How an OFM chatbot works

The one mechanism underneath

Both platforms sell the same way in a private conversation: a photo or a video is sent locked, with a price, and the fan pays to open it. On OnlyFans that is a paid message in dollars. On Telegram it is a price in Stars, the in app currency Telegram sells to its own users, or a payment link attached to the message. Everything else around it, the hours of chatting that make a fan want to open anything, is conversation.

An AI chat tool automates the conversation and the moment the locked media goes out. It does not invent the media, it does not invent the scene, and it does not invent the price. Those are written by a person in advance. What the AI does is decide what to say now, to this fan, given everything said before, and when the prepared scene should start.

1. The connection to the account

The AI has to reach the inbox fans write to, and the two platforms are not alike here.

OnlyFans publishes no public interface for third parties. Its own help center says so in the same sentence it uses about AI chat: there is no public facing API. So tools work from the creator’s own logged in browser session, usually through a browser extension the operator installs. The practical consequences are real: a browser has to exist somewhere, the session can expire, and everything the AI sends leaves from the creator’s own account.

Telegram publishes the opposite. Bots are a documented product, and a business account can attach one to its private chats. Telegram’s own developer documentation states that users can connect bots that process and answer messages on their behalf. In practice that means three objects: the creator’s Telegram account with the business features, a bot created in Telegram’s BotFather, and the connection between the two. Fans never open the bot; they write to the creator’s profile and the answer comes back from that profile.

What each platform publishes, quoted with dates, is on Telegram vs OnlyFans.

2. The gates before any reply

A reply is not automatic. In the one product of this category with a complete public help center, OnlyChat, three switches are checked before anything is written: the creator level switch, the per fan switch, and on OnlyFans the fan lists, where the rule is that the AI answers only fans in at least one active list and in no excluded list. A schedule can add a fourth gate, switching the AI on inside a time slot and off outside it.

This matters more than it sounds. Gates are how an operator keeps friends, other creators, refund cases and hand managed spenders out of the automated flow, and they are the first thing to check when someone says the AI is not answering.

3. Context and memory

Before writing, the AI reads the thread, what it already learned about this fan, and what he spent. The learning is not a form the fan fills in; it is extracted from the conversation. OnlyChat calls this phase discovery and its default target list is first name, age, city, occupation, relationship status and passions.

This is the piece that produces the difference between a message that could go to anyone and a message that could only go to him. It is also the piece that quietly decides how a scene is chosen later.

4. The voice

The persona is a set of fields, not a prompt someone writes once and forgets: identity and appearance, personality traits, languages, and a free text block the AI reads word for word, holding the daily routine, the boundaries and the running jokes. Two separate style settings sit next to the facts: how the AI writes, and how it defends a price.

The full field list, what belongs in the free text and what a persona sheet cannot fix is on Persona and voice.

1. Fan:
   you there
   23:41
2. The AI:
   just got in. tell me about your day first
   23:42
3. Fan:
   boring. send me something
   23:44
4. The AI:
   patience. i want to know what mood you are in
   23:44

Written for this page. The AI is not being coy for effect: the scene has not started, so there is nothing to send yet, and the pacing between steps is a setting.

5. The scene

A scene, called a script in every product that has one, is a short story cut into steps and sold step by step. The AI does not send scenes on a timer. It starts one when the conditions of the conversation are met, picks one the fan has never seen, and reads the internal description of each scene to decide which one fits what the fan has been talking about.

A scene is used once per fan, which creates the constraint that decides the whole editorial workload: an operator is never finished writing. How steps, ladders, floors and pacing are built is on Scripts and PPV sequences.

6. The step

One step is one block: the message written by the operator, the media, and a price. A price of zero sends the media free, and that is the teaser. A price above zero sends it locked. Under the price sits a floor, the lowest amount the AI is allowed to accept on that step, and it only exists if the operator set it.

The AI does not rewrite the message on a step. What the operator wrote is what the fan reads, translated if the fan speaks another language. Everything between two steps, the teasing, the answers to questions, the price defence, is written by the AI in the persona’s voice.

7. What the fan decides

Three outcomes, and only these three. He opens the media at the asked price. He pushes back, and the AI argues, then falls to the floor if one exists and stops there. Or he says nothing, and a follow up goes out after a delay the operator set. A long enough silence ends the scene, and a cooldown starts before any new scene can begin.

8. What is measured

Two numbers do the work. The unlock rate is locked media opened divided by locked media sent. Conversion per step says where in the ladder fans stop, which is the only reliable instruction for rewriting. Revenue per scene ranks the scenes worth duplicating.

One published figure exists in this category: OnlyChat publishes a 43 percent unlock ratio, defined on its about page as 43 of every 100 paid photos or videos sent by the AI being opened. No other tool in the dataset behind this site publishes a comparable figure, so there is nothing to compare it against, and this site does not manufacture a comparison.

What running it costs

Two billing shapes exist, often combined. A monthly amount per creator, and a percentage of what the AI sells. Of the 32 AI chat tools in the dataset behind this site, 16 take a percentage, on top of or instead of a monthly fee. The exact figure for each one, with the date it was read from its own pricing page, is in the comparison table.

The cost that nobody prints on a pricing page is the writing: the persona, a description for every media, and enough scenes that a returning fan still has something he has not seen.

What this page deliberately does not say

It does not say that any of this is allowed, compliant or safe on OnlyFans. That question has one honest answer, which is to read what the platform publishes and the date it published it. It does not give a detection rate either, because nobody publishes one.