The Project Liberal content system is an automation pipeline connecting Notion (content storage), Google Chat (user interface), Make.com (orchestration), Zoho Flow (scheduling delays), Metricool (social publishing), Claude API (AI processing), WordPress (media hosting), and Google Sheets (configuration and state).

There are two organizations — Project Liberal Action and Project Liberal Foundation — each with its own Notion content database, Metricool social accounts, and Google Chat spaces. The infrastructure is mirrored: same automations, same schemas, separate data.

A third organization, Liberal Memes (History is Over), exists in the database layer (Organization Settings, People in Content relations, Scheduling History) but is excluded from the user-facing documentation and chatbot routing.

Six Make.com scenarios, three Notion automation sets, one Zoho Flow, and supporting services.

Three Notion automations run on both PL Action Content and PL Foundation Content (6 total, mirrored). Each sends a webhook with all existing properties.

Two automations on the People in Content database.

// Twitter handle extraction formula (representative)
if(
  empty( Trigger page . Twitter URL ),
  "",
  " (@" +
  replaceAll(
    replaceAll(
      Trigger page . Twitter URL,
      "https?://(www\\.)?(twitter\\.com|x\\.com)/",
      ""
    ),
    "/.*",
    ""
  ) +
  ")"
)

Trigger: Google Chat event

The main scenario. Receives all Google Chat messages and routes them through the AI pipeline. This is the largest scenario with ~80 modules across 9+ workflow routes.

Top-Level Routes

Route 3 Sub-Routes (Workflow Engine)

Intent Router → Sub-Route Mapping

The Intent Router (Claude Haiku, module 269) outputs a workflow value. The Make.com router's filter conditions map these to sub-routes:

"notion"            → 3.1.3 (search), 3.1.2 (edit), 3.1.5 (upload) — determined by Notion workflow prompt
"tweet creation"    → 3.1.6a
"graphic creation"  → 3.1.6b
"video short creation" → 3.1.6c
"permissions"       → 3.1.0
"button manager"    → 3.1.1
"database sync"     → 3.1.4
"notion user sync"  → 3.0 (re-uses provisioning route)
"timezone manager"  → 3.1.0 (handled within permissions route)
null + error_message → 3.1.8 (fallback — error sent as Chat message)

Trigger: Webhook (card button click)

Handles button clicks on Google Chat content cards. Receives: Notion page ID, action value, user secret code, database ID, message ID.

Flow

Routes

Trigger: Webhook (Notion "Submit New Content" automation)

Processes new content when Type is set. Generates Name, Tags, descriptions, thumbnails, transcripts, and people tags. This is the content enrichment pipeline.

Flow

Route 0 — File-Based Content (Graphics & Video)

Downloads file from File (public) → uploads to Google Drive → gets file metadata/thumbnail.

Route 1 — Tweets (Text Only)

Claude Haiku processes text content → generates Name, Tags, AI Summary → updates Notion page via API.

Route 2 — People Tagging (All Types)

Runs for all content after the type-specific route:

• Fetches the content page from Notion.
• Searches the People in Content database for matches.
• Aggregates results into text.
• Claude Haiku determines if any people match the content.
• If match found: updates existing People record with content relation. If new person: creates a new People in Content record, then updates the content page.

Trigger: Webhook (Notion "Schedule Automation")

Handles status changes that trigger scheduling. Receives when any status is changed on a content page. Processes "Schedule -" status values into actual Scheduled Date values.

Flow

Routes

Trigger: Webhook (Zoho Flow "Send_Webhook")

The final stage of the publishing pipeline. Receives a Notion page ID and Scheduled Date from Zoho Flow, fetches the full content, then publishes to all configured social networks. This is the most deeply nested scenario — ~50+ modules across multiple routers.

Flow

Per-Network Publishing

The router fans out into deeply nested sub-routes. Each content type × organization combination has its own set of metricool:scheduleAPost calls because per-network field requirements differ (image URLs, video files, caption formats, alt text, hashtags).

After all Metricool calls complete, each branch:

• Creates a Scheduling History record in Notion (notion:createDataSourceItem).
• Updates the content page via Notion API (/v1/pages/{id}) — sets Last Used date, updates status.
• Sends confirmation to Google Chat via google-chat:makeApiCall — posts to the space ID looked up from Google Sheets.

YouTube Upload Branch

For Video Short content, dedicated branches fetch the video file from Google Drive (google-drive:getAFile) and upload directly to YouTube (youtube:uploadVideo). This runs alongside the Metricool posts, not instead of them.

Trigger: Webhook (Notion "Look for Social Media Handles")

Searches the web for a person's social media profiles and writes the URLs back to Notion.

