The AI Write Tool Factory: 1,000 Pages, Zero Prompts

There are 49,500 people searching for “ai writer” every month. Another 22,200 searching “ai write.” And then it gets interesting — 5,400 people specifically searching “ai write obituary.” Not “best AI tool.” Not “how to use ChatGPT.” They want an AI to write an obituary. Right now. Today. Because someone died and they’re sitting in front of a blank document at 2am and they can’t find the words.

Those people don’t want a prompt. They don’t want to figure out ChatGPT. They want a box that says “Write an Obituary” at the top and a form underneath that asks for the person’s name, their age, and a few things they loved.

And there are thousands of searches like this.

The Keyword Goldmine Nobody’s Mining Properly

I pulled keyword data from Ubersuggest for combinations of “ai” and “write” and found something that made me sit up straight. Here’s a sample of the long-tail searches people are actually typing into Google:

The high-volume obvious ones:

“ai writer” — 49,500/month
“ai writing” — 27,100/month
“ai write” — 22,200/month
“ai write tool” — 22,200/month (SEO difficulty: 32 — that’s practically uncontested)

The specific-task searches that tell a story:

“ai write obituary” — 5,400/month (SEO difficulty: 32)
“ai write book” — 480/month
“ai write story” — 390/month
“ai write paragraph” — 260/month
“ai write letter” — 260/month
“ai write song” — 260/month
“ai write resume” — 170/month
“ai write an email” — 170/month
“ai write a poem” — 110/month
“ai write a bio for me” — 110/month
“ai write script” — 110/month
“ai write a cover letter” — 90/month
“ai write lyrics” — 90/month
“ai write dating profile” — 50/month
“ai write a thank you note” — 50/month
“ai write birthday message” — 20/month
“ai write resignation letter” — 30/month
“ai write performance review” — 30/month
“ai write recommendation letter” — 30/month
“ai write a sentence” — 20/month
“ai write facebook post” — 20/month
“ai write instagram caption” — 10/month

Look at that list. Every single one of those is a person who wants exactly one thing. Not a general-purpose AI. Not a subscription to yet another writing platform. A single tool that does a single job.

And here’s the kicker: most of these keywords have SEO difficulty scores in the 20s and 30s. That’s the sweet spot where a well-optimized page on a decent domain can rank without needing a miracle.

The Tier 2 AI Tool Pattern (And Why It Works)

Let me describe what’s already happening in this space, because it’s instructive.

Right now there are hundreds of “AI writing tools” that are essentially the same thing: a wrapper around GPT or Claude with a pre-written system prompt. The user picks a task from a dropdown, fills in a few fields, and gets output. WriteSonic, Copy.ai, Jasper, Rytr — they all follow this pattern. These are what I’d call Tier 1 tools: subscription products with full feature sets, user accounts, prompt libraries, team features, the works.

But below that tier, there’s a massive ecosystem of what I’ll call Tier 2 tools — single-purpose pages that do exactly one thing. “AI Poem Generator.” “AI Email Writer.” “Free AI Story Generator.” These are the sites that rank for those long-tail keywords. They’re typically:

One page, one task
No account required (or optional)
Free tier with limitations
Monetized through ads, upsells, or affiliate links
Built on the same API underneath
SEO-optimized for that single keyword

The insight is this: ChatGPT can do all of these things. Claude can do all of these things. But people don’t search “how to prompt ChatGPT to write an obituary.” They search “ai write obituary.” They want the McDonald’s drive-through, not the fully stocked kitchen.

“People don’t want a quarter-inch drill. They want a quarter-inch hole.” — Theodore Levitt

Except in this case, people don’t even want the drill. They want someone to point the drill, pull the trigger, and hand them the hole.

The Architecture: A Tool Factory

Here’s what I’m thinking. Instead of building one tool at a time, build a tool factory — a template system that can generate hundreds of single-purpose AI tool pages from a config file.

The Zero-Typing Interface

The most interesting version of this isn’t even a text input. It’s pure selection:

Write a [tone] [format] about [topic]

Where each bracket is a dropdown:

Tone: Happy | Sad | Professional | Funny | Heartfelt | Motivational | Sarcastic
Format: Poem | Story | Song | Paragraph | Letter | Email | Social Post
Topic: Doctors | Nature | Love | Technology | Food | Travel | Custom…

The user picks three things, hits “Generate,” and gets output. No typing required. No prompt engineering. No thinking about how to phrase things.

