Claude for GEO/AEO: field notes for AI citation visibility
Use Claude for GEO/AEO to audit extractability, rewrite answer blocks, and QA AI-visibility signals with prompts, a skill stub, and human review.
Claude for GEO / AEO works best as a citation operator - citability audits, answer-first rewrites, entity/evidence checks, access hygiene - not as a “get cited overnight” trick and not as a GEO plugin lander. Load page + SERP/AI-answer notes + brand facts, run a fixed workflow, paste prompts (or skills when the loop is stable), and keep humans on claims. Classic SEO jobs → Claude for SEO. Skills tooling → Claude SEO skills. Code power path → Claude Code for SEO. Cluster map: Claude for marketing.
When to use Claude for GEO / AEO / when not
Use Claude when the page is (or will be) indexable and you have evidence: draft or URL, target queries, AI-answer or SERP notes, and a brand/claims pack. Strong fits: score extractability gaps; rewrite buried answers into 60–90w self-contained blocks; check entities and attribution; run pre-publish AI-visibility QA.
Don’t use Claude to promise citation rates or invent AI Overview %, Brave overlap, or “platform X will cite you if…”. Don’t ignore crawl/index fundamentals and call it GEO. Don’t treat llms.txt, magic chunking, or AI-only keyword rewrites as required Google levers. Don’t invent stats or entities. Empty context is a setup failure - not a model failure. Don’t start by “installing 40 GEO skills.”
Route elsewhere when the job isn’t citation visibility: classic audits/briefs/ops → Claude for SEO; skills catalog / use-build → Claude SEO skills; repo/CLI bulk → Claude Code for SEO.
GEO / AEO jobs map - citation visibility with Claude
Catalog by citation job, not marketplace plugin commands. Third-party /seo geo READMEs exist - verify same-day if you name one; we teach the job, not clone their score cards.
| Job | Typical inputs | Output | Graduate to skill? |
|---|---|---|---|
| Citability audit | URL/draft, target queries, AI-answer notes | Extractability scorecard + gaps | After 2–3 stable runs |
| Answer-block rewrite | Weak passage, query, allowed claims | 60–90w self-contained answer + H2 Q | High value early |
| Evidence / entity check | Draft, claims list, entity list | Attribution fixes + [GAP] facts | After prompt loop |
| Access / crawler hygiene | robots notes, render notes | Allow/block review + caveats | Optional; verify same-day |
| AI-visibility QA gate | Draft + forbidden claims + link targets | Pass/fail + rewrite bullets | Yes when weekly |
Honest Google framing: optimizing for generative AI features on Search is still SEO - useful, crawlable, clear content with evidence (AI Optimization Guide, Checked 2026-09-21). Platform notes (AI Overviews, ChatGPT, Claude, Perplexity) belong in inputs you paste - not invented universal playbooks. No citation guarantees. Labs seed: geo seo claude · US/EN 70 / KD 2 (labs-kw-2026-09-21). Slug claude for geo aeo - [GAP] no Labs item; do not invent vol/KD.
Workflow - run GEO / AEO jobs in 6–8 steps
- Inventory citation jobs - Audit, rewrite, or QA. Name one job before you prompt.
- Gather inputs - Draft/URL, target queries, AI-answer/SERP notes, brand/claims pack. Missing critical inputs →
[GAP]; do not invent fills. - Confirm fundamentals - Indexable, renderable, not blocked for relevant crawlers. Claude cannot invent crawl access.
- Pick the smallest reliable surface - Chat for spikes; Projects for brand + claims + rewrite loops; Claude Code for bulk files/agents; skills after the loop is boring; MCP only as a verified data pipe.
- Run citability audit - Prompt below → prioritized extractability gaps.
- Rewrite answer blocks - Fix buried or vague sections; keep the allowed-claims list honest.
- AI-visibility QA + human review - Pass/fail before publish. Facts, brand, legal stay human-owned. Label output draft until approved.
- Graduate or prune - Weekly boring loop → thin skill → Claude SEO skills + install-first. Bulk repo QA → Claude Code for SEO. Prune overlaps with skill-doctor.
Prompts (≥3 copy-paste)
Three job-tied specs. The full prompt mirror is kept with the editorial source. Discipline pattern: Claude prompts for marketing.
1. Citability audit
Inputs: URL/draft · queries · AI-answer notes. Output: CRITICAL/HIGH/MEDIUM extractability table. When not: inventing crawl data; promising citations.
You are a GEO/AEO citation analyst. Score extractability from PROVIDED evidence only.
REQUIRED INPUTS: URL or draft excerpt; Target queries; AI-answer or SERP notes (or "[GAP]"); Constraints (or "none").
STEPS: (1) Restate the citation job in one line. (2) Per query: self-contained answer block? evidence/attribution? answer buried? H2/question match? (3) Table: severity (CRITICAL|HIGH|MEDIUM) | gap | evidence quote | fix direction | how would we know this failed? (4) Top 5 rewrite priorities. (5) Mark [GAP] where notes are thin.
RULES: Never invent citation rates, AI Overview %, Brave overlap, rankings, or Labs volumes. Never invent marketplace skill/plugin names. Never promise "you will be cited." Do not sell llms.txt / magic chunking / AI-keyword rewrite as required Google levers.
