Social post and blog writer with web research

Social post and blog writer with web research

An n8n workflow. Send a request to its webhook and it researches the topic, then writes either a social post or a blog draft
and logs it in a Google Sheet with an image link.

What is inside

• workflow.json: the workflow as n8n exports it, 28 nodes, five of them sticky notes that label the sections. It holds no credential.
• README.md: these notes.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• An OpenAI API key.
• A Tavily API key.
• An Apify account and API token.
• A Google account, and a spreadsheet whose first row holds five headings: Platform, Title, Caption, Image and Date.
• An image workflow of your own, described below. The file does not include it.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add an OpenAI credential and select it in every OpenAI node and every chat model node.
3. Open HTTP Request. In its Authorization header, replace <YOUR_TAVILY_API_KEY> after Bearer with your Tavily key.
4. Open HTTP Request1. In its address, put your Apify token where it says <YOUR_APIFY_TOKEN>. In its body, swap the four
   YOUR_INSTAGRAM_ACCOUNT names for the accounts you want to learn from.
5. Add a Google Sheets credential. In Google Sheets and Google Sheets1, put your spreadsheet ID where it says
   YOUR_SHEET_ID and pick the sheet.
6. Build your image workflow and pick it in Image gen and Image gen1.
7. Save, run it once from the webhook's test address, then press Publish at the top of the canvas (or Shift+P).

The image workflow

Image gen and Image gen1 are tool nodes that call a sub-workflow. The original pointed at one in its author's own n8n, so
here they arrive empty. Your sub-workflow receives one field and returns the address of the picture it made:

{
  "description": "<what the picture should show>"
}

Any image service works. Make the sub-workflow end with a node that outputs the link, so the agent can write it into the sheet.

The request

Send a POST request to the webhook with a JSON body that holds a message:

{
  "message": "I want an Instagram post about composting at home"
}

The caller waits for the whole run, which includes a synchronous Apify run, so give it a generous timeout.

How a run goes

1. query gen asks gpt-4.1-mini for one search query, given your message and the current date.
2. HTTP Request sends that query to Tavily and asks for one result. HTTP Request3 fetches that page, Markdown turns it
   into text and OpenAI summarizes it.
3. HTTP Request1 runs Apify's Instagram post scraper for your accounts, with a limit of 20 results. Aggregate keeps each
   caption, comment count and first comment, and OpenAI1 summarizes them.
4. Summarise merges the two summaries into one brief.
5. Is a blog post? answers blog or not, and If tests for the word blog.
6. A blog goes to Planning agent, then Writing agent, then OpenAI4, which picks a title and an image prompt and can call
   the image tool and the sheet. A social post goes to AI Agent, which makes the image first, writes a caption with hashtags
   and a title, and adds a row to the sheet.
7. Respond to Webhook or Respond to Webhook1 returns the result to the caller.

The sheet

Each run appends one row. For a social post, Platform is the platform you named. For a blog, Platform is Blog and Caption
holds the whole post.

| Column | Holds |
|---|---|
| Platform | The platform you named, or Blog |
| Title | The title the agent wrote |
| Caption | The caption with hashtags, or the whole blog post |
| Image | The link your image workflow returned |
| Date | The time of the run |

Change it

• Voice and format: edit the system messages in AI Agent, Planning agent and Writing agent.
• Research: change the accounts in HTTP Request1, or ask Tavily for more results in HTTP Request.
• Models: every model node names gpt-4.1-mini, and any chat model can take its place.
• Output: add a node after a Respond to Webhook node to send the piece to a scheduler or a draft folder.

Good to know

• Each run costs money on OpenAI, Tavily, Apify and your image service.
• Anyone who learns the webhook address can start a run on your keys. Add authentication to the webhook node or keep the address private.
• The research is one article and the posts of the accounts you list.
• Is a blog post?, Planning agent and AI Agent read $json.body.message, yet each runs right after an OpenAI node, whose output has no body. If a
  first run shows your request missing there, point them at $('Webhook').item.json.body.message.
• If tests for the exact word blog in lowercase, so any other answer from the check sends the request down the social path.
• Apify reads Instagram on your behalf, so check Apify's and Instagram's terms for the accounts you pick.
• Nothing checks for repeats, so every run adds a row.
• Your request and the research go to OpenAI, Tavily and Apify, and the results go to Google Sheets.
• The original held its keys inside the nodes. They are placeholders here, and the Tavily key and the Apify token still sit in
  node parameters, so do not share the workflow after you fill them in.