# Boost LinkedIn Automation: Creating Content with n8n and Google Gemini In the fast-paced world of social media, consistency is key. But for busy professionals and tech enthusiasts, maintaining a steady stream of high-quality LinkedIn posts can be a challenge. In this post, I'll walk you through how I automated my LinkedIn content creation using **n8n**, **Google Gemini**, and a **human-in-the-loop** workflow. This system allows me to generate engaging, platform-specific content from a simple topic, review it, add visual assets, and publish it—all without leaving the automation flow. ## The Workflow Overview The core of this automation is an **n8n** workflow that acts as my social media assistant. Here’s the high-level logic: 1. **Idea Injection**: I provide a simple "Topic" or "Title" via a web form. 2. **AI Processing**: An AI Agent (powered by Google Gemini) researches the topic and drafts a professional LinkedIn post, complete with hashtags and a call-to-action. 3. **Human Review**: The workflow pauses and presents the generated text to me. I can review it and uploading a relevant image. 4. **Publishing**: Once approved, the workflow automatically posts the text and image to my LinkedIn profile. ## The Automation Architecture Here is the visual representation of the flow: ```mermaid graph TD Start((Start)) --> Form_Trigger[Form Trigger: Input Topic] Form_Trigger --> Split_Input[Split Input] Split_Input --> AI_Data_Prep[Data for AI] AI_Data_Prep --> AI_Agent[AI Agent - Gemini] subgraph "AI Processing" AI_Agent --> Gemini_Model[Google Gemini Model] AI_Agent --> Parser[Structured Output Parser] end AI_Agent --> Aggregate[Aggregate Results] Aggregate --> Human_Loop_Form[Form: Review & Upload Image] Human_Loop_Form --> Process_Image[Process Image Data] Process_Image --> LinkedIn_Node[Publish to LinkedIn] LinkedIn_Node --> Done((End / Confirmation)) style Start fill:#f9f,stroke:#333,stroke-width:2px style AI_Agent fill:#bbf,stroke:#333,stroke-width:2px style Human_Loop_Form fill:#bfb,stroke:#333,stroke-width:2px style LinkedIn_Node fill:#f96,stroke:#333,stroke-width:2px ``` ### The n8n Canvas View Here's what the actual workflow looks like in n8n: ![](https://cdn.hashnode.com/res/hashnode/image/upload/v1770532209957/798cffe2-499e-45ca-93fa-11bfdfc4027d.png align="center") The workflow is organized into 4 distinct sections (marked by sticky notes): 1. **Get the Data for Social Media Post using the Web Form** (Yellow) 2. **AI Agent will do its Job** (Blue) 3. **Get Image to Be Published** (Light Blue) 4. **Publish on Social Media** (Yellow) → **Confirmation** (Green) --- ## Step-by-Step Node-by-Node Implementation Guide Now, let me walk you through building this workflow from scratch. Each subsection represents a node you need to add to your n8n canvas. ### Section 1: Get the Data for Social Media Post #### Node 1: **On form submission** (Form Trigger) * **Node Type**: `Form Trigger` * **Purpose**: This is the entry point of your workflow. It creates a web form to capture the post topic. * **Configuration**: * **Form Title**: "Social Media Content AI Agent" * **Form Description**: Add a helpful description explaining what the form does * **Form Fields**: * Field Name: `Post Title/Topic` * Field Type: `Text` * Placeholder: "Write a brief and clear title or main topic for the post" * **Authentication**: Basic Auth (optional, for security) * **Button Label**: "Continue to Image Upload" * **What happens**: When submitted, this triggers the workflow and passes the topic to the next node. #### Node 2: **Split Form Input** (Set Node) * **Node Type**: `Set` (Edit Fields) * **Purpose**: Extract and structure the data from the form submission. * **Configuration**: * **Mode**: Keep only specific fields * **Fields to Include**: `output.platform_posts.LinkedIn.post` * **Connect to**: Output from `On form submission` #### Node 3: **Split Data** (Set Node) * **Node Type**: `Set` (Edit Fields) * **Purpose**: Further refine the data structure. * **Configuration**: Same as Split Form Input * **Connect to**: Output from `Split Form Input` #### Node 4: **Data for AI** (Set Node) * **Node Type**: `Set` (Edit Fields) * **Purpose**: Prepare the final input data for the AI Agent. * **Configuration**: * **Assignments**: * `Post Title/Topic` = `{{ $('On form submission').item.json['Post Title/Topic'] }}` * `formMode` = `{{ $('On