Sourced

Inspiration

Search is being rewritten in real time. When a small business owner's customer asks ChatGPT, Claude or Gemini "who's the best family lawyer in Manchester?" — the model gives one answer. Maybe two. The SEO playbook of the last 20 years doesn't work for that. Sourced was built for businesses who can't afford to be invisible to the models.

What it does

Sourced is an editorial AI marketing employee for small service businesses.

• Audits how Claude, ChatGPT and Gemini currently cite your business on real buyer-intent prompts
• Writes citation-grade content the models will actually quote back — in your voice, not generic SaaS blog filler
• Delivers three pieces per month and a weekly Friday report
• Flags weak phrasing, generic claims, and "trust-erosion" language with editor's notes
• Tracks improvements in citation rate across the major LLMs over time

Think of it as a senior content editor who happens to know exactly how language models pick which businesses to mention.

How we built it

• Frontend: Next.js, React, TypeScript, Tailwind CSS, Framer Motion
• Backend: Node.js, FastAPI, Supabase, PostgreSQL with pgvector
• AI layer: Claude API for editorial reasoning, OpenAI and Gemini for cross-model citation auditing, custom RAG pipeline over the client's existing brand voice
• Orchestration: Multi-model agent that runs buyer-intent prompts across LLMs, parses citations, scores quotability, then drafts content optimised for each model's retrieval patterns
• Delivery: Automated weekly reports, Stripe subscription billing

Challenges we ran into

The hardest problem wasn't generating content — it was generating content the models actually quote. We rebuilt our prompting pipeline three times before we found a structure (claim → evidence → specificity → attribution) that consistently made LLMs cite the client in follow-up prompts. The other hard part: writing in a client's actual voice, not a watered-down brand-safe version of it.

Accomplishments that we're proud of

• Measurable lift in LLM citation rate for early clients within 30 days
• A content QA layer that flags AI slop — hype words, vague claims, generic structure — before anything ships
• Zero "sounds like ChatGPT wrote it" feedback from clients to date

What we learned

Generative Engine Optimization (GEO) is a real discipline now. Citations are the new backlinks. The businesses that win the next decade will be the ones whose content reads like it was written by a human expert — because that's what the models reward.

What's next for Sourced

Expanding the audit to cover Perplexity, You.com, and SearchGPT. Building a dashboard so clients can see exactly which prompts they win and lose, week over week.

Built With

• claude-api
• fastapi
• framer-motion
• gemini
• next.js
• node.js
• openai
• postgresql
• python
• rag
• react
• stripe
• supabase
• tailwind-css
• typescript