Blog GEO: AI citation optimization audit

Blog GEO: AI Citation Optimization Audit#

Scores blog posts for AI citation readiness across ChatGPT, Perplexity, Claude, Gemini, Copilot, You.com, Google AI Overviews, and Google AI Mode as one SEO workflow, not a separate discipline. Generates a 0-100 internal AI Citation Readiness heuristic with platform-specific recommendations. The score is not a calibrated probability of citation.

Google's 2026-05-15 guidance frames generative-AI optimization as SEO: no special markup, llms.txt requirement, or separate GEO/AEO playbook is required for Google visibility. Use GEO/AEO as shorthand labels only.

Cross-reference#

This skill covers FLOW surface 3 (AI assistant citations: ChatGPT, Perplexity, Claude, Gemini, Copilot, You.com) and contributes to surface 2 (SERP plus AI Overviews). Surface mapping: skills/blog/references/flow-alignment.md.

For directly relevant AI-citation prompts (AI-supporting-pages rewrite, evidence-based quality follow-up, ChatGPT discovery, visibility prompts), see /blog flow optimize.

Evidence Discipline#

Use numeric AI-citation benchmarks only when the report includes a source block with URL, publisher, methodology, sample size, engine or version, query class, retrieval date, and expiry date. If any field is missing, label the benchmark as directional or remove the number. Default heuristics:

• Self-contained, evidence-backed explanations can aid reuse, but Google prescribes no passage-length or "chunking" requirement.
• Comparison tables with semantic headers may improve extractability, but do not cite an uplift without a dated source block.
• AI Overviews coverage is methodology-dependent: cite a dated range, not a fixed point.

Audit Process#

Step 1: Read Content#

Extract from the blog post:

• Full content text and word count
• Heading structure (H1, H2, H3 hierarchy)
• Individual paragraphs and their word counts
• FAQ sections (if present)
• Schema markup (JSON-LD, microdata, RDFa)
• robots.txt mentions or meta robots directives
• Any TL;DR or summary boxes
• Comparison tables and their HTML structure
• Numbered/ordered lists
• Definition-style formatting

Step 2: Evidence-Backed Citability (4 pts)#

Check each section between headings for AI-extractable passages:

Scoring: Count important sections meeting the evidence and completeness criteria. Do not score section length.

• 4 pts: 80%+ sections have citable passages
• 3 pts: 60-79%
• 2 pts: 40-59%
• 1 pt: 20-39%
• 0 pts: <20%

Step 3: Purpose Fit and Reader Utility (3 pts)#

Check heading format and answer structure:

Scoring:

• 3 pts: All three criteria met
• 2 pts: Two criteria met
• 1 pt: One criterion met
• 0 pts: None met

Step 4: Entity Clarity (3 pts)#

Check topic consistency and disambiguation:

Scoring:

• 3 pts: All four criteria met
• 2 pts: Three criteria met
• 1 pt: One or two criteria met
• 0 pts: None met

Step 5: Content Structure for Extraction (3 pts)#

Check for AI-extractable content patterns:

Scoring:

• 3 pts: 4-5 elements present
• 2 pts: 3 elements present
• 1 pt: 1-2 elements present
• 0 pts: None present

Step 6: AI Crawler Accessibility (2 pts)#

Check technical requirements for AI crawler indexing:

Scoring:

• 2 pts: Google crawlability/indexability is clean, and selected non-Google crawler policies match the site's stated goals
• 1 pt: Google is indexable but one selected non-Google crawler or rendering check needs review
• 0 pts: Google crawling/indexing is blocked or multiple selected crawlers are unintentionally blocked

Step 7: Platform-Specific Analysis#

Evaluate observable readiness for each declared surface. Product behavior changes by mode, query set, geography, and date, so do not infer causal preferences from a vendor sample.

ChatGPT#

• Check that material claims are source-backed and useful without depending on a specific content format.
• Treat current citation samples as non-causal context, not a listicle, freshness, or domain-authority rule.

Perplexity#

• Check crawlability, source fidelity, and material freshness when the query is time-sensitive. Verify observations with current logs or reproducible tooling before describing product behavior.

