⭐ Star this repo if you'd rather talk than type your next article.
5 minutes of talking. Published SEO article. $0.12.
Your site is empty. Not because you have nothing to say — because writing takes 5 hours you don't have.
Open Telegram. Hit record. Explain your thing like you're talking to a friend.
AI transcribes → researches trending keywords → writes the article in your voice → deploys it to your site with schemas, meta, cross-links, and sitemap.
📱 Voice memo → 🔍 Semantics → ✍️ Article → 🚀 Live on site
5 min auto auto auto
No Ahrefs. No Semrush. No blank page. Just you and your phone.
- You send voice notes → Telegram bot transcribes via Whisper
- AI extracts structure → thesis, sections, FAQ, keywords
- Semantic research → Google Suggest + Google Trends for trending topics
- AI writes the article → DeepSeek (or Claude/GPT) — SEO-optimized, human-readable
- Auto-deploy → FTP/SCP to your site + GitHub markdown backup
- Schema markup → Article + FAQPage JSON-LD for AI search visibility (AEO)
Built for Hermes Agent. Portable to Claude Code, Codex CLI, Cursor, or any LLM agent.
Writing blog posts sucks. Editing AI-generated fluff sucks more. This pipeline preserves your voice — slang, personality, directness — while handling the boring parts: semantics, structure, SEO tags, deployment.
The solopreneur content flywheel:
Voice memo (5 min) → Published SEO article (forever)
No dashboards. No content calendars. Just talk about what you know.
| The Old Way | This Tool |
|---|---|
| Open Ahrefs → keyword research (1h) | AI researches semantics automatically |
| Outline in Notion (30 min) | AI extracts structure from your voice |
| Write draft (2h) | You talked — that's the draft |
| Edit, re-edit, stare at screen (1h) | AI writes in your voice, not corporate fluff |
| Manually add meta tags, schema, OG (30 min) | Article+FAQPage Schema auto-generated |
| FTP/SCP upload, sitemap, cross-links (30 min) | Auto-deploy + sitemap + cross-links |
| Total: 5+ hours | Total: 5 minutes of talking |
Tools this replaces for a solopreneur:
- Ahrefs / Semrush — keyword research. Replaced by Google Suggest + Google Trends + AI clustering.
- Jasper / Copy.ai / Writesonic — AI copywriting. Replaced by DeepSeek with your voice prompt.
- SurferSEO / Clearscope — content optimization. Replaced by semantic cluster injection into Article Schema.
- WordPress / Ghost / Webflow — CMS. Replaced by static HTML deploy via SCP/FTP.
- Yoast / RankMath — on-page SEO. Replaced by auto-generated meta, schema, canonical.
- A human copywriter — $200-500/article. Replaced by $0.12 of API calls.
You still need a brain. You still need expertise. But you no longer need 6 tools and an afternoon to publish one article.
- Solopreneurs who need content but hate writing
- Indie hackers launching products with zero marketing team
- Consultants/agencies who think out loud and want that published
- Non-writers with domain expertise and a phone
| Component | What It Does | Cost |
|---|---|---|
| Telegram Bot | Receives voice messages | Free |
| Whisper (OpenAI) | Speech-to-text | $0.006/min (~$0.10/article) |
| DeepSeek API | Article generation + semantics | ~$0.015/article |
| Google Suggest / Trends | Keyword research | Free |
| Your server | Host the published HTML | Whatever you pay |
~$0.12 per published article. Cheaper than a coffee. Better ROI than most ad spend.
Talk to @BotFather → /newbot → get token
- OpenAI (Whisper): https://platform.openai.com/api-keys
- DeepSeek: https://platform.deepseek.com/api_keys
- Google Suggest / Trends — free, no key needed
Hermes Agent:
# Add the skill
cp SKILL.md ~/.hermes/profiles/marketing/skills/
# Set env vars
export OPENAI_API_KEY="sk-..."
export DEEPSEEK_API_KEY="sk-..."
export TELEGRAM_BOT_TOKEN="123:abc"Claude Code / Codex / Cursor:
# Copy PROMPT.md into your agent's system prompt or .cursorrules
# Add API keys to your environmentEdit SKILL.md → update deploy section with your FTP/SCP credentials and paths.
Send a voice message to your bot. The agent picks it up, transcribes, structures, writes, and deploys.
📱 Telegram Voice Note
↓
🗣️ OpenAI Whisper → Raw transcript
↓
🧠 AI extracts: thesis, sections, FAQ, keywords
↓
🔍 Google Suggest / Google Trends → Trending semantics
↓
✍️ DeepSeek writes: HTML article with Article+FAQPage Schema
↓
🚀 Deploy: FTP/SCP to your site + GitHub markdown
↓
📊 SEO: sitemap update, cross-links, search engine ping
From my production pipeline (axelfreeman.com):
- 83 published Q&A pages
- ~100K monthly search impressions
- 4-6 FAQ items per page with FAQPage Schema
- Article published in <2 minutes after voice memo received
This isn't "AI wrote me a blog post." This is:
- Trending topic detection — Google Trends to find what's growing
- Cluster analysis — group keywords by intent, not just volume
- Exact match filtering — broad numbers lie, exact match tells the truth
Example Google Suggest expansions for "AI copywriting":
ai copywriting→ai copywriting tools,ai copywriting free,ai copywriting generatorvoice to text→voice to text online,voice to text free,voice to text app
Full semantic data: semantics/ directory.
Input: 5-minute voice memo about why Telegram is the best platform for AI agents.
Output: Published article on axelfreeman.com with:
- SEO title + meta description
- 8 H2 sections with bullet points
- 2 card blocks (key insight + CTA)
- 6 FAQ with matching JSON-LD Schema
- Google Analytics tracking
See examples/ for the full before/after.
| Agent | How to Use |
|---|---|
| Hermes Agent | Drop SKILL.md into skills directory |
| Claude Code | Paste PROMPT.md into CLAUDE.md |
| Codex CLI | Copy to .codex.md or pass as system prompt |
| Cursor | Add to .cursorrules |
| Any LLM | PROMPT.md works as a standalone system prompt |
voice-to-article/
├── README.md ← You are here
├── SKILL.md ← Hermes Agent skill (full pipeline)
├── PROMPT.md ← Universal system prompt for any LLM agent
├── SETUP.md ← Detailed setup: bots, keys, deploy targets
├── semantics/
│ └── en-suggestions.json← keyword suggestions (Google Suggest)
├── examples/
│ └── article-en.html ← Final published article
└── .github/
└── FUNDING.yml
- Why Claude Sucks for Marketing — why I use DeepSeek for content
- AI Marketing Without Bullshit — my blog
MIT — use it, remix it, ship it. If this saves you time, buy me a coffee.
Last updated: August 2026.
The stack behind this repo runs as a service: marketing engineering, turnkey — Sprint $900 one-time, Engine $1,900/month, full build $2,900. Scope and prices are published before the first call. If you run a local or multi-site business, the countable version of the same work is here: marketing engineer for local business.