# AI influencer workflow: a realistic week The AI influencer workflow for a normal week: the five jobs that take your time, what you can automate safely, and the two checks you must do yourself. - Canonical: https://aiofmcity.com/guides/running-an-ai-persona-operation - Topic: Operations (https://aiofmcity.com/topics/operations) - Author: Gus - Published: 2026-09-08 - Updated: 2026-10-05 --- ## Key points - Automate file handling, renaming, scheduling and backups. Do not automate captions and selection: when they go wrong, you do not see it. Check the output of every re-encoding step once, because most optimisers remove metadata by default. - A human always does the age-appearance check and makes the final decision to publish. - Correspondence is the only bucket that grows with your audience. With a second persona you have to do it twice. - Fanvue says to post daily, or at least every other day. Schedule the week in the Queue in one session. - Fanvue has no custom lists yet. Leave the whole "Spent more than $50" list out of mass messages and write to those fans one by one. --- Running an AI persona means doing five jobs again and again. Generating images looks like the main work, but it is the smallest of the five. This guide explains what the five jobs are, which ones you can automate safely and which ones break when you try. It also covers the two checks that a human always has to do, how often to post, how to use Fanvue's fan lists and promotions, and a checklist for each week. Everything here applies whether you call it an AI persona, an AI influencer or an AI model. The work is the same. Only the name is different. ## The five jobs in every operation Every operation has the same five jobs, whatever tools you use. This guide calls them buckets. [Where the hours go](https://aiofmcity.com/blog/where-the-hours-go-ai-persona) has the details. This is the overview: _Diagram: Every operation has these five jobs, whatever tools you use. Correspondence grows with your audience, and it is the job that earns._ - Preparation — dataset, training - Generation — in batches, almost free per image - Selection — needs your judgement - Distribution — scheduling, labels - Correspondence — grows with the audience - **Preparation:** you build the dataset, train the model and retrain it when the results drift. This takes a lot of time at the start and then almost none. It takes time again whenever something changes. - **Generation:** you produce the images. Do it in batches. Setting up a session takes time, but each extra image costs you almost nothing. - **Selection and review:** you choose what gets published. Almost everyone underestimates this bucket, because it needs judgement and you cannot turn it into a routine. It includes the age-appearance check on every single image. - **Distribution:** you write captions, schedule posts, cross-post and set the right AI label on each platform. This is routine work, and it is the easiest bucket to automate. - **Correspondence:** you answer messages, including the same question for the fortieth time. This bucket grows with your audience, while the others stay the same size. It is also the bucket that earns. Fanvue reports that paid messages are the majority of new-creator income. [How money is made on Fanvue](https://aiofmcity.com/blog/where-fanvue-money-is-made) explains this in detail. ## What you can automate, and what you cannot "Is this repetitive?" is the wrong test. Ask this question: "Does this need a judgement that I would defend if someone asked me?" - **Automates cleanly:** moving files between steps, renaming, resizing, scheduling, backups and collecting the few metrics you look at. These tasks need no decisions, and automating them saves you real time. - **Automates badly:** captions, selection and anything that touches platform policy. Automating them is not hard. A model can write captions all day. The problem is that you do not see it fail. Automated captions do not throw an error. They all start to look the same, and your feed reads as machine-made months before your engagement numbers show it. - **Never automated:** the age-appearance check and the final decision that a specific image is fit to publish. ## The pipeline step that removes AI metadata by accident Most pipelines have an optimiser somewhere: a step that resizes, re-encodes or compresses your files. Most optimisers remove metadata by default. That metadata includes the provenance manifest, which is the record the generator puts into the file to show that it was made with AI. The AI Act puts the duty to add it on the generator's maker: > Article 50(2): "Providers of AI systems, including general-purpose AI systems, generating synthetic audio, image, video or text content, shall ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated." > > — Source: EU AI Act Art. 50, read 2026-10-01 — https://eur-lex.europa.eu/eli/reg/2024/1689/oj So the AI Act does not ban you from removing the manifest. Keep it