For more specific tools, the form adapts:

“AI Obituary Writer”

Full name: [text input]
Age: [number]
Relationship to you: [dropdown: parent, spouse, sibling, friend, child, other]
Things they loved: [text input]
Tone: [dropdown: traditional, celebration of life, religious, simple]
Length: [dropdown: short, medium, detailed]

“AI Cover Letter Writer”

Job title: [text input]
Company name: [text input]
Your experience level: [dropdown: entry, mid, senior, executive]
Industry: [dropdown: tech, healthcare, finance, education, ...]
Tone: [dropdown: formal, conversational, enthusiastic]
Key skill to highlight: [text input]

“AI Dating Profile Writer”

Platform: [dropdown: Tinder, Hinge, Bumble, OkCupid, Generic]
Gender: [dropdown options]
Vibe: [dropdown: witty, sincere, adventurous, intellectual, laid-back]
Interests: [multi-select: outdoors, food, travel, gaming, fitness, ...]
Deal-breakers to mention: [optional text]

Each tool page is SEO-optimized for its specific keyword. The URL structure writes itself: /tools/ai-obituary-writer, /tools/ai-cover-letter-generator, /tools/ai-dating-profile-writer.

The Config-Driven Approach

Every tool could be defined in a JSON or YAML config:

# tools/obituary-writer.yaml
tool:
  name: "AI Obituary Writer"
  slug: "ai-obituary-writer"
  description: "Create a respectful, personalized obituary in seconds"
  target_keyword: "ai write obituary"
  search_volume: 5400
  category: "personal"

  fields:
    - name: "full_name"
      type: "text"
      label: "Full name of the deceased"
      required: true
    - name: "age"
      type: "number"
      label: "Age"
      required: true
    - name: "relationship"
      type: "select"
      label: "Your relationship"
      options: ["Parent", "Spouse", "Sibling", "Friend", "Child", "Grandparent", "Other"]
    - name: "loved"
      type: "text"
      label: "Things they loved (hobbies, passions, people)"
      placeholder: "gardening, jazz music, Sunday dinners with family"
    - name: "tone"
      type: "select"
      label: "Tone"
      options: ["Traditional", "Celebration of Life", "Religious", "Simple & Brief"]
    - name: "length"
      type: "select"
      label: "Length"
      options: ["Short (2-3 paragraphs)", "Medium (4-6 paragraphs)", "Detailed (7+ paragraphs)"]

  system_prompt: |
    You are a compassionate obituary writer. Create a respectful,
    warm obituary based on the provided details. Match the requested
    tone and length. Focus on celebrating the person's life.

  seo:
    title: "Free AI Obituary Writer - Create a Personalized Obituary"
    description: "Write a respectful, personalized obituary using AI. Just enter a few details and get a beautiful tribute in seconds. Free to use."
    h1: "AI Obituary Writer"
    intro: "Writing an obituary is never easy. Let AI help you find the right words to honor your loved one."

One template component reads the config. One template page renders it. You write a new YAML file, and a new tool page exists. Want 200 tool pages? Write 200 configs. The actual UI code stays the same.

The Tech Stack (Keeping It Cheap)

Here’s where it gets interesting from a cost perspective. The whole site can be static — Astro generates the pages at build time. The AI generation happens at the edge. But the LLM costs could kill you if you’re running GPT-4 or Claude on every request for a free tool.

Phase 1: Local LLM Pipeline

Run a local LLM (Llama 3, Mistral, Phi-3) on your own hardware. Use Cloudflare Tunnel to expose a local API endpoint. The flow:

User → Cloudflare Pages (static) → Edge Function → Cloudflare Tunnel → Local Machine → Ollama/llama.cpp → Response

Cost: electricity. Maybe $10-20/month in additional power. The local machine (even an M-series Mac Mini) can handle 10-20 concurrent requests with a 7B or 13B model. For simple form-driven tasks with clear instructions, a smaller model does surprisingly well.

