Full GEO Audit

1. What a full GEO audit covers

A full GEO audit is a periodic physical for an entire surface — it walks every layer that decides whether AI engines can fetch, parse, ground, and cite your pages. Each layer is its own discipline with its own playbook: citability, crawler-access audit, schema implementation, llms.txt deployment, writing for citation, citation tracking. The audit walks them in dependency order, tags each finding with a severity, and ships a report.

The 6 layers, bottom up: access → render → structure → content → off-site authority → outcome. Order matters — a finding higher up the stack is uninterpretable if the layer beneath it failed (§3).

It sits between two adjacent workstreams:

• the heartbeat — AI Citation Tracking produces a recurring snapshot of what engines actually do; the audit consumes that snapshot and diagnoses why;
• the destination — GEO Maturity Model turns the diagnosis into a path forward.

One honest note on framing. “GEO audit” is a generic industry term; the specific 6-layer dependency ladder below is GEO Wiki’s organizing device, not an established standard. Use it because the ordering is load-bearing (§3), not because anyone ratified it.

2. Before you audit — scope, inputs, trigger

Four decisions fix what every later finding means. Get one wrong and the report is uninterpretable — the same decide-before-you-measure discipline the tracking playbook opens with.

When to run. A quarterly cadence, plus trigger events: a site migration, a redesign or SSR/render change, a robots.txt edit, a major content launch, or a known engine model/retrieval update. The cadence catches drift; the triggers catch step-changes.

Inputs on hand before you start: crawl access to production, the live robots.txt, the XML sitemap, a render-check tool (fetch-and-render, not just view-source), and the latest citation-tracking log if one exists. No tracking log is not a blocker — see §4.6.

3. The audit ladder — why order is load-bearing

The core mental model: GEO readiness is a dependency stack, not a checklist. A finding at an upper layer is uninterpretable if the layer beneath it failed.

• Perfect Schema is worthless if AI crawlers are blocked at the door.
• Great content is invisible if it does not survive fetch-time render.

So you audit bottom-up — start at the hardest gate — and you report and prioritize by severity, not by layer (§5–§6). The ladder is the page’s spine:

Each rung’s findings stay chunk-friendly and one-screen so the final report is itself extractable.

4. Running the ladder

Audit the layers in order, 1 → 6. Each layer below uses the same four-row table: the question it answers, the gate signal that decides pass / block-upward, the 2–3 highest-signal checks, and a link to the playbook with the full how-to.

4.1 Layer 1 — Access & crawlability

Authoritative UA list and robots.txt token behavior: Google’s common-crawlers doc. This layer is binary in spirit — you are either reachable or you are a Layer-1 blocker finding.

4.2 Layer 2 — Render & delivery

IndexNow protocol reference: indexnow.org/documentation. A page that 200s but renders blank to a non-JS fetch is a Layer-2 finding, not a Layer-4 content problem — classify it here or you will misdiagnose it upward.

4.3 Layer 3 — Structure & machine-readability

Validate against Google’s structured-data intro with the Rich Results Test and the schema.org Schema Markup Validator; llms.txt against the proposed spec. Note Google states AI features need no special schema — so weight structured data for entity clarity, not as an AI-eligibility lever (AI features and your site).

4.4 Layer 4 — Content & trust

This is the layer with the most published causal evidence: content-level rewrites (add citations, statistics, quotations) lifted answer visibility by up to ~40% in Aggarwal et al. 2024. Treat that as direction, not a guaranteed number — §8.

4.5 Layer 5 — Off-site authority

This is the layer practitioners most often skip and most often need: on-page perfection has a ceiling set by how well the entity is recognized and corroborated elsewhere.

4.6 Layer 6 — Outcome reconciliation

The latest tracking snapshot is the input to Layer 6; running the tracking itself is the AI Citation Tracking playbook. If no tracking log exists, Layer 6 yields exactly one finding: “stand up tracking first.” Per-engine behavior differs — reconcile against Perplexity AI, ChatGPT Search, and Google AI Overviews; the last has no per-citation API, so its outcome layer is necessarily coarser.