anyway. Platforms read it to label your posts automatically, and your generator's terms may forbid removing it. [Don't strip your watermarks](https://aiofmcity.com/blog/dont-strip-your-watermarks-provenance) shows where pipelines do this and how to check. For your own pipeline, do three things. Know which of your steps re-encode files. Inspect the output of each of those steps once. Then use a step that keeps the manifest, or a step that does not re-encode at all. I learned this from my own mistake. One of my resize steps had been removing the manifest from every file it touched. You can read what happened [in the same article](https://aiofmcity.com/blog/dont-strip-your-watermarks-provenance#failed-a-one-line-batch-convert). ## How the work fits into your week Two operations can take the same number of hours, and one fits into your life while the other does not. The difference is whether you can move the work to another time. You can move preparation and generation. Do them in one batch whenever you have a free block of time. You can prepare distribution in a batch too, but the posts still go out on fixed days. You cannot move correspondence at all. A reply three days late is worth much less than a reply on the same day. Plan around the fixed parts. Write down which hours you can move and which hours are fixed. Then decide whether the fixed hours fit your life, before you promise your audience a schedule. Most people only find out which kind of operation they have built after they have announced the schedule. _Diagram: Preparation and generation can go wherever there is a free block. Posts land on fixed days, and replies happen every day. Check that the pinned pieces fit your life before announcing a schedule._ ## A posting schedule you can keep Post on Fanvue every day, or at least every other day. On Instagram, aim for about 10 reels a month. Of the three platforms in this section, only Fanvue and Instagram give creators a number. TikTok asks you to set your own daily target. Fanvue's advice is in its help article on keeping subscribers: > "If you go too long without posting, there's a high chance fans will unsubscribe." Fanvue's recommendation: "Aim to post daily, or at minimum, every other day." It also suggests using the Vault "to upload and store content in advance" and to "Schedule posts to ensure consistent delivery, even when you're busy." > > — Source: Fanvue Help — Tips to Retain Subscribers, read 2026-10-01 — https://help.fanvue.com/en/articles/9527118-tips-to-retain-subscribers On Fanvue, posts keep the fans you already have. Subscribers pay every month, and an empty wall gives them a reason to cancel. You do not have to be online every day to post every day, because Fanvue can publish posts for you: > There are two ways to schedule a post. From "New Post", click "Schedule" and choose a date and time. Or go to Creator Tools, then Queue, click a date and choose "Add Post to Queue". The Queue page shows scheduled posts in a calendar or list view, where you can edit or delete them. > > — Source: Fanvue Help — Scheduling posts, read 2026-10-01 — https://help.fanvue.com/en/articles/9539637-scheduling-posts So fill the Queue in the same session as generation and selection. The posting days then run without you, and your daily time goes to messages. Instagram's number is about reels, and it comes from its creators site: > "What is the optimal number of reels I should share to grow?" Instagram's answer: the frequency you post new reels "is just one factor of many, but we've found that when we look at creators with the greatest net follower growth rates we see, on average, they post 10 or more reels per month." > > — Source: Instagram Creators — Frequently Asked Questions, read 2026-10-01 — https://creators.instagram.com/faq Ten reels a month is two or three a week. It is an average from the fastest-growing creators. Posting 10 reels does not promise growth. TikTok tells creators to pick a daily number they can keep: > On 30 July 2020 TikTok wrote to creators: "The amount of videos you post won't impact how your content is recommended in the For You feed, and views vary from video to video." > > — Source: TikTok Newsroom — 5 tips for TikTok creators, read 2026-10-01 — https://newsroom.tiktok.com/en-us/5-tips-for-tiktok-creators > "While there's no one-size-fits-all answer, success cases show that publishers who post multiple times a day experience higher video views and exposure on the For You feed. We suggest setting a daily post target you can consistently hit and planning your content around that goal." > > — Source: TikTok Creator Academy — Posting cadence tips for all industries, read 2026-10-01 — https://www.tiktok.com/creator-academy/en/article/posting-cadence-best-practices-publishers Turn this into a monthly number before you promise anything. For one persona on Fanvue and Instagram, the arithmetic looks like this: | Plan | Fanvue posts | Instagram reels | TikTok videos | Finished posts a month | | --- | --- | --- | --- | --- | | Every other day on Fanvue | 15 | 10 | 0 | 25 | | Daily on Fanvue | 30 | 10 | 0 | 40 | | Daily on Fanvue, one TikTok a day | 30 | 10 | 30 | 70 | These are finished posts, after selection. You also need images for paid messages on top. Start with the first row. Move up only when you have kept it for a month and your selection still holds. The four posting days in the week above already give you Fanvue's "every other day". If you rush selection to hit a number, the feed starts to look automated. That happened to me, and [the story is further down](#failed-automating-the-selection-step-to). ## The bucket that does not scale, and what to do about it Four of the five buckets stay the same size. They take the same work whether you have fifty followers or five thousand. Correspondence grows with every new follower. That is what limits how big a one-person operation can get. Correspondence is also where the income comes from. So "just do less of it" is the wrong answer. > Fanvue provides AI messaging features that draft new messages, draft replies and > rewrite drafted ones, plus conversation summaries. It reports that about 1 in 4 > creators who message fans use AI assistance, and advises using AI for volume and > speed while reserving your own voice for the moments that matter most, such as a > new subscriber's first message or a high-spending fan. > > — Source: Fanvue Help — AI Messaging: Reply Faster, Chat Smarter, read 2026-09-09 — https://help.fanvue.com/en/articles/16237836-ai-messaging-reply-faster-chat-smarter "AI for volume, your own voice for the moments that matter" is the best rule in this guide. It solves the scaling problem, but the work does not go away. The moments that matter are few, and you can name them: a first message, a fan who has just spent a lot, and anyone who asked a real question. Everything else is volume, and you can use tools for volume. For your week this means one thing: give correspondence a **fixed slot every day instead of one slot a week**. You can do the other four buckets in one long session. You cannot do that with messages, because a reply two days later does not do the same job as a reply the same evening. If you have a bad week and something has to go, drop generation. You have a library of images to post from. Do not drop the inbox. There is a limit, and it answers the question "how big can this get?". You can produce almost unlimited images for one persona. You cannot do the same with conversations. That is the real reason a second persona multiplies your work and does not multiply your income. The limit of this business is a full inbox. You will not run out of content first. [How money is made on Fanvue](https://aiofmcity.com/blog/where-fanvue-money-is-made) covers the strategy behind this. ## What Fanvue lets you automate in the inbox The rule above says: tools for volume, your own voice for the moments that matter. Fanvue shows much more automation than that in its own API docs. Its General Terms set an outer limit, but they do not say where it is. So you set the working limit yourself. > Fanvue's own API documentation walks through building a chatbot that finds every conversation with unread messages, reads the recent history, drafts a reply and sends it — runnable once or on a schedule such as a cron job, to keep conversations going without a human present. The documentation is explicit that the API hands a developer the conversation and lets them send a reply, but "does not write the reply for you" — the generation step is the operator's own implementation, not Fanvue's. > > — Source: Fanvue API Documentation — Building an Auto-Reply Chatbot, read 2026-09-09 — https://api.fanvue.com/docs/tutorials/example-chatbots Fanvue wrote that tutorial and publishes it. Its General Terms still forbid accounts "controlled by other automated means", and no Fanvue page says when a reply bot crosses that line. [Fanvue AI chatbot rules](https://aiofmcity.com/blog/fanvue-ai-chatbot-rules) quotes both. A fully automatic inbox is also the wrong default. Fanvue's own messaging guidance gives the reason: the fans worth the most are the ones most likely to notice. A reply tool is fine for volume while you still run the account. It is wrong for a first message, a large tip or a real question. The tool cannot tell these apart unless you build that rule into it yourself. What the law asks for depends on whose system talks to the fans. > Article 50(1) binds **providers**, not deployers: it requires whoever designs and supplies an AI system that interacts directly with people to build in a disclosure that they are an AI, unless that is already obvious from the circumstances. > > — Source: EU AI Act Art. 50(1), read 2026-09-09 — https://artificialintelligenceact.eu/article/50/ If you use Fanvue's own reply and rewrite tools, Fanvue is the one supplying the system. An account that is already labelled as an AI creator also has a strong case that the AI