// Cloudflare Worker that proxies to local LLM
export default {
  async fetch(request, env) {
    const { tool, fields } = await request.json();

    // Load tool config
    const config = await env.TOOLS_KV.get(tool, "json");

    // Build prompt from template + user fields
    const prompt = buildPrompt(config.system_prompt, fields);

    // Route to local LLM via tunnel
    const response = await fetch(env.LOCAL_LLM_ENDPOINT, {
      method: "POST",
      body: JSON.stringify({
        model: config.model || "llama3:8b",
        prompt,
        stream: false,
      }),
    });

    return new Response(response.body, {
      headers: { "Content-Type": "application/json" },
    });
  },
};

Phase 2: Hybrid Routing

As traffic grows, you add smart routing:

Simple tasks (sentence, paragraph, social post) → local 7B model
Medium tasks (email, letter, bio) → local 13B model or Cloudflare Workers AI
Complex tasks (obituary, cover letter, story) → Claude Haiku or GPT-4o-mini
Premium tasks → Claude Sonnet or GPT-4o (behind a paywall or ad-gate)

The routing logic lives in the edge function. You’re spending pennies on 80% of requests and only paying real API costs for the 20% that need quality.

function selectModel(tool, fields) {
  const complexity = tool.complexity || "medium";
  const wordCount = estimateOutputWords(tool, fields);

  if (complexity === "low" || wordCount < 100) {
    return { provider: "local", model: "llama3:8b" };
  }
  if (complexity === "medium" || wordCount < 500) {
    return { provider: "cloudflare-ai", model: "@cf/meta/llama-3-8b-instruct" };
  }
  if (complexity === "high") {
    return { provider: "anthropic", model: "claude-haiku-4-5-20251001" };
  }
  return { provider: "anthropic", model: "claude-sonnet-4-6" };
}

Phase 3: Revenue-Funded Scaling

Once the site has traffic:

Free tier: 3 generations/day, powered by local or Cloudflare Workers AI
Show an ad between form and result (interstitial or banner)
“Premium” toggle: uses Claude/GPT-4 for higher quality, costs a token or shows more ads
Affiliate links on relevant tool pages (resume writer → link to resume templates, cover letter → link to job boards)

The obituary writer page alone — 5,400 searches/month, SEO difficulty 32 — could potentially drive enough traffic to fund the API costs for dozens of other tools through ads alone.

The Math

Let’s run some rough numbers:

Target: 100 tool pages

If each page ranks and captures even 5% of its target keyword’s search volume:

Tool	Monthly Searches	Est. Clicks (5%)	Annual Clicks
AI Writer (general)	49,500	2,475	29,700
AI Write Tool	22,200	1,110	13,320
AI Obituary Writer	5,400	270	3,240
AI Write Story	390	20	234
… (96 more pages)	~15,000 combined	750	9,000
Total estimate	~92,000	~4,600/month	~55,000/year

That’s conservative. Some of those pages won’t rank. Some will rank higher than position 5. But even at half those numbers, you’re looking at 2,000+ visits/month across the portfolio.

Cost structure with local LLM:

Domain: $12/year
Cloudflare Pages: free
Electricity for local LLM: ~$15/month
Total: ~$192/year

Revenue at even modest ad rates ($2-5 RPM):

55,000 pageviews × $3 RPM = ~$165/year from ads alone

Break-even is nearly immediate with the local LLM approach. And that’s before any upsells, premium tiers, or affiliate revenue on specific tool pages (resume writers, cover letter writers, and business plan writers have CPCs of $5-23 — advertisers are paying real money for those clicks).

Why This Might Actually Work

1. The keyword data doesn’t lie. People are searching for these exact tools. The demand exists. It’s not theoretical.

2. SEO difficulty is low. Most of these long-tail keywords have difficulty scores under 40. The current results are often thin content pages or generic “top 10 AI tools” listicles. A purpose-built tool page that actually does the thing should outrank a listicle.

3. The build cost is near zero. One template component. YAML configs. Static hosting. The marginal cost of adding tool #201 is writing a YAML file and a system prompt — maybe 20 minutes of work.

4. Local LLM changes the economics. Two years ago, you couldn’t run a useful LLM on consumer hardware. Now you can. A Mac Mini M4 running Llama 3 can generate a poem in 2 seconds and costs nothing per request. The API cost wall that killed a lot of AI tool startups in 2023-2024 just… isn’t there anymore if you’re willing to self-host.

5. The upgrade path is clear. Start local, validate demand, then upgrade the model tier on high-value pages. You’re not betting everything on an expensive API from day one.

Why This Might Not Work

Because I’m not delusional enough to only see the upside.

1. Google might not rank tool pages. Google has been increasingly skeptical of AI-generated content and tool pages. If they decide these are “thin content” despite being functional tools, you’re dead in the water.

2. Quality ceiling with local models. A 7B parameter model writing an obituary might produce something… adequate. But “adequate” might not be good enough for something that emotionally matters. The obituary use case in particular is one where quality really counts.

