AI Blog Generation

Wrytze generates blog posts through a 3-phase AI pipeline. Each phase builds on the output of the previous one to produce research-backed, brand-consistent content. This page explains what happens in each phase and how the system ensures quality and reliability.

Overview

When you trigger a blog generation from a topic, Wrytze runs three sequential phases:

1. Research -- Analyze the topic, gather sources, and produce a structured outline
2. Writing -- Generate the full article using the research brief and brand profile
3. Media -- Create image prompts based on the written content

The result is a complete blog post draft with metadata, ready for editing and review.

Phase 1: Research

The research phase lays the foundation for a well-informed article. Given your topic and brand profile, the AI:

• Analyzes the topic in the context of your brand and audience
• Identifies relevant keywords and search terms
• Gathers sources and reference material
• Generates a structured outline with sections and key points

Output: A research brief containing keywords, sources, and a detailed outline. This brief becomes the input for the writing phase.

Phase 2: Writing

The writing phase takes the research brief and produces a complete article. Using the outline, sources, and your brand profile as context, the AI:

• Generates the full HTML article content following the outline structure
• Creates metadata: title, meta description, and excerpt
• Calculates word count and estimated reading time
• Matches the tone, voice, and style defined in your brand profile

Output: A complete blog post with HTML content, SEO metadata, word count, and reading time. The post is saved as a draft, ready for editing.

Phase 3: Media

The media phase analyzes the written content and generates prompts for visual assets:

• Reviews the article structure and key sections
• Generates an image prompt for the featured/hero image
• Generates image prompts for inline images at relevant points in the article

Output: A set of image generation prompts that can be used for AI image creation. These prompts are attached to the blog post for the editor to review and use.

Brand awareness

Every phase uses your website profile as context. The profile is generated by crawling your actual website (see Website Profiles) and includes:

• Brand voice -- The personality and character of your writing (e.g. authoritative, conversational, playful)
• Tone -- The emotional quality of your content (e.g. professional, friendly, enthusiastic)
• Target audience -- Who you are writing for (e.g. developers, marketers, enterprise buyers)
• Content style -- Structural preferences like article length, heading style, and formatting conventions

This ensures that AI-generated content sounds like it came from your team, not a generic content mill.

Supported AI providers

Wrytze supports both OpenAI and Anthropic models through the Vercel AI SDK. Your organization can configure which provider and model to use for generation. The provider-agnostic architecture means:

• You can switch providers without changing your workflow
• Both providers use the same 3-phase pipeline
• Model selection can be configured per organization

Orchestration and reliability

The 3-phase pipeline runs as an Upstash Workflow for reliable execution. Each phase is a discrete step in the workflow, which provides:

• Automatic retries -- If a phase fails (e.g. due to a transient API error), it is retried automatically without restarting the entire pipeline
• Independent recovery -- Each phase can be retried independently. A failure in Phase 3 does not require re-running Phases 1 and 2
• Progress tracking -- You can see which phase is currently running and monitor progress in the dashboard
• Durable state -- Intermediate results are persisted between phases, so no work is lost if the system needs to restart

What happens after generation

Once all three phases complete, the blog post is saved as a Draft. From there, your team can:

1. Edit the content in the block editor -- adjust wording, add sections, upload images
2. Submit for review through the approval workflow
3. Publish immediately or schedule for a future date

AI generation is the starting point, not the end. The goal is to give your team a strong first draft that significantly reduces the time from topic to published post.