Research Procedures¶
AEO/GEO Researcher runs a three-step pipeline for comprehensive sessions. Each step has a defined purpose and output. Simple, scoped sessions can enter at any step.
The Pipeline¶
01 · Signal Scan → 02 · Deep Investigation → 03 · Source Audit
01 · Signal Scan¶
Purpose: Map what's present and absent before going deep. Identify the highest-priority angle.
When to use: The client's AI visibility landscape is unknown. The question is broad. This is the first session with a client.
When to skip: The client has a specific, well-scoped question and the Orchestrator has already identified the right specialist.
Output: A triage document — 6 quick observations across crawl access, schema, platform visibility, content structure, off-site presence, and competitive landscape. Every scan observation is unconfirmed by default. Ends with a recommended specialist for Procedure 02.
02 · Deep Investigation¶
Purpose: Specialist research on the specific angle. This is where findings are produced.
How it works: The Orchestrator routes to the right specialist based on the client's angle. The specialist runs its full investigation sequence, tags every finding with verification status and source credibility tier, and closes with a research ledger.
What the research ledger surfaces:
- What is confirmed and what it rests on
- What was inferred and what would confirm it
- What would change the conclusion if new information emerged
Ends with: An explicit offer of Procedure 03 for any findings the client plans to act on.
03 · Source Audit¶
Purpose: Adversarial review of Procedure 02 findings before the client acts on them.
Posture: The auditor is not the specialist. The specialist produces findings to answer the question. The auditor challenges those findings to stress-test them.
What the auditor checks:
| Check | What it catches |
|---|---|
| Citation traceability | Statistics without a named, reachable source |
| Confirmation status accuracy | confirmed findings that rest on inference, not direct observation |
| T4/T5 reliance | Vendor statistics treated as benchmarks without a T1–T3 corroborator |
| Platform behavior staleness | Platform claims without an "as of" date qualifier |
| Uninvestigated gaps | Dimensions the specialist didn't look at that could change the priority order |
The standard: A clean audit with zero findings is a failed audit. If everything checks out, the auditor didn't look hard enough.
Output: List of findings that passed, findings that were flagged with specific remediation, gaps not surfaced, and a revised priority if the audit changed the order.
Entering at Any Step¶
| Client situation | Entry point |
|---|---|
| "We don't know why we're not showing up" | Start at 01 — landscape is unknown |
| "We think it's a technical issue with our schema" | Skip to 02 → 1-signal-technical |
| "We have findings from a previous session — can you stress-test them?" | Skip to 03 directly |
| "We know our gap is content structure for Perplexity" | Skip to 02 → 3-platform-intelligence then 4-content-gap |