form submission').item.json.formMode }}` * **Connect to**: Output from `Split Data` --- ### Section 2: AI Agent Will Do Its Job #### Node 5: **AI Agent** (AI Agent Node) * **Node Type**: `@n8n/n8n-nodes-langchain.agent` * **Purpose**: The brain of the operation. Generates platform-specific content. * **Configuration**: * **Prompt Type**: Define * **Prompt**: A detailed system prompt that: * Defines the AI's role as a content creator for your brand * Specifies platform-specific rules (LinkedIn, Instagram, Facebook, Twitter) * Includes hashtag strategies * References the input data: `{{ $json['Post Title/Topic'] }}` * **Has Output Parser**: ✓ Enabled * **Connect to**: Output from `Data for AI` * **Sub-nodes to connect**: * **Google Gemini Chat Model** (Language Model) * **Structured Output Parser** (Output Parser) #### Node 6: **Google Gemini Chat Model** (Sub-node) * **Node Type**: `@n8n/n8n-nodes-langchain.lmChatGoogleGemini` * **Purpose**: The actual AI model that processes the prompt. * **Configuration**: * **Model Name**: `models/gemini-3-flash-preview` (or your preferred Gemini model) * **Credentials**: Google Gemini API credentials * **Connect to**: `AI Agent` (via Language Model connection) #### Node 7: **Structured Output Parser** (Sub-node) * **Node Type**: `@n8n/n8n-nodes-langchain.outputParserStructured` * **Purpose**: Ensures the AI returns properly formatted JSON. * **Configuration**: * **Schema Type**: Manual * **Input Schema**: A JSON schema defining the structure: ```json { "type": "object", "properties": { "platform_posts": { "type": "object", "properties": { "LinkedIn": { "type": "object", "properties": { "post": {"type": "string"}, "hashtags": {"type": "array"}, "call_to_action": {"type": "string"} } }, "Twitter": {...}, "Facebook": {...} } } } } ``` * **Connect to**: `AI Agent` (via Output Parser connection) #### Node 8: **Aggregate** (Aggregate Node) * **Node Type**: `Aggregate` * **Purpose**: Combines all the AI output into a single item for easier reference. * **Configuration**: * **Aggregate**: All Item Data * **Connect to**: Output from `AI Agent` --- ### Section 3: Get Image to Be Published #### Node 9: **Upload Image** (Form Node) * **Node Type**: `Form` * **Purpose**: Pauses the workflow to let you review the AI-generated text and upload an image. * **Configuration**: * **Operation**: Wait for Form Submission * **Form Title**: "Review the Text" * **Form Description**: Display the AI-generated text using expressions: ```plaintext LinkedIn: {{ $json.data[0].output.platform_posts.LinkedIn.post }} Twitter: {{ $json.data[0].output.platform_posts.Twitter.post }} ``` * **Form Fields**: * Field Label: `image` * Field Type: `File` * Accepted File Types: `.jpg` (or `.png`, `.jpeg`) * Required: ✓ Yes * **Button Label**: "Proceed to Next Step" * **Connect to**: Output from `Aggregate` #### Node 10: **Nest Top Meta** (Set Node) * **Node Type**: `Set` (Edit Fields) * **Purpose**: Preserve all form data including the binary image. * **Configuration**: * **Assignments**: * Name: `metaTop` * Type: Object * Value: `{{ $json }}` * **Options**: Include Binary ✓ * **Connect to**: Output from `Upload Image` #### Node 11: **Rename Image Binary Top Image** (Code Node) * **Node Type**: `Code` * **Purpose**: Rename the binary data field for LinkedIn compatibility. * **Configuration**: * **Mode**: Run Once for Each Item * **JavaScript Code**: ```javascript $input.item.binary.top = $input.item.binary.data; delete $input.item.binary.data; return $input.item; ``` * **Connect to**: Output from `Nest Top Meta` --- ### Section 4: Publish on Social Media #### Node 12: **Publish to LinkedIn** (LinkedIn Node) * **Node Type**: `LinkedIn` * **Purpose**: Posts the content to your LinkedIn profile. * **Configuration**: * **Resource**: Post * **Operation**: Create * **Person**: Your LinkedIn Person URN (e.g., `CryRqQfSsC`) * **Text**: ```plaintext {{ $('AI Agent').item.json.output.platform_posts.LinkedIn.post }} {{ $('Aggregate').item.json.data[0].output.platform_posts.LinkedIn.call_to_action }} ``` * **Share Media Category**: IMAGE * **Binary Property Name**: `image` * **Credentials**: LinkedIn OAuth2 credentials * **Connect to**: Output from `Rename Image Binary Top Image` #### Node 13: **X** (Twitter Node) - Optional * **Node Type**: `Twitter` * **Purpose**: Posts to Twitter/X. * **Configuration**: * **Text**: `{{ $('Aggregate').item.json.data[0].output.platform_posts.Twitter.post }}` * **Credentials**: Twitter OAuth2 credentials * **Connect to**: Can run in parallel with LinkedIn #### Node 14: **Edit Fields** (Set Node) * **Node Type**: `Set` (Edit Fields) * **Purpose**: Extract Twitter post ID for confirmation. * **Configuration**: * **Assignments**: * Name: `edit_history_tweet_ids` * Type: Array * Value: `{{ $json.edit_history_tweet_ids }}` * **Connect to**: Output from `X` (if using Twitter) #### Node 15: **Merge1** (Merge Node) * **Node Type**: `Merge` * **Purpose**: Combines outputs from LinkedIn (and optionally Twitter) before final confirmation. * **Configuration**: Default merge settings * **Connect to**: * Input 1: Output from `Publish to LinkedIn` * Input 2: Any other social media nodes --- ### Section 5: Confirmation that Post is Published #### Node 16: **Form** (Form Node - Completion) * **Node Type**: `Form` * **Purpose**: Shows a success message with links to the published posts. * **Configuration**: * **Operation**: Completion * **Completion Title**: "Thanks" * **Completion Message**: ```plaintext Your post has successfully been submitted to LinkedIn. LinkedIn: https://www.linkedin.com/feed/update/{{ $('Publish to LinkedIn').item.json.urn }} Thanks, AI Agent ``` * **Form Title**: "AI Agent (Job Done)" * **Connect to**: Output from `Merge1` --- ## Deep Dive: How It Works ### 1\. The Trigger (Command Center) Everything starts with an **n8n Form Trigger**. Instead of manually logging into LinkedIn, I simply open a private URL hosted by my n8n instance. This form asks for: * **Post Title/Topic**: e.g., "The Future of Open Source AI". * **Keywords**: Optional context to guide the AI. ### 2\. The Brain: Google Gemini AI Agent This is where the magic happens. The input is passed to an **AI Agent** node connected to the **Google Gemini Chat Model**. I've configured the prompt to act as a "Content Creation AI" for my brand. It follows specific rules: * **Tone**: Professional, insightful, and value-driven. * **Structure**: 3-4 sentences, optimized for engagement. * **Hashtags**: A mix of general tech tags (#Innovation, #AI) and niche ones. The agent doesn't just output text; it uses a **Structured Output Parser** to return a clean JSON object containing the post text, hashtags, and even suggestions for other platforms like Twitter and Facebook. ### 3\. Human-in-the-Loop (The "Wait" Node) Automation is great, but I don't want to blindly post AI-generated content. I need final approval. The flow uses a second **n8n Form** node (titled "Upload Image") in the middle of the execution. * The workflow **pauses** here. * It displays the AI-generated text for me to read. * It asks me to upload the final image or creative asset to go with the post. * Once I hit "Proceed," the workflow resumes. This step is crucial. It combines the speed of AI with the quality control of a human. ### 4\. The Publisher Finally, the workflow takes the text approved in the previous step and the image I just uploaded. It formats the binary data and sends it to the **LinkedIn** node, which uses the LinkedIn API to create a "Share" on my personal profile. ## Security: Keeping Credentials Safe One of the most important aspects of sharing or backing up automation flows is **security**. * **Credential Separation**: n8n separates the *workflow logic* (the nodes and connections) from the *credentials* (API keys and passwords). * **No Hardcoded Secrets**: In the workflow JSON file, you will never see my actual API keys. You will only see references like `linkedInOAuth2Api` or `googlePalmApi`. * **Environment Variables**: For sensitive data that might be needed inside expressions, I use n8n environment variables rather than typing them directly into the node parameters. When you import my flow, n8n will ask you to set up your *own* credentials for LinkedIn and Google Gemini. Your secrets stay on your server, and mine stay on mine. ## Conclusion This workflow has transformed how I manage my professional presence. By automating the "drafting" phase and streamlining the "publishing" phase, I save hours of time while maintaining high content quality. If you want to try this yourself, you'll need: 1. A self-hosted or cloud n8n instance. 2. A Google Cloud Console project with the Gemini API enabled. 3. A LinkedIn App for API access. Happy Automating!