3. ChatGPT is free. The fundamental challenge: why would someone use your tool when they could just ask ChatGPT? The answer is friction reduction — they don’t know how to prompt, they don’t have an account, they want it faster. But that advantage erodes as ChatGPT gets easier to use and more widely adopted.

4. Competition is coming. If this keyword data is obvious to me, it’s obvious to others. Every “AI tool” startup with a marketing team has seen these keywords. The window for organic SEO dominance won’t stay open forever.

5. Maintenance overhead. 200 tool pages means 200 system prompts to maintain, test, and improve. When the underlying model changes, every prompt might need tweaking. That’s not nothing.

The Cloudflare Pipeline Dream

One thing that keeps nagging at me: Cloudflare’s expanding AI infrastructure. They already have Workers AI (run inference at the edge), KV (store configs), R2 (store outputs), and Tunnels (connect to local machines). What if the entire pipeline lived in Cloudflare’s ecosystem?

Request → Cloudflare Worker → Workers AI (for simple) / Tunnel to Local (for complex)
       → KV (rate limiting, caching)
       → R2 (store generated outputs for caching)
       → Pages (static frontend)

You could literally run this entire operation on Cloudflare’s free or $5/month plans for the Workers, with the LLM costs eating only electricity on your local machine. The first $1,000 in revenue is almost pure profit.

And the dream version: as Cloudflare Workers AI gets better models (they keep adding them), you could gradually shift traffic off local hardware entirely. Your Cloudflare bill goes up, but your maintenance burden goes down, and you’re no longer dependent on keeping a machine running 24/7.

What I’d Build First

If I were to start this tomorrow, here’s the priority list based on the keyword data:

AI Obituary Writer — 5,400 searches/month, SEO difficulty 32, deeply emotional use case where people genuinely need help, low competition
AI Cover Letter Writer — 90 searches for the exact term but “ai write a cover letter” and variations combine to ~250+/month, high commercial intent (CPC ~$3-4)
AI Poem Generator — 110+ searches/month across variations, fun/viral potential, very forgiving quality bar
AI Bio Writer — 110+ searches/month, professional use case, easy form design
AI Thank You Note Writer — 50 searches/month but extremely simple to build, good proof of concept
AI Dating Profile Writer — 50 searches/month, fun angle, shareable, potential viral loop
AI Email Writer — 170 searches/month, professional utility, natural upsell to premium
AI Story Generator — 390 searches/month, higher quality bar but strong demand
AI Resume Writer — 170 searches/month, $8+ CPC, premium tier candidate
AI Performance Review Writer — 30 searches/month but $13+ CPC — someone’s paying real money for that keyword

Ten pages. Maybe a weekend to build the template system, another weekend to write the configs and test the prompts. If even three of them rank, the data starts flowing and you know whether to scale to 100.

FAQ

How is this different from ChatGPT?

No account needed. No prompt required. You see a form, fill in a few fields, click a button. It’s the difference between cooking from scratch and ordering from a menu. ChatGPT is more powerful, but power isn’t what most people want — they want simplicity.

Would the output quality be good enough with a local LLM?

For most of these tasks, yes. Writing a birthday message, a social media caption, or a basic paragraph doesn’t need GPT-4. A well-prompted 8B model handles these fine. For higher-stakes content like obituaries or cover letters, you’d route to a better model. The config-driven approach means you can set the model per tool.

How would you monetize this?

Three layers: display ads (low but steady), affiliate links on relevant tool pages (resume tools → job boards, cover letter tools → career sites), and a premium tier for higher-quality output. The local LLM approach means your baseline costs are nearly zero, so even modest ad revenue is profitable.

What about the legal implications of AI-generated content?

This is genuinely worth thinking about, especially for obituaries and professional documents. The output should always be presented as a “draft” or “starting point” with a clear disclaimer. Nobody should submit an AI-generated cover letter without reading it first. But then, nobody should submit a human-written one without reading it either.

Could someone just copy this idea?

Absolutely. And they will. The moat isn’t the idea — it’s execution speed, SEO momentum, and the compounding advantage of having 200 indexed pages when your competitor has 5. First-mover advantage in SEO is real, even if the underlying product is simple.

What’s the realistic timeline to see if this works?

Build the template system: one weekend. Create the first 10 tool pages: another weekend. Wait for Google to index and rank: 2-4 months. So you’d know by summer if the SEO thesis holds. Total investment before you have data: two weekends of work and $12 for a domain. That’s about as cheap as validation gets.