---
name: research
description: Structured topic research with freshness checks, source diversity, and an honesty pass. Outputs a brief that ends in concrete content angles for Jake's site — turns research into content fuel, not abstract knowledge.
trigger: /research
---

# /research

Most "research" sessions are a vibe check — search a few terms, skim three articles, form an impression. That's fine for casual reading. It's bad for content production, where you'll publish based on what you find. Research forces structure: multiple sources, dated, contested points called out, hype flagged, and the final brief lands on specific content angles you could write *next week* — not in a year.

The skill is built for Jake's actual job: feeding a content engine for Elite AI Advantage (AI consulting, GEO/SEO, AI for SMBs). The "Content angles" section at the end is the deliverable. Everything above it is the work.

## Usage

`/research <topic>`

Examples:
- `/research GEO (generative engine optimization) for local services`
- `/research AI agents replacing SDR teams in 2026`
- `/research AI-driven content moderation pitfalls`
- `/research What's actually working for AI SaaS pricing right now`

## What it's for

You ship blog posts and how-tos. The hard part is not writing — it's choosing what to write about with enough conviction that you don't end up sounding like every other AI consulting site. Research solves the *finding* problem: structured discovery of the topics where you'd have something specific to say.

The output is calibrated to your audience: SMB owners, marketing directors, people who can buy a $25k engagement. Not other AI engineers. Not enterprise CIOs. Specific.

## What You Must Do When Invoked

### Step 1 — Plan the search

Before searching, write 3–5 specific search queries that will surface different angles. Don't just paraphrase the topic. Think:
- One query for **technical claims** (what does it actually do)
- One query for **case studies / real deployments**
- One query for **counter-arguments / skeptics**
- One query for **recent news** (last 90 days)
- One query for **practitioner discussion** (Reddit, HN, Twitter, blog comments)

Print the planned queries before running them. This is auditable.

### Step 2 — Search and read

Run the queries via WebSearch and/or WebFetch. For each candidate source:
1. **Get the publish/update date.** No date → low trust until you find one elsewhere.
2. **Identify the source type.** Vendor blog, independent journalism, academic paper, practitioner post, official docs.
3. **Read enough to get a position**, not just headlines.

Aim for at least 5 sources, ideally from at least 3 distinct domains. If you only have one source, the research isn't done.

### Step 3 — Honesty pass

Before writing the brief, force three checks:

1. **Freshness check:** Is anything older than 12 months still being treated as current? In AI specifically, anything pre-2025 is stale unless it's foundational. Flag staleness explicitly.
2. **Diversity check:** Are all your sources from the same ecosystem (e.g. all OpenAI marketing material, all VC blog posts)? If yes, the brief is biased — note it.
3. **Hype check:** Which claims have multiple independent sources? Which are single-source vendor claims? Which are speculation dressed as forecast?

Be willing to say "I couldn't find good evidence for X" — that's more useful than confident filler.

### Step 4 — Write the brief

Output exactly this structure. Aim for 800–1,500 words total — long enough to be useful, short enough to read in 5 minutes.

```markdown
# Research: <topic>

**Run:** <YYYY-MM-DD> · **Queries:** <count> · **Sources:** <count> from <distinct-domain-count> domains

## Headline finding
<1–2 sentences. The one thing worth knowing if you read nothing else. Specific, not generic.>

## What's solid
<Bullet list. Claims with multiple independent sources. Cite each: "(Source: <name>, <date>)". 4–8 bullets.>

## What's contested
<Where credible sources disagree. Name the disagreement and both sides. Don't take a side here — that's the user's job. 2–4 bullets.>

## What's hype
<Claims that don't survive scrutiny. Vendor-only claims, speculation, repeated assertions without evidence. 2–4 bullets.>

## What's missing
<Honest list of what you couldn't find solid evidence for. This is more useful than it sounds.>

## Freshness notes
<Anything older than 12 months. Anything that's likely to age out within 6 months. Anything where the field is moving fast enough that this brief decays.>

## Sources
<Numbered list. For each: title, domain, publish/update date, 1-line characterization.
Example: "1. 'GEO is dead, long live AEO' — searchengineland.com — 2026-02-14 — independent journalism, contrarian">

---

## Content angles for Elite AI Advantage

<This is the deliverable. 3–5 specific blog/post angles ranked by signal.

For each angle:
- **Working title:** <short, specific>
- **The point:** <1–2 sentences. The thesis Jake would actually argue.>
- **Why it's good:** <1 sentence. What makes it different from generic AI-blog content.>
- **What you'd need:** <case study, interview, original data, none.>

Rank by how distinctive Jake could make the post. Generic explainers go last. "I'd argue X because Y, contra the consensus" angles go first.>
```

