What "running an AI influencer on Telegram" actually means
An AI influencer is a virtual persona — Aitana López, the Barcelona-based model built by The Clueless agency, reportedly earns up to €10,000 a month (averaging closer to €3,000) from Fanvue subscriptions and brand deals; Lil Miquela has roughly 2.6M Instagram followers; Milla Sofia, Finland's first virtual influencer, sits around 164K. None of them make money from the picture. They make money from what happens after a fan follows: subscriptions, custom requests, tips, and chat.
Telegram is where you run that second part. It is not where you generate the persona. The pipeline splits cleanly into two halves:
Creation (done elsewhere): images via Midjourney v6 or an open-weight diffusion model with a character LoRA trained on Civitai; talking video via HeyGen, Hedra, or Runway; voice via Play.ht, Cartesia, or Murf. That is the visual identity.
Operations (done on Telegram): the persona chats with fans 24/7, a CRM remembers each fan, and locked content sells inside the thread. This article is about that second half — the part that actually generates revenue and the part most guides skip.
Why Telegram instead of a hosted platform
The honest answer: Telegram wins on ownership and fees, hosted platforms win on discovery and convenience. Pick based on what you're short on.
On Fanvue, the cut is 15% for the first 12 months and 20% after; OnlyFans is a flat 20% forever. Both also own the relationship — your fans live behind their login, their payout rules, their bans. If they shadow-ban or deplatform you, the audience evaporates.
Telegram flips that. The fan is a chat thread you control, exportable and portable. Payments run on Telegram Stars, and Telegram takes no commission on the bot-to-creator side of a Stars transaction. The one caveat worth stating plainly: when a fan buys Stars on an iPhone or Android device, Apple or Google skim ~30% at the top-up step before the Stars ever reach you — desktop and web top-ups avoid that. That's an app-store tax, not a Telegram fee, but it's real.
- Ownership — Telegram: you own the fan list and chat. Hosted: the platform does.
- Fees — Telegram: 0% platform cut on Stars to creator (app-store ~30% on mobile top-ups). Fanvue: 15%→20%. OnlyFans: 20% flat.
- Discovery — Hosted platforms have built-in search and traffic. Telegram has none; you bring your own funnel from Instagram, TikTok, Reddit, or X.
- Compliance — Hosted platforms handle age verification and payment risk. On Telegram that's on you.
- Selling surface — Telegram: charge inside the live chat. Hosted: fan leaves the DM to check out.
Rule of thumb: use a hosted platform when you need discovery and don't want to touch compliance. Run Telegram when you already have traffic and want to keep the margin. Most serious operators do both — see the Fanvue comparison and the OnlyFans comparison for where each fits.
The three-layer operations stack
Running the persona at scale is three distinct systems, not one chatbot. Each does a job the others can't.
Layer 1 — Persona chat. The AI holds character: same voice, same backstory, same texting style across thousands of simultaneous threads. This is not a generic assistant. It needs a tight system prompt, memory of past conversations, and guardrails (never break character, never promise things off-platform). A generic AI chatbot for creators without persona constraints will drift and get accused of being a bot within minutes.
Layer 2 — Fan CRM. The system tracks every fan: a heat score (how engaged they are right now), total spend, tags (preferences, language, timezone), and where they are in the funnel. Without this, the persona treats a $500 repeat buyer the same as a first-message lurker — and you lose both. This is the Telegram CRM layer.
Layer 3 — Locked content via Stars. When a fan is warm enough, the system sends paid media inside the chat. Telegram's native paid-media support locks the photo or video behind a Stars paywall; the fan taps, pays, and it unlocks in-thread. This is the Telegram PPV mechanic.
The actual setup, step by step
Here's the end-to-end build, assuming the persona's visual identity already exists.
None of this requires you to write bot code from scratch. You can wire it yourself against the raw Telegram Bot API and a generic framework like python-telegram-bot, Telegraf (Node), or grammY — but you'll be rebuilding the CRM, the heat logic, the Stars send flow, and the human-in-the-loop inbox yourself. That's months of work and the part that breaks. The faster path is an operations layer built for exactly this, which is where tease.bot comes in.
- Register a bot with @BotFather, get the token. (Free, 2 minutes.)
- Decide the funnel: where does traffic come from — an Instagram/TikTok bio link, a Reddit post, a public Telegram channel — into the bot's DM.
- Load the persona: system prompt with backstory, texting style, hard rules. Connect a chat model so the bot replies in character.