is obvious from the context. If you build your own bot on the API, the position changes, because you built the system that talks to fans. [Fanvue's AI chatbot rules](https://aiofmcity.com/blog/fanvue-ai-chatbot-rules) explains where the line is. Read that article before you connect anything to the inbox. The other half is sorting your fans into lists. Fanvue already does most of that for you. > Fanvue pre-populates lists for subscribers, followers, non-renewing fans, auto-renewing fans, expired subscribers, free-trial subscribers, and fans who have spent more than $50. It notes that the $50 threshold is a fairly low bar, and suggests building a custom higher-spend list and excluding that group from mass messages. > > — Source: Fanvue Help — Mass Messaging: Driving Revenue Without Spamming, read 2026-09-09 — https://help.fanvue.com/en/articles/16237736-mass-messaging-driving-revenue-without-spamming Fanvue's own pages disagree on that last tip. Its help page on custom lists, updated on 22 September 2026, says you cannot build your own lists yet: > Asked "Can I create custom lists?", Fanvue answers: "Not at the moment, but this feature has been requested A LOT and we’ve listened! It’s in progress and we hope to release it really soon." > > — Source: Fanvue Help — Can I create custom lists?, read 2026-10-01 — https://help.fanvue.com/en/articles/7860590-can-i-create-custom-lists So for now, work with the seven ready-made lists. Treat the whole "Spent more than $50" list as your top fans, and leave all of it out of mass messages. With these lists, your daily inbox slot becomes a short routine: 1. Answer new subscribers first. 2. Then answer everyone on the "Spent more than $50" list. 3. Then answer real questions. 4. Send broadcasts only to lists that leave out the people a broadcast would put off. [How money is made on Fanvue](https://aiofmcity.com/blog/where-fanvue-money-is-made) explains why the list of people you leave out matters more than the list of people you send to. ## Which Fanvue list gets which message Each list is a group of fans at a different stage with you. Send each list a message that fits its stage, and write to your top spenders one by one. Fanvue's mass messaging guide says what suits each list: > Subscribers: "point them at a new PPV drop or a themed Collection." Followers: "these fans haven't paid yet!", so use the list "to nudge them toward subscribing" with a teaser. Free Trial Subscribers: "remind them what they'd lose access to before the trial ends". Expired Subscribers / Non-Renewing Fans: "you can send a win-back message, since this is a place creators commonly underperform." Spent more than $50: "These sell best 1:1, not as a broadcast." > > — Source: Fanvue Help — Mass Messaging: Driving Revenue Without Spamming, read 2026-10-01 — https://help.fanvue.com/en/articles/16237736-mass-messaging-driving-revenue-without-spamming | List | What to send | How | | --- | --- | --- | | Subscribers | A new PPV item or a new Collection | One broadcast when it goes live | | Followers | A teaser of what subscribers get | Broadcast, no hard sell | | Free Trial Subscribers | A reminder of what they lose when the trial ends | Broadcast before the trial ends | | Expired Subscribers and Non-Renewing Fans | A win-back message, if you have a reason to send one | Broadcast with a clear reason | | Spent more than $50 | Personal messages and personal offers | One by one, never in a broadcast | How often you broadcast matters as much as what you send. Fanvue sets no limit, but it warns about sending too much: > "There's no official limit. If you're just starting out, broadcast only when you have something genuinely worth saying (a drop, an offer, a milestone), not on a fixed schedule." "Messaging too often risks mutes or unsubscribes." > > — Source: Fanvue Help — Mass Messaging: Driving Revenue Without Spamming, read 2026-10-01 — https://help.fanvue.com/en/articles/16237736-mass-messaging-driving-revenue-without-spamming Give every broadcast a reason you can say in one line: a new set, an offer or a milestone. If you cannot name the reason, do not send it. A fan who mutes you will also miss your next real offer. Your top fans need more of your time. Fanvue's pricing guide calls this VIP treatment: > Under "Give your top fans VIP treatment": "Use the 'Spent more than $50' list (see the Mass Messaging guide) to find them, and spend more time chatting with this group." "Give them first access to new content, or an occasional higher-priced, personal PPV item made just for that relationship." > > — Source: Fanvue Help — Subscription Pricing & Retention Guide, read 2026-10-01 — https://help.fanvue.com/en/articles/16233102-subscription-pricing-retention-guide These are the moments that matter from the rule above. Write these messages yourself. ## Running promotions on Fanvue: which offer keeps fans longest A subscription bundle keeps fans longest. Fanvue says only 1 in 8 bundle fans leave in the first month. About 