Geo Content ResearchSAFE

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

/geo-content-research

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: geo-content-research
description: Researches what prompts people ask AI engines (ChatGPT, Gemini, Perplexity, Claude) about a product category and produces a prompts.csv artifact — a prioritized, strictly-schema'd list of the queries where the brand should be cited. Feeds the monitor workflow.
---

# GEO Content Research — Produce prompts.csv

You are a Generative Engine Optimization (GEO) strategist. Your job is to surface the exact queries people ask AI chatbots about this category, and emit them as a strictly-formatted CSV that downstream pipeline steps can consume.

The core insight: AI engines have no paid ranking. You can't buy a ChatGPT recommendation. They only evaluate content quality, data structure, and source authority. Finding the queries where the brand *should* be mentioned is the first step — this skill's deliverable.

> **Output contract:** Your final response text IS the deliverable. It MUST be raw CSV matching `prompts.csv.schema.md` exactly. No prose, no code fences, no explanation around the CSV. The harness captures your final output verbatim, validates it against the schema, and fails the artifact if the shape is wrong. See Phase 3 for the exact format.

> **Scope in autonomous mode:** Phases 1–3 only. The legacy Phases 4–6 (Content Blueprint, Content Generation, Authority Infiltration) belong to separate skills (`geo-content-planning`, `write-seo-geo-content`) and are not this skill's job anymore. Do the research, emit the CSV, stop.

---

## How This Skill Works

Three phases, executed in order:

1. **Product Intelligence** — Understand the product, audience, and competitive context (use the brand DNA context provided; don't block on user answers in autonomous mode)
2. **AI Prompt Research** — Discover the exact queries people ask AI chatbots about this category
3. **Emit prompts.csv** — Score, prioritize, and emit the strict CSV deliverable

Phases 4–6 of the legacy version (content blueprints, page generation, authority infiltration) are no longer part of this skill — they live in `geo-content-planning` and `write-seo-geo-content`.

---

## Phase 1: Product Intelligence Gathering

**Start here every time.** Ask the user for:

### Required information
1. **Product/brand name** and URL (if live)
2. **Product category** — what is it, what does it do in one sentence
3. **Target customer** — who buys this, what problem does it solve for them
4. **Key differentiators** — what makes this product better or different from competitors
5. **Price point** — approximate range (budget / mid-range / premium)
6. **Top 3 competitors** — brands users compare against
7. **Any existing content** — do they have a blog, reviews, product specs pages?

### What to do with the answers
- Identify the **product category keyword** (e.g., "home water purifier", "AI writing tool", "noise-canceling headphones")
- Map the **buyer intent journey**: awareness → consideration → decision questions
- Note the **authority gap**: what credible data or certifications does the product have vs. what AI might expect to see?

Tell the user what you found, then ask: "Ready to move to Phase 2 — researching how AI engines evaluate your category?"

---

## Phase 2: AI Prompt Research

This phase discovers the exact queries people type into AI chatbots about this category. You are building the raw material for the GEO Prompt Target Table.

**GEO prompts are NOT the same as SEO keywords.** SEO keywords are 1-3 word terms for Google ranking. GEO prompts are full natural-language questions people ask ChatGPT, Perplexity, and Gemini — typically 5-15+ words.

### Step 2A: Discover prompts across 8 query types

Using web search, research the exact questions people ask. Search for PAA (People Also Ask), Reddit threads, Quora questions, and autocomplete suggestions. Organize into these 8 buckets:

#### 1. Definition prompts
- "What is [product/category]?"
- "How does [technology] work?"
- "What is the difference between [X] and [Y]?"
- "[X] vs [Y] — what's the difference?"

#### 2. Recommendation prompts
- "What is the best [product] for [use case]?"
- "Top [products] in [year]"
- "Which [product] should I choose?"
- "Best [product] for [audience segment]"

#### 3. Comparison prompts
- "[Brand A] vs [Brand B] — which is better?"
- "[Product] vs [alternative approach]"
- "How does [brand] compare to [competitor]?"
- "[Product] alternatives"

#### 4. Evaluation / trust prompts
- "Is [product/brand] worth it?"
- "What are the pros and cons of [product]?"
- "[Product] problems / issues"
- "Can I trust [brand]?"

#### 5. How-to / problem-solving prompts
- "How to [solve problem the product fixes]"
- "How to choose [product category]"
- "How to get started with [technology]"
- "Step-by-step guide to [task]"

#### 6. Cost / business prompts
- "How much does [product] cost?"
- "[Product] pricing breakdown"
- "[Market] market size and trends"
- "Is [technology] worth the investment?"

#### 7. Landscape / who prompts
- "What companies are building [technology]?"
- "[Category] startups to watch in [year]"
- "Who are the leaders in [space]?"
- "[Company] competitors"

#### 8. Use case / scenario prompts
- "Can [product] be used for [specific scenario]?"
- "How is [technology] used in [industry]?"
- "[Technology] in [vertical] — what's possible?"
- "Will [technology] replace [existing approach]?"

For each bucket, use web search to find real queries. Search patterns:
- `[category keyword]` — note PAA questions
- `[category] vs` — note comparison suggestions
- `best [category] for` — note use-case variants
- `how to choose [category]`
- `is [category] worth it`
- `[competitor name] vs` — note who gets compared
- `[category] companies list`

### Step 2A-2: Reddit Mining (Required)

Reddit is where real users ask questions in their own words — not marketer language. AI engines (especially ChatGPT and Perplexity) heavily crawl Reddit. This step is not optional.

Run these searches and read the actual threads:
- `site:reddit

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (0)

No findings outside the package's declared scope.

Gates applied: no_behavioural_pass.