AEO Prompt Research (Free)

Source: github.com/psyduckler/aeo-skills Part of: AEO Skills Suite — Prompt Research → Content → Analytics

Discover which prompts and topics matter for a brand's AI visibility — using zero paid APIs.

Context: Why This Matters Now

Google's AI Overviews and Search AI Mode — powered by Gemini 3 Flash — now handle a significant share of search queries. These AI-generated answers cite specific sources, creating a new competitive landscape. This skill identifies the prompts where your brand should be cited and audits your current coverage.

Requirements

• web_fetch — crawl the target site
• web_search — Brave Search free tier (optional but recommended)
• LLM reasoning — the agent's own model does the heavy lifting
• Optional: Gemini API key (free) — use gemini-3-flash-preview with grounding to validate which prompts currently cite your brand (20 samples per prompt for reliable signal)

No paid tools or accounts required for core functionality.

Workflow

Input

The user provides:

• Domain URL (required) — e.g. acme.com
• Niche/category (optional) — e.g. "SEO software for content teams"
• Competitors (optional) — e.g. "Surfer SEO, MarketMuse, Frase"

Step 1: Crawl the Site

Use web_fetch on key pages to understand the brand:

Pages to fetch (try each, skip 404s):
- / (homepage)
- /about or /about-us
- /pricing
- /products or /features or /services
- /blog (index only)

Alternatively, run scripts/crawl_site.sh <domain> for a batch crawl.

Extract from crawled content:

• Core product/service offering
• Target audience (industry, company size, persona)
• Key differentiators / value props
• Competitor mentions
• Content themes from blog titles

Step 2: Discover the Topic Universe

Using the brand understanding, brainstorm topic categories. For methodology and category types, read references/aeo-methodology.md.

Core prompt categories to generate:

1. Problem-aware — "How do I solve [problem]?"
2. Solution-aware — "What tools exist for [category]?"
3. Comparison — "[Brand] vs [competitor]"
4. Best-of — "Best [category] for [use case]"
5. How-to — "How to [task the product helps with]"
6. Evaluation — "Is [brand] good for [need]?"
7. Industry — "[Industry] trends / best practices"

Step 3: Generate Prompts

For each category, generate 5-15 specific prompts people would actually ask an AI assistant.

Guidelines:

• Write naturally — how people talk to ChatGPT, not how they Google
• Be specific — include context (company size, industry, use case)
• Vary intent — research, comparison, how-to, buying decision
• Avoid jargon-heavy or unrealistic prompts

Step 4: Prioritize

Score each prompt (1-5) on:

• Relevance — How closely tied to the brand's core offering?
• Volume potential — How many people likely ask this?
• Winability — Can this brand realistically be the best answer?
• Intent value — Does this indicate buying/conversion intent?

Formula: Priority = (Relevance × 2 + Volume + Winability + Intent) / 5

Sort into Tier 1 (≥3.5), Tier 2 (2.5-3.4), Tier 3 (<2.5).

Step 5: Audit Existing Coverage

For Tier 1 prompts, use web_search with site:domain.com [topic keywords] to check if content already exists.

Rate coverage:

• Strong — Dedicated page directly answers the prompt
• Partial — Related content exists but doesn't fully address it
• None — No relevant content found

Step 6: Deliver Results

Output a structured report with:

1. Brand summary (2-3 sentences)
2. Prioritized prompt list with scores and coverage status
3. Content gap analysis (high-priority prompts with no coverage)
4. Top 5 recommended content pieces to create first

Use the output format from references/aeo-methodology.md.

Tips for Better Results

• If web_search is unavailable, the skill still works — just skip the coverage audit or have the user manually check
• For competitor analysis, crawl competitor sites too and compare topic coverage
• Re-run quarterly — AI prompt trends shift as models and user behavior evolve
• The agent's own knowledge of the industry is a valid research input — use it
• Validate with Gemini 3 Flash: After generating prompts, run Tier 1 prompts through gemini-3-flash-preview with Google Search grounding (20 samples each) to see which sources currently get cited. This tells you who you're competing against in Google's AI Overviews