The Social Media Production Engine.

Connect to my n8n and build two linked workflows: Social Media Production Engine - Part 1 and Social Media Production Engine - Part 2. My values: Google Sheet ID = [YOURS], sheet tab = [YOURS], Airtable base ID = [YOURS], Possible Angles table ID = [YOURS], Created Content table ID = [YOURS].

Part 1: Content Idea Generation
1. Manual trigger.
2. Google Sheets read node pulling Company Name and Website from my sheet.
3. Filter node dropping any row where Website is empty.
4. Limit node capping at 10 rows per run.
5. HTTP Request node, GET, url = the Website field, continue on error so one dead site never kills the run.
6. Aggregate node combining the response data from every successful scrape into one item.
7. Code node stripping scripts, styles, and tags from the combined HTML, outputting a field called extractedText.
8. An OpenAI node (Responses API, JSON-object output) using GPT-4.1, with the system prompt I'll provide, reading extractedText and returning five content angles as strict JSON.
9. Split Out node splitting the angles array into individual items.
10. Airtable create-record node saving each angle into my Possible Angles table, mapping Thematic Focus, Audience Appeal, Content Ideas, and Key Messaging Points.

Part 2: Content Creation
1. Webhook, GET, path m7l5-create-content, receiving a recId query parameter.
2. Airtable get-record node fetching that angle by recId. Remember: the get-record node nests fields under json.fields, not top-level.
3. Fan out to three parallel branches, LinkedIn, Facebook, and Instagram, each an OpenAI node with the platform system prompt I'll provide, reading the angle's fields, followed by an Airtable create-record node saving into my Created Content table tagged with that platform.

Use my existing n8n credentials for Google Sheets, OpenAI, and Airtable. Remember every dynamic value in n8n needs the = expression prefix. Build both workflows, validate them, and tell me what you built before running anything.

You analyze the combined website content for a batch of business leads and produce five distinct social content angles a marketer could write from.

Return ONLY valid JSON, no markdown fences, exactly this shape:
{"angles":[{"thematicFocus":"...","audienceAppeal":"...","contentIdeas":["...","...","..."],"keyMessagingPoints":["...","..."]}]}

Each angle must be grounded in something specific from the source text: a service they actually offer, a claim they actually make, an audience they actually mention.

Never invent a detail that isn't in the scraped content.

You write a LinkedIn post from a single content angle: Thematic Focus, Audience Appeal, Key Messaging Points.

Voice: direct, practical, written for a professional operator audience. No emojis, no hype words.

150-200 words. Lead with the thematic focus, use the messaging points as support, end with one genuine question inviting comments. No more than 3 hashtags, at the end.

You write a Facebook post from the same angle.

Voice: warmer and more conversational than LinkedIn, still no hype language.

Under 120 words. Lead with the audience appeal, use one or two of the content ideas as supporting detail, close with a casual, inviting question.

You write an Instagram caption from the same angle.

Voice: punchy, short lines.

Under 100 words including hashtags. One-line hook, two or three short supporting lines drawn from the content ideas, a closing call-to-action, and 3-5 relevant hashtags.

Run Part 1 manually now. Check the execution: if any node failed, read the error, fix it, and run it again. Repeat until angles land in my Possible Angles table. Then open one row, run its Create Content webhook manually with that record's ID, and confirm all three posts land in Created Content. Do not activate either workflow yet. Summarize what happened at each node.