OUTPUT: severity table | top 5 priorities | [GAP] list | QA self-check.
2. Answer-block rewrite
Inputs: weak passage · query · allowed claims. Output: 60–90w extractable opener + optional question H2. When not: empty claims; inventing sources.
You are a citation editor. Rewrite a weak passage into a self-contained answer block.
REQUIRED INPUTS: Weak passage; Target query; Allowed claims; Must-not list; Preferred length (60-90 words default).
STEPS: (1) Write a 60-90w extractable answer that stands alone. (2) Optional question-form H2 matching the query. (3) Attribute only from the allowed list; mark missing facts [GAP]. (4) List must-not violations avoided. (5) One evidence placeholder (URL or "human supply").
RULES: No invented stats, entities, or platform guarantees. No thin extractable spam.
OUTPUT: answer block | optional H2 | [GAP] facts | must-not check | evidence placeholder.
3. AI-visibility QA gate
Inputs: draft · forbidden claims · live link targets. Output: pass/fail + rewrite bullets. When not: autopilot publish.
You are a pre-publish AI-visibility QA gate for GEO/AEO drafts.
REQUIRED INPUTS: Draft; Forbidden claims; Live internal-link targets (existing paths only - or "none"); Platform notes (or "[GAP]").
STEPS: (1) Pass/fail: extractable answer-first opener; evidence for declarative claims; no invented metrics; no dead internal links; no myth levers as required Google factors (llms.txt-as-magic, magic chunking, AI-only keyword rewrite). (2) Flag undated "platform said X". (3) Rewrite bullets only - do not expand into a full SEO audit. (4) Route classic SEO → Claude for SEO; skills catalog → Claude SEO skills; Code bulk → Claude Code for SEO.
RULES: Never invent citation rates or marketplace names. Label output draft.
OUTPUT: pass/fail table | rewrite bullets | [GAP] list | QA checklist.
Agent block - GEO/AEO citation ops runner
Paste into a Project custom instruction, Claude Code agent, or system prompt:
You are the GEO/AEO citation ops runner for this Project.
MATCH each request to one job: citability | rewrite | QA.
REQUIRE inputs (URL/draft, queries, AI-answer/SERP notes, claims pack) or mark [GAP] - never invent fills.
NEVER invent citation rates, AI Overview %, Brave overlap, rankings, Labs volumes, or marketplace skill/plugin names.
NEVER promise "you will be cited" or sell llms.txt / magic chunking / AI-keyword rewrite as required Google levers.
ROUTE classic SEO jobs → /guides/claude-for-seo; skills catalog → /guides/claude-seo-skills; Code power path → /guides/claude-code-for-seo.
OUTPUT: recommendation + next action + QA checklist. Label all copy draft until a human approves.
Skills path / stub - geo-citability-audit
When the citability loop is boring weekly, graduate to a thin skill. Stub: geo-citability-audit in the editorial source pack, not a zip yet. Later alias: aeo-answer-block-qa.
Surfaces: Projects custom skills · Claude Code .claude/skills/ · marketplace (verify same-day). Bridges: Claude SEO skills · install-first · marketplace · skill-doctor. Agent Skills: platform docs (Checked 2026-09-21). Do not catalog third-party GEO plugin SKUs as the H1 promise.
Stack glance
| Surface | Best for | Role in GEO/AEO |
|---|---|---|
| Chat | One-off citability questions | Prompt only |
| Projects | Brand + claims + rewrite loops | Files + custom skills |
| Claude Code | Bulk page/repo QA, agents | → Claude Code for SEO |
| Skills | Repeatable citation SOPs | After loop stable |
| MCP | Live crawl/SERP data | Verify before decisions |
Pitfalls + quality bar
Promising citations · inventing AI-visibility scores · skipping indexability · hallucinated stats/entities · autopilot publish · skill bloat (install-first + skill-doctor).
Quality bar: evidence fields · no invented metrics · extractable answer-first · human QA · live internal links only · Google-aligned myth hygiene.
FAQ
What is Claude for GEO / AEO?
A citation-ops runner for AI-answer visibility - audits, answer blocks, evidence checks - with fixed inputs and human QA. Not a plugin SKU and not a citation guarantee.
Is GEO/AEO different from SEO?
Useful labels for AI-answer surfaces. For Google, generative AI optimization is still SEO (AI Optimization Guide, Checked 2026-09-21). Claude runs the citation jobs on top.
Claude chat vs Claude Code for GEO work?
Chat/Projects for rewrite/QA loops. Code when bulk files/agents matter → Claude Code for SEO.
Do I need GEO skills installed first?
Usually no. Stabilize prompts in a Project, then graduate → Claude SEO skills · install-first.
How is this different from Claude for SEO / SEO skills / Code for SEO?
This page = AI citation visibility. Claude for SEO = classic jobs. Claude SEO skills = skills catalog. Claude Code for SEO = Code power path.
Where to go next
Run one citability audit today with a real page, queries, and AI-answer notes. Classic jobs: Claude for SEO. Skills: Claude SEO skills. Code path: Claude Code for SEO. Map: Claude for marketing. Takeaway: citation jobs > plugin landers · inputs required · prompts then skills · no guarantees · human QA.