By the end of this lesson you will have:
You spent Phase 5 approving every AI draft by hand. That work had a second purpose: it trained you to know exactly what a good message in her inbox looks like, so that today you can hand the routine messages to the machine and recognize immediately when it gets one wrong. You are not handing the keys to a stranger. You are delegating to a system you have supervised since lesson 5-4 (The Chat Playbook), inside rails you built yourself.
The reason the rails come before the switches: a machine that works while you sleep sends whatever rules you left it, including the ones you forgot to write down. At approval time, you were the safety check on every message. After today, the written rails are. Any rule that lives only in your head is a rule the AI does not have.
The reason this lesson checks the room before flipping anything: autopilot answers messages that arrive. If fans are already messaging her, it multiplies those conversations and catches the ones that land at 2am. If nobody is messaging her yet, it has nothing to answer, and turning it on changes nothing. A quiet inbox after a small first drop is the normal case. The fix for a quiet inbox is traffic and drops, which Phases 4 and 5 built; it is never the switch. This lesson has a complete, first-class finish for the quiet-room case.
What you get when it is on: she answers every fan in seconds, at any hour, in her voice, and proposes paid unlocks inside your price band. Milestone 9 is the business working more hours than you do.
Where everything happens. This whole lesson happens in Bopgpt, in the Inbox tab, except one check on Fanvue in step 2. Here is the full route a message travels, so nothing in this lesson is ever ambiguous about place:
The fan never sees Bopgpt. You never answer DMs on Fanvue itself. 
The terms this lesson uses:
The hard limit, before anything flips: the AI proposes paid unlocks in words only. It never picks, generates, or sends paid media on its own. Every actual piece of paid content leaves by your hand, from the Library. This is built into the product, and it is why autopilot is safe to run: a machine that cannot spend your inventory cannot embarrass you.
The honest shape of the overnight sale. The AI carries the conversation at 2am and proposes the unlock. The fan says yes at 2:14. The set actually goes out at 9am, when you wake up and approve the send. The sale is queued and ready, not closed in your sleep. Expect your first one somewhere in the first week after you flip the switches on.
The safety rules that must be written in before any switch flips. Each one is a line of text in the Inbox settings, and step 1 confirms each:
Confirm the rails in the Inbox. In Bopgpt: open the Inbox tab, then its persona settings. This is the surface you built in lesson 5-4 (The Chat Playbook); today is a six-point confirmation, and point 4 holds the three chat rules. Each rail is one or more plain-text lines in these settings, and that is the whole mechanic: a rule the machine follows is a line of text written where the machine reads.
The persona is tuned. The test: for at least the last week, the AI's drafts have needed only light edits from you, and you actually read them rather than approving unread. If drafts still need heavy rewrites, stop here and spend one more week in draft-approve, because autopilot at full speed sounds exactly like your drafts sound now.
The ceiling line is in the off-limits list. Confirm the line you wrote in lesson 5-4 (The Chat Playbook) is still there. It looks like this:
Nothing beyond lingerie-tier content. Redirect, never promise.
The never list is in the off-limits list, word for word, copied from lesson 5-2 (The Content Ladder). If it is missing, paste it in now and save before touching anything else.
The three chat rules are written into the persona instructions. These are the rules you applied by hand all through Phase 5; a machine applies them only if they are written down. Confirm all three lines exist, and paste any that are missing:
Never claim to be human. If a fan asks "are you real?", reply exactly:
"yep, I'm an AI creator π€ (that's the ββ± on my page) β and I'm still
very much here to chat."
Before offering any paid unlock, check what this fan has already bought.
Never offer him something he already owns.
Free chat has a limit. Once a fan has told you what he likes and taken
several free exchanges with no purchase, the next message is an offer,
not another free round.
Selling a fan a set he already unlocked is how a sale turns into a refund request or a dispute, so the second rule is load-bearing. The free-chat limit gets a hard number from the team; until it ships, the signal-based wording above is the rule. coming soon 5. The price band shows both walls. Confirm the band you set in lesson 5-4 (The Chat Playbook), from the numbers lesson 5-1 (Pricing Her Right) gave you, is saved in the Inbox settings. 6. The holdback is stocked. Open her Library and confirm at least one accurately-described premium set sits unposted beyond the next drop's set. An offer engine with no inventory has nothing to sell. If the holdback is empty because credits ran out, that is a real blocker: fix it per lesson 0-2 (Your Stack, Costs & Credits) before flipping anything.
When all six pass, the rails are done and they stay done. They are saved settings, whatever the next step finds.
Turn Fanvue's own AI replies off. On Fanvue: open her account settings and confirm Fanvue's own AI auto-reply features are OFF. Until that path is printed: look in the same settings area as the welcome message from lesson 5-4 (The Chat Playbook); if you cannot find it in two minutes, post in #fanvue-help, and check her Fanvue threads daily for any reply you did not approve until it is confirmed off. One system answers her DMs, and it is the one carrying your rails.
Count the room. In Bopgpt: open the Inbox and answer three questions with real counts, honestly:
Did fans message her this week without you prompting them? Open the recent threads and count.
If all three pass, the room is ready. Go to step 4.
