CORE

Generating

Execute AI processing to enrich your data with Horizontal and Vertical workflows.

Overview

Understanding the generation process and what to expect.

Generation is the final step in your Cension workflow. After setting up your organization, project, and dataset, and creating the columns you need, you're ready to generate data.

You can generate data using two powerful approaches:

Horizontal processing: Enrich existing rows by adding AI-generated content to your current dataset, processing one row at a time.
Vertical processing: Generate entirely new rows from your existing data, creating multiple outputs from single inputs.

All data generation runs as background processing, so you can continue working while your AI workflows execute. Large datasets are processed efficiently without blocking your interface.

Choose the right AI model for your needs, test workflows on sample data, and monitor processing status in real-time. Your generated content appears automatically once processing completes.

Horizontal processing

Add depth to your existing data, row by row.

Horizontal processing is the standard mode for adding depth to your existing data. It works 'row-by-row' to enrich each record with AI-generated content.

Common use cases

Add new columns to fill missing information or generate value:

Text generation: Write product descriptions, SEO meta tags, email subject lines, or social media posts.
Categorization: Classify rows by sentiment, topic, priority level, or custom internal tags.
Data extraction: Parse unstructured text to find emails, phone numbers, keywords, or specific entities.
Research: Look up company information, market data, or geographic details using web tools.

The enrichment workflow

1. Define context: Tell Cension which existing columns to use as input (e.g., use 'Title' and 'Features' to write a 'Description').
2. Build workflow: Use the Workflow Builder to chain AI models, logic, and tools.
3. Preview: Test your workflow on a single row to verify quality before spending credits. The Preview feature generates a sample result instantly without saving it permanently.
4. Process: Run the job across your entire dataset (or a filtered selection).

Processing modes

Fill empty: Smart processing that strictly targets blank cells. Ideal for resuming interrupted jobs or filling gaps without touching completed work.
Update existing: Forces a regeneration of all selected cells, overwriting any previous content. Use this when you've refined your prompts.

Efficiency

Cension uses Context Windows to bundle multiple fields into a single AI request, ensuring speed and consistency across your data.

Vertical processing

Generate new rows from your existing data.

Vertical processing creates new rows from your existing data using a 'One-to-Many' logic.

Goal: Expand lists, generate ideas, or conduct deep research.
Example: Input: 1 Row ('Summer Campaign') → Action: 'Generate 10 Instagram post ideas' → Output: 10 New Rows (one for each post idea), linked to the original campaign.
Search: You can combine this with search tools to find entities (e.g., 'Find 5 competitors for this company') and populate your dataset with them.

Models

Choose the right AI model for your processing needs.

Cension integrates with multiple AI models to give you flexibility in processing approach, cost, and quality. Choose the model that best fits your use case.

Model selection guide

Choose your model based on your priorities:

Quality first: Use gemini-2.5-flash or gpt-5 for marketing copy, creative content, or complex business logic.
Speed and cost: Use gpt-5-nano or gemini-2.5-flash-lite for bulk processing, simple categorization, or when budget matters most.
Balanced approach: Use gpt-4o or gpt-5-mini for most general-purpose tasks requiring reliability and consistency.

Available models

Here's a quick overview of available models:

gemini-2.5-flash: Google's most advanced model, excellent for quality and complex tasks.
gpt-5: OpenAI's latest flagship model, best for creative and analytical work.
gpt-4o: OpenAI's fast and capable model, great all-purpose balance.
gpt-4o-mini: Cost-effective version of gpt-4o for lighter tasks.
gpt-5-mini: Smaller gpt-5 model for efficiency-focused workflows.
gpt-5-nano: Most cost-effective option for bulk processing.
gemini-2.5-flash-lite: Lightweight Google model for speed and budget.

Model switching

You can switch models mid-workflow by selecting different blocks or changing the global model setting. Test different models on sample data to find the best fit for your content.

Model capabilities

Different models have different strengths beyond just raw intelligence. Here's what varies between them:

Special features: Image processing, advanced reasoning, multilingual support
Context window sizes: Standard models vs. enhanced capabilities in Deep Search mode

Deep search mode

Enhanced research capabilities with doubled costs:

What it does: Doubles credit costs but enhances depth and thoroughness of research operations
When to use it: Research-intensive tasks, competitive analysis, academic work
Cost impact: All credits multiplied by 2 - use only when maximum depth is needed

See our comprehensive credit system documentation for complete pricing details and cost management strategies.