7 in 10 fans from a time-limited discount leave in that same month. Free trials sit in between. You set up all three offers in one place: > Go to Creator Tools, then Promotions. "Create promotional bundle": select discount and duration. "Create promotional offer": set up a Free Trial or a First Month Discount offer. "Create new free trial link": select number of offers, expiry date and trial duration. > > — Source: Fanvue Help — How do I set up a promotion offer on my account?, read 2026-10-01 — https://help.fanvue.com/en/articles/7860747-how-do-i-set-up-a-promotion-offer-on-my-account Fanvue's promotions guide compares what happens to fans after each type of offer: > Subscription bundles: up to 50% off for a 3, 6, or 12-month subscription, paid upfront. "Subscription Bundles perform best: only 1 in 8 churn in the first month, and 4 in 10 are still subscribed after 90+ days." "Free trials and free trial links land in the middle: about 1 in 4 stick around past 90 days." "Time-limited discount promotions: churn fastest: about 7 in 10 leave within the first month, and only 1 in 5 stay past 90 days." > > — Source: Fanvue Help — Running Profitable Promotions, read 2026-10-01 — https://help.fanvue.com/en/articles/16237907-running-profitable-promotions Compare these numbers with the normal case. Fanvue publishes that too: > Roughly 4 in 10 subscriptions last beyond the first 30-day cycle (the fan renews at least once), and about 1 in 4 last 90 days or longer. > > — Source: Fanvue Help — Subscription Pricing & Retention Guide, read 2026-10-01 — https://help.fanvue.com/en/articles/16233102-subscription-pricing-retention-guide | How the fan joined | Still subscribed after 90+ days | Compared with all subscriptions | | --- | --- | --- | | Subscription bundle | 4 in 10 | Above the average of 1 in 4 | | Free trial or free trial link | About 1 in 4 | The same as the average | | Time-limited discount | 1 in 5 | Below the average | Keep one thing in mind with the bundle. A bundle fan has already paid for at least three months, so part of that number is the prepayment. Fanvue's advice in the same guide follows from these numbers: > "Default to bundles when you want to discount." Use time-limited discount promotions "sparingly and with a target", aimed at "a specific segment and moment: Expired Subscribers as a win-back, a launch, a seasonal push, rather than leaving one running constantly". "Expect free trials to convert a minority long-term." "Constant discounting signals your regular price isn't the "real" price, fans learn to wait for the next deal instead of subscribing now." > > — Source: Fanvue Help — Running Profitable Promotions, read 2026-10-01 — https://help.fanvue.com/en/articles/16237907-running-profitable-promotions That gives each offer one job: - **The bundle is your standing discount.** It can stay on all year. - **A time-limited discount is for one list at one moment.** Give it an end date before you launch it, and announce it with one mass message to that list. - **A free trial brings in traffic.** Plan for most trial fans to leave, and count the ones who stay as the result. The promotions guide gives no money figure per offer or per campaign. So keep your own record: how many fans took each offer, and how many were still there after 30 and 90 days. ## Keeping subscribers past the first month Post regularly, talk to your fans, and give renewing fans a small reward. Those are Fanvue's three tips for keeping subscribers. > Tip 1: "Stay Consistent with Regular Content Uploads." Tip 2: "Build Personal Connections with Your Fans", including "Respond to comments and start conversations in DMs." Tip 3: "Reward Renewing Subscribers with Exclusive Content", for example free access to a premium video or a discount on pay-to-view content for re-subscribers. "Send reminder messages a few days before their renewal date to promote the incentive and increase conversions." > > — Source: Fanvue Help — Tips to Retain Subscribers, read 2026-10-01 — https://help.fanvue.com/en/articles/9527118-tips-to-retain-subscribers Renewals are a small part of income at the start, and a bigger one later: > "In a creator's first two months, renewals make up only around 6% of earnings, so don't worry if it doesn't feel very relevant on day one. It matters more as your subscriber base matures." > > — Source: Fanvue Help — Subscription Pricing & Retention Guide, read 2026-10-01 — https://help.fanvue.com/en/articles/16233102-subscription-pricing-retention-guide Fanvue also has an automated message that fires when a fan renews: > Automated messages "fire on a trigger (a fan follows, subscribes, replies, etc.)". "Subscription Renewed: Thank your most loyal subscribers for being there another month." > > — Source: Fanvue Help — Mass Messaging: Driving Revenue Without Spamming, read 2026-10-01 — https://help.fanvue.com/en/articles/16237736-mass-messaging-driving-revenue-without-spamming So set up the renewal routine once, early, and let it run: 1. Pick one reward for fans who renew, such as a free premium video. 