Google AI Overviews#

• Follow Google's normal SEO guidance: make content helpful, crawlable, indexable, and eligible for snippets. No special GEO/AEO markup or llms.txt is required for Google visibility.
• Measure Search and AI-feature visibility separately. Organic overlap observed in a sample does not establish a preference or guarantee inclusion.

Google AI Mode#

• Treat separately from AI Overviews in reports. Emphasize normal Search eligibility, clear page purpose, accessible text, and consistency between visible content and structured data.

Claude, Gemini, Copilot, and You.com#

• Evaluate content clarity, source accessibility, freshness, and whether robots policy intentionally allows or blocks each crawler where documented.
• Use engine-specific recommendations only when current docs, logs, or test results are available.

For each platform, provide:

• Current citability rating (High / Medium / Low)
• Specific improvements to clarity, source fidelity, usefulness, and crawlability
• Content format recommendations

Step 8: Strengthen Reusable Evidence#

For important sections that lack support, propose a self-contained improvement with a specific claim, the context needed to understand it, and a verified source or transparent original methodology. Do not pad every section, impose a word band, or manufacture statistics.

Step 9: Calculate AI Citation Readiness Score (0-100)#

Map the 15-point subcategory scores to a 0-100 display score:

Rating thresholds:

• 90-100: Excellent: highly citable by AI systems
• 70-89: Good: citable with minor improvements
• 50-69: Needs Work: significant gaps in citability
• Below 50: Poor: major restructuring needed

Step 10: Generate Report#

Output the following report:

## AI Citation Readiness Report: [Title]

**AI Citation Readiness Heuristic: [X]/100**: [Rating]

This is an internal editorial heuristic, not a calibrated probability.

### Score Breakdown
| Category | Raw | Display | Max |
|----------|-----|---------|-----|
| Evidence-Backed Citability | X/4 | X | 27 |
| Purpose Fit and Reader Utility | X/3 | X | 20 |
| Entity Clarity | X/3 | X | 20 |
| Content Structure | X/3 | X | 20 |
| AI Crawler Accessibility | X/2 | X | 13 |
| **Total** | **X/15** | **X** | **100** |

### Per-Section Citability Analysis
| Section (H2) | Purpose Clear | Self-Contained | Claim+Evidence | Ready |
|---------------|---------------|----------------|----------------|-------|
| [heading] | Yes/No | Yes/No | Yes/No | Yes/No |

### Platform-Specific Optimization
#### ChatGPT
- [specific recommendations]

#### Perplexity
- [specific recommendations]

#### Google AI Overviews
- [specific recommendations]

#### Google AI Mode
- [specific recommendations]

#### Claude / Gemini / Copilot / You.com
- [specific recommendations]

### Evidence Improvements

#### [H2 Section 1]
> [Self-contained, source-backed improvement sized to the material]

#### [H2 Section 2]
> [Self-contained, source-backed improvement sized to the material]

### Technical Recommendations
- [ ] [Technical fix with specifics]

### Priority Action Items
1. [Most impactful improvement]
2. [Second most impactful]
3. [Third most impactful]

Run `/blog analyze <file>` for full content quality scoring.

Optional: Search Performance Context (blog-google)#

If blog-google credentials include Tier 1 (GSC) and the post has a published URL:

1. Query GSC by page and query dimensions, then filter rows to the URL: python3 skills/blog-google/scripts/run.py gsc_query --property <property> --dimensions query,page --json
2. Add to platform-specific analysis:
   • Current impressions, clicks, CTR, average position
   • Search queries driving traffic to this URL
3. Check indexation: python3 skills/blog-google/scripts/run.py gsc_inspect <url> --json
4. Report indexation status, canonical selection, mobile usability.
5. If skipped, report SKIPPED: credentials unavailable or SKIPPED: unpublished URL.

Optional: AI Citation Readiness Heuristic#

For a per-engine readiness view (distinct from the 15-point AI Citation Readiness category scored by /blog analyze), run:

python3 scripts/ai_citation_score.py <file> --format markdown

It returns a non-calibrated 0-100 overall heuristic plus per-engine subscores for Google AI Overview, Perplexity, and ChatGPT, a factor breakdown, and up to three highest-impact fixes. Legacy overall_probability output is retained only as a compatibility alias.

