Chat case
AI for social media posts: one voice brief, three posts for Instagram, LinkedIn and a Telegram channel

Social media posts mostly go unwritten for lack of time, not lack of things to say. We recorded the brief of a small coffee roastery as a one-minute voice note and sent it to our test agent. 26 seconds later it returned an Instagram caption, a LinkedIn post and a post for a Telegram channel. Three more messages gave each platform its own shape, applied a fix by voice to all three posts at once and produced a two-line version to pin. Below are the real replies, the cost and the requests to copy.

The short answer
Record your brief as a voice note and send it to the agent: what you are launching, when, what it costs, how people sign up and who reads each platform. The agent transcribes the recording and writes the posts itself. In our test a one-minute voice note came back 26 seconds later as an Instagram caption, a LinkedIn post and a post for a Telegram channel.
Then the conversation carries on in the same chat. One line, "each one in the style and length that fits the platform", turns the drafts into three clearly different texts. A change can go in as another voice note, and the agent updates all three posts at once.
That is what an AI social media post generator looks like in practice: not a box that says "enter your topic", but someone who listens to the brief, keeps the facts straight and writes for specific readers. The agent starts in Telegram, which is also the easiest place to record a voice note between two other jobs.
The test: a coffee roastery and a one-minute brief
We made the business up. Hannah runs North Roast, a small coffee roastery. On Saturday 7 November she launches Ember, a winter blend: Ethiopia and Colombia, medium-dark, cherry and dark chocolate in the cup. A 250 g bag costs $16. The same day there is a free cupping at the roastery with twelve places. The second piece of news: North Roast now does office subscriptions, one kilo every two weeks, ground for the office machine, first delivery free. Tom roasts every Thursday.
Hannah has three platforms and three different audiences. Instagram is regulars and people from the neighbourhood. LinkedIn is office managers. The Telegram channel is her own people, about 1,200 of them.
She did not type the brief. She said it in one voice note of 59 seconds, the way she would tell it to a barista behind the counter: "Okay, so, the brief for this week's posts", then "And the other thing", then "That's it". We voiced the brief with a synthetic voice and on 31 October 2026 sent it to our test agent as a Telegram voice note. The roastery and the people are invented. The agent's replies are real and appear as they arrived.
Step 1. A voice note and not a word of text
Hannah sent the voice note on its own. No caption, no request.
The agent understood from the recording alone that this was a brief for posts and did not ask what to do with it. 26 seconds later there were three drafts, each one labelled with its audience: "regulars and the neighborhood", "office managers", "our own people (1,200)".
It also made a choice nobody asked for. The cupping is in the Instagram and Telegram posts but not on LinkedIn, and the agent said why: it kept that post on one message, the subscription, and offered to add the cupping as a team outing. The second open point was the price of the subscription. It was missing from the brief, so the agent left "message me" in place of a number and asked whether to keep it that way.
Every figure from the voice note is there: 7 November, $16 for 250 g, 11:00 to 13:00, twelve places, one kilo every two weeks, roasting on Thursdays.
Step 2. Three platforms, three different readers
Now Hannah introduced herself, named the platforms and asked for each post in the style and length of its platform.
Seventeen seconds, and the drafts became three texts with three different jobs.
The Instagram caption opens with three words, "Ember is coming.", then taste, price and date, the invitation to the cupping and a last line that sells the cupping as a tasting before buying: "Come taste it before you commit to a bag." Four hashtags at the end, including the roastery's name.
The LinkedIn post talks to someone who has to solve the office coffee question. It starts from the problem, "it's stale, and nobody is really paying attention to it", answers it with Thursday roasting and the subscription terms, and ends with the outcome a manager cares about: "a fresh, consistent cup for the team without anyone having to manage it." No hashtags, no emojis, signed as a person: "message me".
The Telegram post is written for people who already buy from the roastery. Everything is there, the blend, the cupping and the subscription, in short paragraphs, with "reply here" as the way to sign up. The agent added one idea of its own for the subscription: "split it with a neighbour". If that is not something you offer, cut the line before posting.
Step 3. A fix by voice, all three posts at once
A minute later the cupping time had to move: there is a market on Saturday. The number of places went up to fifteen. And Hannah wanted LinkedIn without emojis. She said all of it in a second voice note.
Ten seconds. The agent moved the time and changed the number of places in the Instagram and Telegram posts and left everything else exactly as it was. On the emojis it answered in one line: LinkedIn had none to begin with. It also remembered the open question from step 1 and asked for the subscription price once more.
Fixing by voice is handy precisely when changes arrive on the move: at the roaster, in the car, between two customers. You do not need to open three drafts and look for "11:00" in each of them.
Step 4. Three first lines and a version to pin
On Instagram everyone sees the first line and only some tap "more". Hannah asked for options to choose from and a short version of the Telegram post to pin in the channel.
Three different openings: the news, a story, the taste. The agent did not stop at the list. It picked the first one and explained why, then said when each of the other two would be the better choice. The second line mentions a blend the roastery has "been sitting on since August", a detail that was not in the brief; if you pick that line, make sure it is true.
The pin is two lines with the launch, the price, the date and the cupping. The agent left the subscription out on purpose and explained it: a pin is what a newcomer sees first, and newcomers are not buying office coffee.
What one voice note and three messages produced
About four minutes of chat, and Hannah has:
- an Instagram caption for neighbours and regulars, with hashtags;
- a LinkedIn post for office managers, built around the subscription;
- a post for the Telegram channel with all the details;
- all three with the new time and number of places after one fix by voice;
- three first lines with a recommendation;
- a two-line version to pin.
