AI-Powered Content Production for Agencies, Freelancers, and Growing Businesses
From topic to published-ready article, social content, metadata, and images, in minutes, not days.
Content production is repeatable work, and it's eating your time
Every article follows the same sequence: research competitors, identify keyword opportunities, write the draft, rewrite it to match the brand's voice, generate metadata, create social posts for three platforms, source or generate images, and brief a designer on a thumbnail. Repeat for every client, every week.
Whether you're running a small agency, freelancing across multiple clients, or trying to grow a business without a full content team, the manual volume is the problem. Each brand has different voice rules, banned phrases, audience expectations, and content standards. Managing that without a system means inconsistent output, burnout, or both.
AI writing tools promised to fix this. They haven't. Generic output doesn't sound like any particular brand, and prompting your way to consistency is a full-time job in itself. The real problem isn't generating words. It's generating the right words, in the right voice, with the right facts, every time.
- 2–3 hours of manual competitor and SERP research per topic
- A full draft write, then a separate brand voice rewrite pass
- SEO metadata written from scratch each time
- 3–4 separate social posts written per article, per platform
- Image sourcing or separate design briefs for every piece
- Voice calibration rebuilt from memory with each new writer or AI prompt
- No centralized system. Brand voice lived in someone's head
A full content production platform, built for anyone managing more than one brand or more than one article a week
The system replaces every manual step in the content workflow with an automated, configurable pipeline. Each client or brand gets their own voice configuration, knowledge base, and content rules. The platform runs them all from a single interface, whether that's two clients or twenty.
Multi-Client Architecture
Every brand is fully isolated. Separate voice configuration, banned terms, knowledge base documents, custom instructions, and content settings, all stored per client. Onboard a new brand without touching any other client's setup. Works for agencies managing multiple accounts, freelancers with a roster of retainer clients, or business owners running more than one brand.
Research-First AI Pipeline
Before a single word is written, the system pulls live SERP data, scrapes top-ranking competitor pages, and extracts People Also Ask questions and content gaps. Every article is grounded in actual search reality, not hallucinated context or outdated assumptions.
Per-Client Brand Voice System
A layered voice system. Custom instructions, banned terms, voice rules, and real writing samples. Enforces each client's voice at the model level. A secondary voice compliance pass rewrites any section that drifts from the approved style. The output sounds like the brand, not like a generic AI.
Accuracy and Pattern Checks
A deterministic post-generation filter catches banned words, AI writing clichés, and structural violations that prompts alone miss. Zero-tolerance terms trigger a targeted rewrite, then a mechanical fallback, so they never reach the client. Statistics policy settings prevent unverified numbers from entering the article.
Full Content Suite Per Run
One article run produces: the SEO article, SEO metadata (title, description, slug, OpenGraph fields), LinkedIn post, Facebook post, Instagram caption, video script, generated inline images, and a featured thumbnail. Everything needed to publish, in one pass.
SEO and AI Search Optimized
Content is structured to perform in traditional search and in AI-generated answers (Google's AI Overviews, Perplexity, ChatGPT search). Each article is built with PAA questions, AEO formatting, and structured headers, so it's findable however your audience is searching. The system also outputs image generation prompts and video script prompts.
The 5-step content pipeline
Each step runs sequentially. Outputs are saved at every stage, so a failure at any point preserves everything before it. No lost work, no starting over.
-
1
Research
Live competitor scraping, SERP analysis, People Also Ask extraction, and content gap identification. Produces a structured research brief with verified key facts the article can actually cite.
Gemini + SerpAPI + Firecrawl -
2
Writing
Draft article written against the research brief and the brand's full voice system. A second pass runs section-by-section against voice examples and rules. A deterministic filter then catches any remaining banned patterns before the output is finalized.
Claude with Voice Compliance Pass -
3
Metadata
SEO metadata generated automatically: meta title, meta description, URL slug, and OpenGraph fields. Structured output format means no parsing errors and consistent field population every time.
Gemini Structured Output -
4
Images
Image prompts are derived from article context and brand guidelines. Inline images are generated and placed at marked positions in the article. A featured thumbnail is generated separately for the article header.
fal.ai Generation -
5
Social Repurposing + Visual Content Prompts
After the article completes, the repurposer produces LinkedIn, Facebook, and Instagram posts, each written to its platform's conventions, not just the article copy pasted into a caption box, plus a short-form video script. The system also outputs standalone image generation prompts and video production prompts so you can create supporting visual content in any external tool.