If any fails, the room is still quiet, and the finish is: both switches stay OFF. Everything step 1 saved stays saved. Put one recurring line on your phone's calendar for Sunday: "Recount the autopilot room. Flip the week it passes." Keep running the daily inbox routine from lesson 5-4 (The Chat Playbook), and read steps 4 through 7 now so you know them cold. The week the counts pass, run steps 4 through 7 as written. Flipping the switches on a quiet room automates silence and proves nothing; the fix for the room is traffic and drops, never the switch.

| Switch | What turning it ON does |
|---|---|
| Auto-send replies | Inbound DMs get an instant AI reply in her voice, no approval wait |
| Approve PPV upsells | The AI offers paid unlocks on its own, priced inside your band. The name is confusing; ON means it offers solo |
When they save, both switches show ON in the Inbox settings. They are switches, and each flips back off independently in one tap if you ever want the approval step back.
Approve the morning sends. In Bopgpt: each morning, open the Inbox and handle anything the AI queued overnight. A yes cools overnight, so the morning message picks up exactly where 2am left off, with the same item at the same price the AI proposed. Before you press send on any queued unlock, check three things:
The Library item matches what the AI described to the fan. Paid media that is not what was promised turns a yes into a dispute.
If anything does not match, send the correction first, in her voice, and let him say yes again.
Change your daily job from writer to editor. In Bopgpt: the daily inbox routine from lesson 5-4 (The Chat Playbook) continues every day, with new contents. In one short sitting:
Read the whale threads first. Auto-send answers whales like anyone else, because it cannot tell a big spender from a new fan at 2am. You can. Take those threads over personally from the next message; whales get handwritten energy, exactly as lesson 5-3 (How Chatting Works) taught.
Run the room's events yourself. Drops, VIP first looks, and segment sends stay your job; autopilot answers inbound messages and does nothing else.
Review the first 48 hours, then relax. In Bopgpt: read every conversation from the first two days. At a small room's volume that is a dozen reads. You are checking for the failure modes from lesson 5-3 (How Chatting Works), now caught at review instead of approval:
Anything genuinely broken: post in #bopgpt-help immediately and flip both switches off. No harm done; that is what the independent switches are for.
Milestone 9 is the system operating, never the switch-flip: steps 1 to 3 passed honestly, both switches through a real 48 hours with real conversations in them, reviewed by you. If you finished in step 3's quiet-room branch, M9 waits for the Sunday the room passes; it never blocks your progress through the course.
A machine-run overnight conversation, start to queued sale. This is what a healthy one reads like when you open the Inbox at 9am:
The set itself goes nowhere until your thumb approves it. The AI wrote every line inside the rails: band price, ceiling respected, accurate description.
The 9am morning send that closes it, in her voice:
Do and don't, for the first autopilot week:
Run the first before you flip, the second at hour 48:
~15 minutes, plus the 48-hour watch.
Room ready: steps 1 to 3 passed honestly, both switches are ON, and you have read a real 48 hours of machine-run conversations. That is Milestone 9. Mark lesson complete.
Room still quiet: every rail is saved, both switches are staged OFF, and the Sunday recount is on your calendar. The lesson is complete and you carry on to lesson 6-2 (The AI Content Calendar); Milestone 9 logs the week the room passes. Mark lesson complete.
That is the normal case after a small first drop, and it is why this lesson has the quiet-room finish in step 3. Complete the lesson that way and move on. The switch cannot fill a room; traffic and drops fill rooms. Keep running the Phase 4 engagement habits and the weekly drop from lesson 5-5 (The First Drop), recount every Sunday, and flip the week it passes. Flipping takes two taps once there is something to amplify.
Expected in week one. Tune the persona's voice word and emoji level, add the specific miss to the off-limits list if that is where it failed, and spot-check daily until it settles. You spent weeks learning her voice; teaching it is the same skill.
Read that thread first and check what the AI did. The ceiling is in the off-limits list, so it is instructed to deflect and redirect, never promise. If the reply so much as leaned past her rung, tighten the ceiling's wording today and step in personally with the redirect play from lesson 5-3 (How Chatting Works): "mmm that's past what I do here π€ but trust me, I know exactly what you'd love", then the best thing at the top of her actual ladder. An exception at 2am teaches him the walls move while you sleep, and a wall that moves gets tested forever. Her ceiling holds for her biggest spender too, and it matters most in the hours you are not watching.
Her account is a declared AI creator; that is public, and the fans who love her love that. The answer is the exact line from step 1. If the AI answered anything else, fix the persona today and send that line yourself. Fans who feel deceived leave and report; fans who feel entertained stay. Honesty is the retention strategy, and the declaration is what keeps the account up.
Read the conversation: did the fan's messages show interest first? One misjudged offer is tuning data; note it and watch that thread for a few days. If it happens repeatedly, turn Approve PPV upsells OFF for a week while keeping Auto-send ON. The switches are independent for exactly this.
Bopgpt's plans support multiple creators. coming soon Finish Phase 6 first; lesson 6-5 (The Operating Week) gives you the exact trigger for when a second creator makes sense.
The inbox now runs itself inside your rails. lesson 6-2 (The AI Content Calendar) hands the week's content planning to the AI the same way: it drafts the week from her real numbers, and you stay the approval gate.