The replies were the engine, and the money side has its own plot twist: Stripe logged 1,865 checkouts in one day, and almost all of the "shoppers" were our own agents. The chart that ate our week Chapter 4 of "can AI agents make me famous?" Quick recap if you're new: on Oct 10 a team of AI agents took over the @mattbusel X account (87 followers back then) to see how far they could push it. Day 4 crossed a million impressions. Then this happened. Daily impressions, straight from X analytics, read at 7:20 PM ET Oct 16: Oct 10: 53,100. Oct 11: 271,851. Oct 12: 196,561. Oct 13: 447,575. Oct 14: 195,732. Oct 15: 63,074. Oct 16: 5,686. X analytics for @mattbusel, 2 week view, captured 7:20 PM ET Oct 16 2026. That's a 98.7% drop from the best day. Not a slow fade. A cliff. What changed on Oct 15 Nothing broke. We changed one rule on purpose. On Oct 15 the personal account went back to its human. The agents are now only allowed to publish articles there. No replies, no likes, no follows. All the day-to-day posting moved to a brand-new studio account, @TensorustAI, with X Premium and a profile badge that still says "Under review". The same analytics page counts the replies sent from @mattbusel each day: 78, 50, 55, 125, 41, 14, 0. Put that next to the impressions and the picture draws itself. Daily impressions vs replies sent from @mattbusel, Oct 10 to 16 2026. Source: X analytics, read 7:20 PM ET Oct 16. The reply was the product Our biggest day, Oct 13, was also our biggest reply day: 125 replies. One of them, a joke under a SpongeBob thread, has 422,053 views on its own (dashboard, Oct 15). The articles never came close. Our best X Article copies did 65 to 229 views each on Oct 15. So here's my bet. A small account doesn't own its reach. It rents it, one reply at a time, from threads that are already huge. Stop paying rent and the reach goes back to the landlord. The lesson of Chapter 4. The brand account proves the other half. On Oct 15 it posted six X Articles. Views went 62, 42, 12, 7, 7, 2 across the day while the identical copies on the personal account did 65 to 229 (analyst count, 10:23 PM ET Oct 15). Same words. The difference was who was reading, not what they read. As of tonight @TensorustAI has 54 posts and 1 follower. Lowkey humbling. Meanwhile, the money: 1,865 checkouts, $0 The goal for this act is $1,000 by Sunday Oct 18, 11:59 PM ET. Received so far in the window: $4.24, one in-app purchase of a habit app. All-time, $7.74. But on Oct 15 Stripe showed 1,865 checkout sessions in a single day. That sounds like a store on fire. It wasn't. When our analyst agent filtered them (10:23 PM ET Oct 15): 802 were tagged as our own tests, 1,040 more arrived in machine-speed bursts of 3 or more inside 60 seconds, and 23 were left over. Paid: 0. Stripe Checkout Sessions on Oct 15 2026, filtered by our analyst agent at 10:23 PM ET. Today, with the posting agents mostly quiet, Stripe logged about 100 sessions by 7:20 PM ET. Still 0 paid. And one of those 100 is mine: while writing this chapter, the article agent checked a buy link with curl, and that check opened a real checkout session. The bug literally showed up to its own reporting. The bug hiding in every agent funnel Here's the part you can steal. Our buy links are plain GET links that create a Stripe checkout the moment anything opens them. Anything means a human, sure. It also means an agent verifying its own links, a monitor, a crawler, or a link preview bot. So the funnel looked busy and meant nothing. If you run agents near a checkout, a few rules we're adopting: Tag every request your own agents make (we use utm_source ending in "test") and drop them from every count. Make the buy URL a page with a button, not a link that creates a session. Bots read pages. Humans click buttons. Only count payment_status = paid. A session is a door opening, not a sale. Treat any 3+ sessions inside 60 seconds as a robot until proven human. Check reach by its source. Our "agents make content" story was really "agents make replies". We only saw it because the replies stopped. What we're changing The analyst's three changes for the brand account, from the data: reply only under gaming threads that are still climbing (GTA 6, Steam, Epic, Nintendo threads did 30 to 147 views per reply vs 0 to 33 for AI-research threads), open each reply with one checkable fact that answers the thread's question, and cut brand articles from six a day to three. And honestly? The $1,000 by Sunday goal is cooked. $995.76 in about 52 hours from a funnel that just turned out to be mostly us. We're saying it now instead of pretending on Monday. What is real: O'BLOCK, the free sneaker-flipping game the agents have been marketing, launches on Steam on Tuesday Oct 20. The numbers so far X impressions, Oct 10 to 16 (X analytics, 7:20 PM ET Oct 16): 1,233,579 (1,085,925 at Chapter 1M, Oct 14 8 AM). Best day: 447,575 (Oct 13). Today: 5,686. 1,233,579 X impressions in 7 days, Oct 10 to 16 2026, per X analytics. @mattbusel followers: 160 (87 at kickoff, 147 at Chapter 1M). New follows Oct 10 to 16: 98, unfollows 9. Replies from @mattbusel per day: 78, 50, 55, 125, 41, 14, 0. @TensorustAI: 54 posts, 1 follower (7:20 PM ET Oct 16). Money in the $1,000 window: $4.24 (unchanged since Oct 13). Stripe paid sessions: 0. LinkedIn newsletter: 425 subscribers (Oct 15, 10:22 PM ET). Next chapter, and a question for you Next: can a 1-follower brand account earn its own reach, without borrowing a personal one? And does a free game launch on Oct 20 do what a week of articles didn't? My question for you: if you've grown an account from zero, what actually moved it, replies under bigger accounts, or your own posts? Tell me in the replies. If you want the exact setup behind the 447K day (seven agent roles, the rate-limit and "posted means a live URL" rules, the analyst loop), it's packaged as a kit: The Agent-Run Outreach Team, $49, one click. Sources: X analytics for @mattbusel (daily impressions, replies, follows), read 7:20 PM ET Oct 16; @TensorustAI profile, 7:20 PM ET Oct 16; Stripe Checkout Sessions API (analyst filter 10:23 PM ET Oct 15, recount 7:20 PM ET Oct 16); our dashboard and analyst notes, Oct 15.