2. Mention it in your captions. 3. Set up the Subscription Renewed message to thank fans who stay. 4. Remind fans a few days before their renewal date. ## Two personas is not twice the work, and not half again either The technical buckets scale well when a second persona shares the pipeline. You use one set of scripts, one scheduler and one backup routine for both. Correspondence does not scale at all, because two audiences mean two sets of conversations. This is the bucket that decides whether you can keep a second persona going. Add a second persona only when you know how much correspondence the first one needs, and that amount is stable. ### What did not work: Automating the selection step to hit a schedule I wanted to post every day. So I let a scoring script pick the day's images from each batch. It reliably chose technically clean images. But it could not see two things. Three of the images in one week were near-duplicates of poses from the week before. And one image had a lighting shift that made the face look like a different person. Within a fortnight the account looked automated. I had solved the consistency problem in training, and now it was back at the selection step. Selection went back to being a human decision, and I cut the schedule to what a human could select for. That was the right pace all along. ## A weekly checklist for one persona Everything above fits into one routine: one batch session a week, one message slot every day, and a short review once a month. Use this list to check your week. It has no hours on purpose, because your hours depend on your audience. **Did you do this week's work?** - [ ] Every day: answer messages in a fixed slot. New subscribers first, then the "Spent more than $50" list, then real questions. — Write first messages and replies to top fans yourself. Use AI drafts for the rest. - [ ] Every day: check that the scheduled Fanvue post went out. — Creator Tools, then Queue, shows what is planned for each day. - [ ] Once a week: generate and select in one batch, with the age-appearance check on every image. — Make enough finished posts for the week, plus images for paid messages. - [ ] Once a week: fill the Fanvue Queue for the next seven days and schedule your reels. — Every other day on Fanvue and two or three reels a week is the minimum plan. - [ ] Once a week: send a broadcast only if you can name its reason in one line. — Leave the "Spent more than $50" list out of every broadcast. - [ ] Once a week: back up the training set, the model, the image library and the captions. — Keep the manifests intact in the backup. - [ ] Whenever you add or change a pipeline step: inspect one output file for the provenance manifest. - [ ] Once a month: write down how many fans renewed, and how many fans from each offer are still there. — Fanvue does not publish money per promotion. Your own numbers are the only source. - [ ] Once a month: decide on one targeted offer, or none. — For example, a time-limited discount for Expired Subscribers, with an end date. If a week goes badly, keep the daily message slot and drop the batch. You have a library of images to post from. ## Keep your own copy of everything Keep a copy of everything the operation depends on in your own storage: the training set, the model, the image library with its manifests intact, the caption archive and the scheduling plan. You should be able to restore all of it yourself. You do not know that a platform will disappear, but one might. [Risk, bans and recovery](https://aiofmcity.com/guides/risk-bans-and-recovery) is about the day that happens. Start this week with the checklist above. The articles in [Operations](https://aiofmcity.com/topics/operations) go deeper on the individual steps. ## Frequently asked ### What should I automate first? Start with the tasks you do most often that need no judgement: file handling, renaming, format conversion, scheduling and backups. Most people want to automate generation and captions first. That is the wrong place to start, because your judgement adds the most there. ### Can the age-appearance check be automated? No. This check decides whether an image is published at all, and it is your responsibility. You cannot hand it to a script. In Germany, the law judges the impression an image gives. If you have any doubt, do not publish the image. ### How do I keep provenance data intact through the pipeline? Find out which of your steps re-encode files. Check the output of each one with a manifest inspector. Most optimisers remove metadata by default. Use a step that keeps the metadata, or a step that does not re-encode at all. ### How long before the workflow settles? Expect six to eight weeks for your first persona. In that time you need roughly double the weekly hours you will need later, because you are building and running the operation at the same time. Plan for that. It does not mean you are failing.