5. Scoring — a defensible severity model

Every finding gets a severity, and severity is anchored to the layer it sits on. A Layer-1 access failure outranks any amount of Layer-4 polish, because the polish cannot be observed until access is fixed.

A composite 0–100 readiness score is optional and explicitly demoted. It is directional, not absolute — it ships only with its weighting method attached, the same provenance discipline GEO Metrics enforces for every reported number. A bare score with no method is a rumour, not a result.

6. Prioritize — from finding list to ranked action plan

Severity is not priority. A Major finding that is cheap and certain to fix outranks a Blocker that needs a six-month migration. Re-rank the severity-tagged findings by impact × confidence × ease (ICE) so the report ends with a sequenced action list, not a pile.

The ranked plan is the input to a roadmap, not the roadmap itself: where you are is this audit; the path forward is the GEO Maturity Model.

7. The audit report deliverable

What ships, every time:

• Header / provenance — audit date, scope, declared engine set, and the tracking-log version consumed (§4.6). A report without provenance is not comparable to the next one.
• Ladder results, bottom-up — Layer 1 → 6, each pass/fail with evidence.
• Severity-ranked findings, then the ICE-prioritized action plan.
• Baseline delta — if a prior audit exists, what moved and whether it moved because you changed something or the engine did.

Re-audit discipline. A re-audit trusts unchanged lower layers and re-runs what the trigger touched (§2). State explicitly what you re-ran versus carried forward — an unstated carry-forward is how a stale “pass” survives three audits.

8. Validity threats & pitfalls

Do not ship a report without clearing every item. Each line is the checklist; the linked source has the full treatment.

• Top-down auditing — checking content first and missing a Layer-1/2 block beneath it; the most common and most expensive error.
• Bare composite score — a 0–100 with no weighting method (GEO Metrics).
• Stale tracking log treated as current — Layer 6 reconciled against last quarter’s reality (AI Citation Tracking).
• Mixing competitor-relative and absolute findings — they are different constructs; never sum them (GEO Metrics).
• One-locale audit generalized to all — Chinese and English answer surfaces are not interchangeable.
• “Fixed it” with no re-audit delta — a change is not a result until a re-audit shows it.
• Over-claiming visibility → revenue — the audit proves a visibility gap closed, not that revenue moved; the business bridge is a separate model (GEO ROI Models).
• Over-extrapolating a single-actor lift — a gain measured alone may not survive once competitors optimize against the same engine; see the caveat in Aggarwal et al. 2024 §6.

9. Further reading

• The ladder, by layer: AI Crawler Access Audit · SSR for AI Crawlers · Sitemap & IndexNow · Schema Implementation · llms.txt Deployment · Citability · Writing for AI Citation · Brand Mention Tracking
• The loop: heartbeat → AI Citation Tracking; destination → GEO Maturity Model
• Definition layer: GEO Metrics · Generative Engine Optimization
• Academic anchor: Aggarwal et al. 2024 — GEO: Generative Engine Optimization

References

Academic:

1. Aggarwal, P. et al. (2024). GEO: Generative Engine Optimization. KDD ‘24. arXiv:2311.09735 · ACM DL
2. Liu, N., Zhang, T., Liang, P. (2023). Evaluating Verifiability in Generative Search Engines. Findings of EMNLP ‘23. arXiv:2304.09848

Platform & standards documentation (verified 2026-05):

• Google Search Central — Overview of Google crawlers · Common crawlers list · AI features and your site · Structured data intro
• Validators — Rich Results Test · Schema Markup Validator
• IndexNow — Protocol Documentation
• llms.txt — proposed standard
• Perplexity — Chat Completions API Reference · OpenAI — Web Search tool (Responses API)