Flow

Trigger: Webhook (Notion "Schedule Date From Status" automation)

Receives a Notion page ID when a Scheduled Date is set. Calculates the delay until the scheduled time, waits, then fires a webhook to the Metricool Content Poster scenario.

Flow Steps

Deluge Functions

string Minutes_till_Scheduled(string Current_Time, string Scheduled_Time)
{
  cleanScheduled = Scheduled_Time.subString(0,19).replaceAll("T"," ");
  cleanCurrent = Current_Time.subString(0,19).replaceAll("T"," ");
  scheduledDT = cleanScheduled.toDateTime("yyyy-MM-dd HH:mm:ss");
  currentDT = cleanCurrent.toDateTime("yyyy-MM-dd HH:mm:ss");
  diffMs = scheduledDT.toLong() - currentDT.toLong();
  diffMinutes = (diffMs / 60000).toLong();
  if(diffMinutes < 4)
  {
    diffMinutes = 4;
  }
  return diffMinutes.toString() + " minutes";
}

string Send_Webhook(string NotionID, string ScheduledDateStart)
{
  payload = Map();
  payload.put("Data_ID", NotionID);
  payload.put("Scheduled_Date_Start", ScheduledDateStart);
  response = invokeurl
  [
    url: "https://hook.us1.make.com/i2bxdatn59m1b9beovm77p33elk90tcb"
    type: POST
    body: payload.toString()
    headers: {"Content-Type":"application/json"}
  ];
  return response;
}

Module 269 in Chatbot 1.1 claude-haiku-4-5-20251001

Determines which workflow handles each message. Outputs a JSON object with workflow, error_message_to_user, set_organization, thumbnail, and button state.

Key Logic

Workflow Values

"notion" · "tweet creation" · "video short creation" · "graphic creation" · "permissions" · "button manager" · "database sync" · "notion user sync" · "timezone manager" · null (with error message)

Routing Disambiguation

Graphic attached + no help request → notion (upload). Graphic attached + help request → graphic creation. "sync" + "user/users" → notion user sync. "sync" without "user" → database sync. Unclear content type → defaults to tweet creation.

Output Schema

{
  "workflow": string | null,
  "error_message_to_user": string | null,
  "set_organization": string | null,
  "Public_Thumbnail": string | null,
  "Button1_Name": string | null,
  "Button1_Action": string | null,
  "Button2_Name": string | null,
  "Button2_Action": string | null,
  "Button3_Name": string | null,
  "Button3_Action": string | null,
  "Button4_Name": string | null,
  "Button4_Action": string | null
}

Module 293 in Chatbot 1.1 claude-haiku-4-5-20251001

Handles search, edit, and upload actions against the Notion content database. Outputs a complete Notion API request body ready to execute.

Key Logic

Output Schema

{
  "action": "search" | "edit" | "upload" | null,
  "API_URL": string | null,
  "Method": "POST" | "PATCH" | null,
  "Scheduled Date": string | null,
  "Public_Thumbnail": string | null,
  "body_json": string,
  "question_for_user": string | null
}

Module 450 in Chatbot 1.1 (Route 3.1.6a) claude-haiku-4-5-20251001

Writes tweets in Project Liberal's voice. Identifies task type (new/rewrite/alternatives/refine), content category, style, and tone.

Key Rules

Voice: Confident classical liberal. Uses "liberal" not progressive/centrist. Opponents labeled "illiberal"/"MAGA"/"authoritarian" not "Republican"/"GOP". No progressive buzzwords. No doom without confidence.

Length: Default under 100 chars. Max 200 unless style requires more. Median ~78 chars for punchy styles.

Formatting: Double line breaks for dramatic pause. ALL CAPS sparingly (single words). 0–2 emoji max. 0–1 hashtag max.

8 tweet styles: punchy_declarative, quote_attribution, contrast_hypocrisy, you_cant_be, question_led, to_be_is_to_be, historical_parallel, sarcastic_mockery.

Output Schema

{
  "reasoning": string,
  "identified_task": "new" | "rewrite" | "alternatives" | "refine" | "unclear",
  "identified_style": string,
  "identified_tone": string,
  "identified_category": string,
  "tweet_output": string | null,
  "message_to_user": string
}

Module 453 in Chatbot 1.1 (Route 3.1.6b) claude-haiku-4-5-20251001

Generates graphic concepts: visual format, exact text, image-text relationship, branding notes. Identifies task type (new_concept/write_text/refine/alternatives/help_with_attached).

Key Rules

Every graphic must make a specific argument — not just express a mood. Must frame from classical liberal principles.