Platform-Specific Repurposing Engine
How brand voice is actually enforced, not just requested
Most AI content tools accept a "write in our brand voice" instruction and hope for the best. This system enforces voice at multiple layers, from the top of the model prompt down to post-generation code checks. Each layer catches what the one above it misses.
| Priority | Layer | What It Does |
|---|---|---|
| 1 | Custom Instructions | Client-specific mandatory rules injected at the top of every prompt. Written with ceiling-based language and explicit numeric limits. The model reads them first and they override everything below. |
| 2 | Banned Terms | Words and phrases the model must never produce. Backed by a code-level deterministic filter for zero-tolerance terms. Prompts are probabilistic, code is not. |
| 2.5 | Voice Rules | Binding calibration rules: terminology swaps, parenthetical limits, statistical citation policy, positioning language. One structured document per brand. |
| 3 | Voice Examples | Real writing samples from the client, or client-approved rewrites of AI output. The single most effective lever in the system. The model matches the actual cadence and vocabulary rather than inferring from rules. |
| 4–8 | Structure, Stats, Knowledge Base, Guidelines | SEO structure rules, verified-facts-only statistics settings, industry terminology reference docs, and general content best practices. All compiled and injected in prompt priority order. |
Accuracy checks built into every article
Zero-Tolerance Filter
Banned words and AI clichés (e.g., "game-changer," "seamless," "transformative") trigger a targeted rewrite, then a mechanical regex fallback if the rewrite still fails. They don't reach the client.
Frequency Limits
Patterns allowed sparingly are capped at one per article. Two or more triggers a flag and targeted rewrite.
Structural Checks
Parenthetical overuse, heading personality stacking, and AI meta-commentary (12 detected patterns) are caught and stripped before output is finalized.
Statistics Policy
When the verified-stats flag is on, the model can only cite numbers from the research brief's Key Facts section. No invented statistics reach the article.
How the system is deployed and managed
Infrastructure
The platform runs on Railway with a Node.js backend and PostgreSQL database. The frontend is a React interface. All client configurations, knowledge base documents, voice files, and article history live in the database, not flat files, not spreadsheets.
Real-time progress streams to the UI as the pipeline runs, so you can watch each step complete rather than waiting for a finished result to appear.
Generated images are stored in Cloudflare R2. Articles, metadata, and social content are stored in the database and accessible per-article from the client dashboard.
Onboarding a New Client
Setting up a new brand takes about two hours. Create the brand record, configure tone, voice, and content settings, upload the banned terms and voice rules documents, and add 2–3 real writing samples as voice examples. Run a test article, audit it against the voice rules, and adjust any instructions based on what fails.
Once a client is calibrated, the system runs without manual intervention on each article. The pipeline handles research through social output end-to-end.
What the system produces, and what it replaces
These estimates are based on tracked manual time for a typical 1,500-word SEO article with social repurposing, metadata, and images included in scope.
Research, writing, voice rewrite, metadata, social posts, image sourcing, and thumbnail brief, done by a human writer and designer.
Full 5-step pipeline runs including research, writing, voice compliance, metadata, images, and social repurposing.
Human review and light editing are still recommended. The pipeline handles production; humans handle final judgment.
SEO article, full metadata set, LinkedIn + Facebook + Instagram posts, video script, image generation prompts, and generated images. One trigger, all delivered.
No per-client infrastructure. New brands are database records, not new deployments. The system scales horizontally across as many clients or brands as needed.
A new client produces on-brand output from their first article run, not after months of writer calibration and style guide enforcement.
Built for anyone managing content at volume
Small Agencies
Managing five, ten, or twenty client accounts means five, ten, or twenty content workflows to run every month. This system collapses that into a single platform with per-client isolation. Same output quality, fraction of the production time.
Freelancers
Retainer clients expect consistent output. This gives you a production system that scales your capacity without scaling your hours. Take on more clients without adding more late nights.
Small Business Owners
You know your industry. You know your voice. You don't have time to produce a blog, three social posts, metadata, and images every week on top of running the business. This system handles the production once you've told it what you sound like.
Want this running for your business or your clients?
We build and configure the full system. Pipeline, voice setup, client onboarding, and deployment. You run it from day one.
Let's Talk