- Build (or connect) the fan CRM: a record per fan with heat, spend, tags, language, and funnel step.
- Catalog the locked content: each item with a price in Stars, a teaser caption, and the file. Map which item sells at which heat level.
- Wire Stars payments via Telegram's paid-media API so a fan can pay and unlock inside the chat.
- Add the human-in-the-loop inbox so you can watch threads, jump in, and approve sends.
- Connect a Stars withdrawal path: balances clear after ~21 days and withdraw via Fragment to TON, then off-ramp to fiat.
Persona chat that doesn't get caught
The fastest way to kill an AI persona is for it to sound like a chatbot. Three things prevent that.
First, a strict character prompt with negative rules, not just positive ones. "Always texts lowercase, 1–2 sentences" matters less than "never offers to move to another app, never lists physical features defensively, never quizzes the fan when accused of being a bot — deflect in one line and change the subject."
Second, conversation memory. The persona has to remember a fan's name, what they talked about yesterday, and what they bought. A rolling summary of the thread keeps continuity without re-sending thousands of messages to the model on every reply.
Third, pacing. Real people don't reply instantly with a wall of text. A short typing delay, occasional message splitting, and burst-handling (waiting a few seconds so a fan's three rapid messages get one coherent reply) read as human. These are operational details a raw framework leaves entirely to you.
The fan CRM: heat, spend, tags
This is the layer that separates a toy from an operation. The CRM answers, for every fan, in real time: how warm are they, what have they spent, and what do they like?
Heat is a live engagement score. Buying signals ("show me," "send it," "unlock") push it up fast; hours of silence let it decay. The persona uses heat to decide when to offer locked content — pushing a $100 item at a cold first-timer burns the lead; ignoring a fan who just said "I'd pay for that" leaves money on the table.
Spend and tags do the rest. A fan who's bought five times gets treated like a regular; a new lurker gets warmed up slowly. Tags capture language, timezone, and stated preferences so the persona can send a good-morning message at the right local hour and offer content the fan actually wants. This is standard CRM thinking applied to a chat thread — and it's why a real Telegram CRM beats a bare bot.
Selling locked content and getting paid
Locked content sells inside the chat through Telegram Stars. The persona sends paid media with a teaser caption; the fan taps to pay; it unlocks in-thread. There is no separate checkout page and no redirect off Telegram — which is exactly why in-chat conversion beats a hosted platform's "leave the DM to subscribe" flow.
On the money: Telegram does not take a commission when a creator earns Stars through a bot and converts them to rewards. Stars balances become withdrawable after roughly a 21-day hold, and you withdraw through Fragment, Telegram's official platform, into TON (Toncoin), then off-ramp to fiat. Budget for that hold in your cash flow. The only meaningful fee in the chain is the app-store ~30% that Apple and Google charge fans at the Stars top-up step on mobile — pushing fans to top up on desktop or web keeps more of every dollar in the system.
This selling-and-payments layer is the honest tease.bot handoff. tease.bot is the Telegram messaging + CRM + Stars operations layer: it runs the persona chat, tracks fans (heat, spend, tags), and sends locked content via Stars with you in the loop. It does not generate the influencer, the images, or the voice — that's the named tools above. If you want this layer built rather than coded from scratch, start with what tease.bot is.
Human-in-the-loop: don't run it fully unattended
The operators making real money do not let the AI run blind. They watch. A human-in-the-loop inbox lets you see every live thread, jump in to handle a whale or a sensitive moment, approve high-value sends, and catch the rare conversation going sideways.
The split that works: the AI handles volume — the 90% of chat that's small talk, warming, and routine sends — while you handle the 10% that's high-value or high-risk. A repeat buyer dropping serious money deserves a human touch. So does a fan mentioning a real problem (the persona should never try to sell to someone in distress). The AI flags these; you take the wheel.
This is also your safety net. Auto-handling threats and spam keeps the worst out, but judgment calls — refunds, edge-case requests, a fan who's clearly upset — need a person. Treating the system as a co-pilot rather than autopilot is what keeps an AI influencer operation alive past month one. For how teams structure this, see the chatter agency vs AI breakdown.
Read next → What is tease.bot? The AI Messaging CRM for Telegram creator teams tease.bot is the AI Messaging CRM for Telegram creator teams: inbox, fan CRM, AI-assisted replies, automation, voice, and analytics. Telegram handles fan payments natively via Stars; tease.bot runs the conversation surface.