If you need other kinds of text, a payment reminder to a client, a long message cut short or a translation, we have a separate test: AI writing assistant in Telegram: five text jobs, from a payment reminder to a post in Spanish. The agent also draws the picture for the launch post in the same chat; we tested that in Free AI photo generator right in your chat: five photo jobs we tested and what the free plan covers.
What to say in a voice brief
A minute of speech is roughly 130 to 150 words. That is enough for everything the agent needs, as long as you say it in order. We built Hannah's brief this way, and not one fact got lost on the way into the posts.
- What is happening. A launch, a new product, an offer, an event. One sentence: "On Saturday we launch our winter blend."
- The facts with numbers. Date, time, price, weight, number of places. Say the numbers the way you would say them out loud: "a 250 gram bag is sixteen dollars."
- What the reader should do. Sign up by message, come over, subscribe. Without this a post ends nicely and leads nowhere.
- The second piece of news, if there is one. The agent decides where it goes first and where it only gets a line, as it did with the office subscription.
- Who reads each platform. The most important point. "Instagram is the neighbourhood, LinkedIn is office managers, the channel is our own people" changes the length, the first line and the order of the news.
There is no need to edit out "okay, so" and the pauses. The agent listens for meaning, not diction. When something is missing, such as a price, it says so in its reply and asks.
Voice also helps because you can record a brief where you could never write a post: on the way to work, at the roaster, while a batch cools down. The answer waits in the chat, and you can come back to it in the evening and ask for changes in text.
Habits that make posts sound the same
All of them are on the side of whoever writes the brief, and all of them are easy to avoid.
Not saying who reads where. "Write a post for three social networks" gives you one text cut three ways. "Instagram is the neighbourhood, LinkedIn is offices, the channel is our own people" gives you three posts with a different main idea each.
Keeping the facts in your head. The agent only knows what you told it. Price, date, time, number of places and how to sign up have to be in the brief, or the post will say "soon" where the facts should be.
Fixing the posts one at a time. Say the change once and ask to update all three. That way the time does not drift apart between platforms.
Asking to "make it more engaging". Say what exactly: shorter, a different first line, no emojis, a question at the end. The agent does exactly that and leaves the rest alone.
Posting without one read. Now and then the agent adds a sensible detail of its own, like "split it with a neighbour" or a blend kept "since August". Read the post before it goes out and keep only what you actually do.
What it costs
We read the balance in the agent's status line after each message. The voice brief with the first drafts cost 5 . The three platform posts and the fix by voice together cost another 5 , the first lines and the pin 2 . With one extra "update all three" along the way, the whole conversation came to 13 .
Transcribing a voice note costs 1 per minute, and text work on the default model costs 1 per 20,000 tokens. The Free plan gives 200 in the first month and 100 every month after, with no card. Pro costs $13.90 a month for 400 , or $1.39 for the first month if you subscribe within 7 days of signing up, and adds recurring reminders, scheduled tasks and top-ups. All rates are on the Pricing page.
Your messages stay in your agent's isolated container. To answer, the agent sends requests to the model: on the default model they go through DeepSeek's infrastructure in the People's Republic of China, which you confirm during onboarding, and on Pro you can attach your own Claude or Codex, so requests go to that provider under your own account.
If you record whole meetings and not just briefs, the same chat turns the recording into minutes: AI meeting minutes: send the recording of a team call, get decisions, owners and deadlines. And if social posts are one of several jobs you want to hand over, start with AI agents for small business in plain words: three jobs from a flower shop's week, before and after.
Messages worth copying
That voice note is my brief. I'm [name], I run [business]. Write three posts from it: for [platform 1] for [who reads it], for [platform 2] for [who reads it] and for [platform 3]. Each one in the style and length that fits the platform.
Update all three posts with this: [what changed].
Give me three different first lines for the [platform] post so I can pick one, and tell me which one you'd take.
Write a two-line version of the [channel] post that I can pin.
Create your agent, record a one-minute brief and see which posts come back.
Try it with your own agent
Free, no card. The agent starts working in the chat straight away.
Create your AI agentQuestions
Can an AI write a social media post from a voice note?
Yes. Send the agent a voice note with your brief: what you are launching, when, what it costs and who reads each platform. The agent transcribes the recording itself. In our test a one-minute voice note from a coffee roastery became an Instagram caption, a LinkedIn post and a Telegram channel post in 26 seconds.
Won't the posts come out the same on every platform?
Not if your brief says who reads where. The agent opened Instagram with three words and a tasting invitation, built LinkedIn around the office subscription for managers and gave the Telegram channel every detail in short paragraphs, with “reply here” to sign up.
Will it mix up prices and dates?
It takes them from your brief. In our test every figure from the voice note made it into the posts, and when the price of the subscription was missing, the agent said so and asked instead of making one up. After a fix by voice it changed the time and the places everywhere. Still check the numbers against your brief before posting; it takes a minute.
Can I edit the posts by voice instead of typing?
Yes. We recorded thirteen seconds with two changes and a request for no emojis on LinkedIn. Ten seconds later the agent sent all three posts updated and noted that LinkedIn had no emojis to begin with.
How is this different from an AI post generator website?
A generator writes one text per prompt. The agent keeps the whole conversation: it remembers the brief, updates three posts at once, suggests first lines and says which one it would pick. It starts in Telegram and works in DMs, groups and forum topics.
How much does it cost?
Our whole conversation, from the voice note to the version to pin, cost 13 . The Free plan gives 200 in the first month and 100 every month after, with no card. Pro costs $13.90 a month for 400 .


