Methodology

v1.0 · 11 June 2026 · This page is updated whenever the method changes. See Day-1 self-audit for a worked example →

1. The 9 engines we cover (fixed list)

We commit to a fixed list. We will not silently swap engines without notice, and we will not add a 10th engine without versioning this page.

2. The 4-column methodology matrix

Every datapoint in every audit carries these 4 columns. If any cell is empty, the datapoint is rejected before the report is drafted.

3. The 30-prompt matrix (5 categories × 5–7 prompts)

The full prompt set is at /audit/self-audit-01.html. Categorically:

1. Buying intent (5–7 prompts): "best GEO audit service for [vertical] in [year]" variants. Surfaces vendor-recognition patterns.
2. Competitive (5–7): "[Vendor A] vs [Vendor B] vs [Vendor C]" variants. Surfaces comparative-recognition and opportunity gaps.
3. Methodology (5–6): "what is [concept]" / "how to do [task]" variants. Surfaces citation-source patterns for educational queries.
4. Brand-specific (5): "[your brand] [predicate]" variants. Surfaces entity-recognition status.
5. Purchase intent (5): "GEO audit pricing", "is GEO audit worth it" variants. Surfaces funnel-bottom visibility.

For Enterprise ($1,499), we extend each category to 24 prompts (120 total) and add 3 more languages: 繁中, 日本語, 한국어.

4. The 5-notebook deliverable structure

Each Standard Audit ($299) ships as 5 independent PDF notebooks rather than one long PDF. Each can be read in 5 minutes, updated independently, and shared with a different stakeholder (CMO, content lead, dev lead, legal, etc.).

1. Notebook 1 — Index. Score table, mention counts, top citation sources, competitor deltas. The one-page exec summary.
2. Notebook 2 — Intent. Per-category breakdown of which prompt types the brand wins, ties, or loses on. Tells you which queries to optimize for first.
3. Notebook 3 — Content. Top 20 cited sources across the 9 engines for your category. Tells you which publishers and aggregators the LLMs trust for your space.
4. Notebook 4 — Quotables. Verbatim phrases from LLM answers where your brand is (or is not) mentioned. Actionable for content writers — they can pattern-match the language.
5. Notebook 5 — Strategy. 5–10 prioritized actions: schema.org additions, llms.txt entries, content gaps, link targets, FAQ candidates. Each action has an effort estimate and an expected citation-rate lift.

5. The 6 anti-hallucination rules

These are the hard rules that gate the report from going to the customer. A draft that violates any of them is rewritten before delivery.

1. No absolute numbers in headlines. "5 of 30 prompts mention you" — not "You are mentioned 16.7% of the time." Numbers are exact; percentages imply false precision.
2. Every claim cites a raw JSON line. "Perplexity cited tryprofound.com on prompt #14" must link to snapshots/2026-06-11/perplexity-014.json.
3. No "best" / "only" / "guaranteed." These are superlatives. We do not assert them about our own product, and we do not assert them about competitors in customer reports.
4. Disclose engine stochasticity. Every report includes a "this is a snapshot" disclaimer with the ±10% expected re-run variance.
5. Distinguish "no result" from "no data." "Perplexity returned no result for prompt #18" is different from "We did not call Perplexity for prompt #18." We never silently conflate them.
6. No automated decisions. We score; humans (us) interpret. We do not auto-reject a brand from being audited because of low mention counts; we report the data and let the customer decide.

6. What the audit does not claim

• It does not predict future rankings. AI search engines change behaviour monthly.
• It does not claim that acting on the recommendations will produce a specific lift. The Princeton 2023 paper reports an aggregate +40% in a controlled setting; individual results vary widely.
• It does not cover every AI engine that exists. We commit to 9; we will not advertise "12+" or "all major engines."
• It does not score on ranking position within a single response (yet). v2 will introduce position-weighted scoring.

7. How to read a Clarivy report

1. Open Notebook 1 (Index) first. The score table tells you where you stand in one minute.
2. Read Notebook 4 (Quotables) next. The verbatim LLM answers are the most actionable artifact for content teams.
3. Read Notebook 5 (Strategy) last. The action list is prioritized by expected citation-rate lift, not by implementation ease.
4. Notebooks 2 (Intent) and 3 (Content) are reference material — read them when you have a specific question.

8. Versioning & change log

• v1.0 (2026-06-11) — Initial 9-engine matrix, 30-prompt set, 5-notebook structure, 6 anti-hallucination rules.

Substantive changes will be announced on the blog and reflected here.