Primary text target: under 120 characters. Uses specific numbers/data over vague claims.

Meme template rules: Drake = preference contrasts, Distracted Boyfriend = misplaced priorities, Wojak = in-group dynamics, NPC = mindless talking points, etc. Format must match argument structure.

Each concept delivers 5 components: graphic concept, primary text, supporting text, branding note (@projectliberal), why it works.

Output Schema

{
  "reasoning": string,
  "identified_task": string,
  "identified_style": string,
  "identified_tone": string,
  "identified_category": string,
  "graphic_output": string | null,
  "message_to_user": string
}

Module 456 in Chatbot 1.1 (Route 3.1.6c) claude-haiku-4-5-20251001

Writes spoken-word scripts. 3–5 beat structure: hook → core fact → context/evidence → escalation/stakes → closing punch. Identifies structure, tone, and target length.

Key Rules

Hook rules: First sentence must be bold declarative, provocative question, counterintuitive framing, or direct challenge. No preamble ever.

Closing rules: Warning, mic-drop, rhetorical question, principle affirmation, or callback. Never a CTA or summary.

8 script structures: breaking_news_analysis, hook_first_escalation, contrast_reversal, data_led_argument, rapid_fire_rhetorical, direct_address_response, storytelling_human_interest, philosophical_essay.

3 length brackets: Short (150–250 words), Standard (250–400), Long (400–600).

Visual directions: inline bracketed (e.g. [ON SCREEN: $330 billion / year]) only where they add value.

Output Schema

{
  "reasoning": string,
  "identified_task": string,
  "identified_structure": string,
  "identified_tone": string,
  "identified_category": string,
  "target_length": "short" | "standard" | "long",
  "script_output": string | null,
  "message_to_user": string
}

Module 469 in Chatbot 1.1 (Route 3.1.0) claude-haiku-4-5-20251001

Manages user permissions. View is open to all. Modify requires Admin. Admin Permission itself cannot be changed via chatbot. All changes require confirmation. Outputs current permission values for the target user.

Managed Permissions

Search · Scheduling · Editing · Uploading · Creation (not yet implemented in routing)

Admin — read-only via this prompt. Cannot be modified.

Output Schema

{
  "User ID": string | null,
  "Admin Permission": boolean,
  "Search Permission": boolean,
  "Scheduling Permission": boolean,
  "Editing Permission": boolean,
  "Uploading Permission": boolean,
  "Creation Permission": boolean,
  "Success Message to User": string | null,
  "Questions to User": string | null
}

Module 418 in Chatbot 1.1 (Route 3.1.1) claude-haiku-4-5-20251001

Manages the 4-button configuration and available actions list per space. Stores config as minified JSON in Google Sheets.

Key Rules

Buttons must have both name + action, or neither. Actions must exist in available_actions before assignment. Can't remove an action assigned to a button. At least one action must remain. Schedule actions require minutes value. Outputs validated JSON string.

Output Schema

{
  "reasoning": string,
  "button_settings": "{\"buttons\":[...],\"available_actions\":[...]}",
  "error_message_to_user": string | null,
  "success_message_to_user": string | null
}

9 sheets in the ChatBot workbook. Sheets serve as the configuration layer, state store, and history log for the chatbot system.

Two mirrored databases: PL Action Content (29 properties, ~2,100 rows) and PL Foundation Content (28 properties — same minus Created time).

Full Property List

23 properties. Stores individuals referenced in content with per-platform social URLs and auto-extracted handles.

10 properties. One record per social media publication event, created by the Metricool Content Poster scenario.

Connected Networks by Org

Per-Network Publishing Notes

Each metricool:scheduleAPost call is configured per-network because field requirements differ: Instagram requires an image URL, YouTube requires video file + title, Twitter has character limits + alt text, TikTok only accepts video, etc. This is why the Metricool Content Poster has so many modules — each content type × network is a separate call.

Thumbnails are uploaded to projectliberal.org via the WordPress REST API. This provides stable public URLs for content card images in Google Chat.

How It Works

When a graphic is processed by New Content Submission or the Chatbot:

• Image is uploaded to Google Drive → thumbnail link extracted from Drive metadata.
• Thumbnail downloaded from Drive.
• Thumbnail uploaded to WordPress via POST /wp-json/wp/v2/media.
• WordPress returns a public URL (e.g. https://projectliberal.org/wp-content/uploads/...).
• URL stored in the Notion page's Public Thumbnail property.

Fallback thumbnails (when no image exists) are stored in the Thumbnail Links sheet: a PL logo, a Twitter icon, and a video placeholder.