Supporting file: skills/blog/references/flow-alignment.md#

FLOW Alignment in claude-blog#

This reference explains how claude-blog adopts the FLOW framework. Load it on demand when a user asks about FLOW, evidence standards, surface coverage, or how the blog skills route to FLOW stages.

1. What FLOW is and why claude-blog adopted it#

FLOW is an evidence-led 2026-search operating model published at github.com/AgriciDaniel/flow (CC BY 4.0 prompt content, MIT code), authored by Daniel Agrici. It treats modern discovery as a multi-surface system rather than a single SERP, and it encourages traceable support for material public claims. claude-blog v2.1.0 includes FLOW through the blog-flow sub-skill (originally integrated in v1.7.0) plus this alignment doc, which informs how every other blog skill in the suite plans, writes, optimizes, and audits content. FLOW provides the principles and prompts; claude-blog provides the production tooling that applies them at scale.

2. Evidence records in claude-blog#

Material public statistics need enough provenance for readers and editors to verify and interpret them. Useful fields include:

1. Relevant date or study period when it changes the claim's meaning.
2. Identifiable source with publisher and document title where needed.
3. Retrievable support through a stable URL, plus retrieval notes for changeable or undated sources.
4. Methodology and limitations when they affect interpretation.

The publication's citation style controls presentation. No fixed sentence form or complete field set is required for every source. Unverifiable statistics get dropped, and contradicted statistics get replaced with a verified alternative.

3. The FLOW 5-surface model#

In 2026 a query reaches a buyer through 5 parallel surfaces, often without a site visit. Coverage on each surface is planned independently.

claude-blog operationalizes surfaces 1, 2, 3, and 5 directly. Surface 4 is delegated to claude-seo. When a blog post targets a query that also surfaces in a community (Reddit thread, YouTube comment), the writer should consider dual-surface thinking: optimize the post for extraction AND consider whether the same answer should also live in the community where the query is asked.

4. FLOW stages mapped to existing claude-blog skills#

The Leverage stage (off-site authority, 1 prompt) and Local stage (11 GBP/citation prompts) are intentionally not exposed as top-level claude-blog commands. Leverage is reachable through /blog flow prompts. Local is delegated to claude-seo.

5. What claude-blog adds beyond FLOW#

• Visual media pipeline: AI image generation via Gemini (blog-image), inline SVG charts (blog-chart), Pixabay/Unsplash/Pexels integration. FLOW does not prescribe asset generation.
• Writing-persona system (blog-persona) for voice/tone profiles. FLOW is voice-agnostic.
• 5-category 100-point scoring (blog-analyze). FLOW provides quality bars but no numeric scoring.
• Topic-cluster execution engine (blog-cluster). FLOW prescribes content planning; claude-blog also executes clusters end-to-end with auto-interlinks.
• Multilingual publishing (blog-multilingual, blog-translate, blog-localize, blog-locale-audit). FLOW is single-locale by default.
• Audio narration (blog-audio). Not in FLOW scope.
• Google API integration (blog-google: PSI, CrUX, GSC, GA4, NLP, YouTube, Keywords). FLOW references measurement; claude-blog provides API tooling.
• Source-grounded research via NotebookLM (blog-notebooklm). Adjacent to FLOW's evidence discipline; provides the research side.
• CMS taxonomy sync (blog-taxonomy). Operational layer FLOW does not address.

These additions implement the FLOW principles in production tooling; they do not modify the principles.

6. When to consult this doc vs. the synced FLOW source#

• This doc (flow-alignment.md): the orchestrator and any sub-skill loads this when the user asks "what does claude-blog do for AI citations / evidence / surfaces", to give a claude-blog-specific answer with skill names and routing.
• The synced FLOW source (skills/blog-flow/references/flow-framework.md plus 30 prompts): load when applying a specific FLOW prompt or quoting the framework verbatim. CC BY 4.0 attribution is required for any quote.
• The bibliography (skills/blog-flow/references/bibliography.md): load when verifying sources for a statistic.

Last updated 2026-07-23 for claude-blog v2.1.0.