### Step 5 — Stop

End on the content angles. Do not start writing one of the posts. Do not propose follow-up research without being asked. The skill outputs a brief. Jake decides what to do with it.

## Calibration for Jake's audience

The "Content angles" section MUST be calibrated to Jake's actual readers:

- **Elite AI Advantage audience:** SMB owners, marketing directors at mid-market companies, agency owners. They have budget but limited time. They've been sold AI snake oil before and are appropriately skeptical.
- **They want:** specific tactics that work for businesses their size, honest takes on what's hype, real ROI numbers when possible, anti-trend takes when the trend is bad.
- **They don't want:** generic "what is AI" explainers, enterprise-only frameworks, vague futurism, more "ChatGPT changes everything" posts.

Angles that lean into Jake's voice: contrarian, tactical, money-numbered when possible, willing to call out when an industry trend is bullshit.

## What to avoid

- **No vibes-only research.** Every "What's solid" claim needs a source. Every "What's hype" claim needs a reason.
- **No SEO-think in the angles.** The brief is for Jake, not for Google. He's not asking what to rank for — he's asking what to argue.
- **No "more research needed" cop-out for the headline finding.** If you don't have a finding, say what you couldn't find and why.
- **No paraphrasing entire articles.** Cite, don't reproduce. (Also: copyright.)
- **No "ChatGPT says…" sourcing.** LLM output is not a source. Find a human-authored or organization-authored piece.

## Example output (abbreviated)

```markdown
# Research: GEO for local service businesses

**Run:** 2026-04-24 · **Queries:** 5 · **Sources:** 8 from 6 domains

## Headline finding
GEO (generative engine optimization) for local services is mostly conventional local SEO with three additions: schema density, citation freshness, and AI-overview-friendly answer snippets. Vendors are over-selling it as a new discipline; the practitioners who measure carefully treat it as an extension of existing local SEO.

## What's solid
- Google's AI Overviews now answer ~30% of local-intent queries without a click. (Source: BrightLocal Local Consumer Review Survey, 2026-Q1)
- Schema markup (LocalBusiness, FAQPage) increases AI Overview citation rates measurably. (Source: Search Engine Land, 2026-03; SEMRush blog, 2026-02 — independent confirmation)
- ...

## What's contested
- Whether GEO requires fundamentally different content than SEO. Vendor camp says yes; independent practitioners (Marie Haynes, Britney Muller) say no — same fundamentals, different surface.
- ...

## Content angles for Elite AI Advantage
1. **Working title:** "GEO is mostly local SEO with extra steps — here's the actual three things that changed"
   **The point:** Push back on GEO-as-new-discipline narrative. Map specifically which 3 things are genuinely new (schema density, citation freshness, answer-snippet structure) and what's just rebranded local SEO.
   **Why it's good:** Anti-hype, concrete, useful to a service business owner deciding whether to buy a "GEO audit."
   **What you'd need:** Nothing — opinion piece grounded in this research.

2. **Working title:** "We tested whether AI Overview citations actually drive bookings. Here's what we found."
   **The point:** ...
```

That's the shape. Structured, sourced, ends in money-makers.