Create LinkedIn content with GPT-4 via Telegram bot & approval loop

AI LinkedIn Content Bot with Approval Loop

This n8n workflow transforms your Telegram messenger into a personal assistant for creating and publishing LinkedIn posts. You can simply send an idea as a text or voice message, collaboratively edit the AI's suggestion in a chat, and then publish the finished post directly to LinkedIn just by saying "Okay."

What You'll Need to Get Started

Before you can use this workflow, you'll need a few prerequisites set up. This workflow connects three different services, so you will need API credentials for each:

1. Telegram Bot API Key: You can get this by talking to the "BotFather" on Telegram. It will guide you through creating your new bot and provide you with the API token. New Chat with Telegram BotFather

2. OpenAI API Key: This is required for the "Speech to Text" and "AI Agent" nodes. You'll need an account with OpenAI to generate this key. OpenAI API Platform

3. Blotato API Key: This service is used to publish the final post to LinkedIn. You'll need a Blotato account and to connect your LinkedIn profile there to get the key. [Blotato platform for social media publishing]

Once you have these keys, you can add them to the corresponding credentials in your n8n instance.

How the Workflow Operates, Step-by-Step

Here is a detailed breakdown of how the workflow processes your request and handles the publishing.

1. Input & Initial Processing

This phase captures your idea and converts it into usable text.

2. AI Core & Iteration Loop

This is the central dialogue system where the AI drafts the content and engages in the feedback loop.

AI Agent System Prompt (Internal Instructions - English)

The agent operates under a strict prompt that dictates its behavior and formatting (found within the AI: Draft & Revise Post node):

> You are a LinkedIn Content Creator Agent for Telegram. > Keep the confirmation process, but change the output format as follows: > > Your Task > Analyze the user's message: > > * Topic > * Goal (e.g., reach, show expertise, recruiting, personal branding, leads) > * Target Audience > * Tonality (e.g., factual, personal, bold, inspiring) > > Create a LinkedIn post as ONE continuous text: > > * Strong hook in the first 1–2 lines. > * Clear main part with added value, story, example, or insight. > * Optional Call-to-Action (e.g., question to the community, invitation to exchange). > * Integrate hashtags at the end of the post (5–12 suitable hashtags, mix of niche + somewhat broader). > * Readable on LinkedIn: short paragraphs, emojis only sparingly. > > Present the suggestion to the user in the following format: > > Headline: Post Proposal: > Below that, the complete LinkedIn post (incl. hashtags at the end in the same text). > > Ask for feedback: > For example: > "Any changes? (Tone, length, formality, personal vs. professional, more technical content, different hashtags?)" > > If the user requests changes: > Adjust the post specifically based on the feedback. > Again, output only: > Post Proposal: > the revised complete post. > > If the user says “approved”, “ok”, “sounds good”, or similar: > Return exclusively this JSON, without additional text, without Markdown: > > json > { > "Post": "The final LinkedIn post as one text, including hashtags at the end" > } > > > Important: > > * Never output JSON before approval, only normal suggestion text. > * The final output after approval consists of only one field: Post.

3. Publishing & Status Check

Once approved, the workflow handles the publication and monitors the post's status in real-time.

4. Final Notification

🛠️ Personalizing Your Content Bot

The true power of this n8n workflow lies in its flexibility. You can easily modify key components to match your unique brand voice and technical preferences.

1. Tweak the Content Creator Prompt

The personality, tone, and formatting rules for your LinkedIn content are all defined in the System Prompt.

• Where to find it: Inside the AI: Draft & Revise Post node, under the System Message setting.
• What to personalize: Adjust the tone, change the formatting rules (e.g., number of hashtags, required emojis), or insert specific details about your industry or target audience.

2. Switch the AI Model or Provider

You can easily swap the language model used for generation.

• Where to find it: The OpenAI Chat Model node.
• What to personalize:
  • Model: Swap out the default model for a more powerful or faster alternative (e.g., gpt-4 family, or models from other providers if you change the node).
  • Provider: You can replace the entire Langchain block (including the AI Model and Window Buffer Memory nodes) with an equivalent block using a different provider's Chat/LLM node (e.g., Anthropic, Cohere, or Google Gemini), provided you set up the corresponding credentials and context flow.

3. Modify Publishing Behavior (Schedule vs. Post)

The final step is currently set to publish immediately, but you might prefer to schedule posts.

• Where to find it: The Create post with Blotato node.
• What to personalize:
  • Consult the Blotato documentation for alternative operations. Instead of choosing the "Create Post" operation (which often posts immediately), you can typically select a "Schedule Post" or "Add to Queue" operation within the Blotato node.
  • If scheduling, you will need to add a step (e.g., a Set node or another agent prompt) before publishing to calculate and pass a Scheduled Time parameter to the Blotato node.