Solnest AEO: AI Visibility Audit

Start here: what actually moves AI citations

Do not skip this section. Most AEO advice on the internet is selling work that the controlled studies say does nothing. Get the ordering right and you save the client both money and disappointment.

Google's own position (developers.google.com/search/docs/fundamentals/ai-optimization-guide, updated June 2026):

"From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO."

Google states plainly that you do not need llms.txt files, AI-specific content rewriting, special schema, or content chunking. Its wording on llms.txt is that such files "neither harm nor help."

How to talk to a client about this. Never promise placement. LLMs are non-deterministic, and the same prompt returns different brand lists to different users. You are raising the probability of citation, not buying a slot. Anyone guaranteeing inclusion is selling something they cannot deliver.

Step 1: Crawler access, verified at the edge

This is where most sites lose before they start, and it is the single highest value thing in this skill.

Training crawlers and search crawlers are different bots. Blocking the training bot does not remove you from that engine's answers. Blocking the search bot does.

Check robots.txt, then check the edge. A permissive robots.txt proves nothing if Cloudflare, a WAF, or a bot-management rule returns 403 to these user agents before the request reaches the origin. This failure mode is common and completely invisible in robots.txt.

Run the check with the bundled runtime:

"${CLAUDE_PLUGIN_ROOT}/scripts/claude-seo" run aeo_crawler_check.py <url>

Also confirm the page is indexable and snippet-eligible. noindex, or max-snippet:0 / nosnippet, removes Google AI Overview eligibility outright.

Step 2: Entity consistency

AI systems merge your brand into a single entity across sources. Inconsistent naming splits that entity and dilutes every mention you earn.

Check that the legal or trading name is byte-identical across:

• <title> and the organization name in JSON-LD (Organization / LocalBusiness)
• llms.txt, if one exists
• Open Graph og:site_name
• Google Business Profile, and any directory listings
• The footer NAP block

Diff the machine-readable surfaces against each other:

"${CLAUDE_PLUGIN_ROOT}/scripts/claude-seo" run aeo_entity_check.py <url>

The two failure modes worth knowing, both seen in the wild:

• Truncation. og:site_name and llms.txt declare Acme while the title says Acme Property Group. A retrieval system can read that as two entities.
• Near-miss typo. One surface carries Acme Partners LPP against Acme Partners LLP everywhere else. One character, and the brand splits.

Step 3: Citability scoring

Delegate the scored analysis to the seo-geo sub-skill, which carries the full rubric: passage citability (optimal self-contained answer blocks of roughly 134 to 167 words), question-based heading hierarchy, attribution density, structural readability, multi-modal presence, and entity presence across Wikipedia, Reddit, YouTube and LinkedIn.

Then apply the Solnest weighting on top when reporting:

Step 4: Measure honestly

Track the trend, never the single reading. Prompt-level "rank" from any AI visibility tool is one roll of a non-deterministic die.

Measure:

1. Share of voice across a fixed prompt set that matters commercially, re-run on a fixed schedule, always the same prompts.
2. AI referral traffic in analytics, segmented from organic.
3. Branded search volume, the closest available proxy for whether the models are learning who the brand is.
4. Rankings, because retrieval still runs through the index.

Expect nothing for six to eight weeks. Movement shows up on long-tail queries before head terms.

Set expectations on traffic. Pew tracked 68,879 real Google searches and found users clicked a result on 8% of visits where an AI summary appeared, versus 15% without. AI sends little traffic today. What it sends converts unusually well, because the model already did the shortlisting. Sell this as a recommendation channel, not a traffic channel.

Output files

Persist the audit. Chat output disappears; a client deliverable should not.

Write into the same {domain}-audit/ directory the seo-audit skill uses, so the SEO and AEO halves compose into one folder and one report:

• {domain}-audit/AEO-REPORT.md: the readable audit. Crawler access table, entity diff, citability assessment, and the evidence-ordered recommendations.
• {domain}-audit/aeo-data.json: structured results, in the shape below.

Create the directory if it does not exist. If a seo-audit run already produced one, write alongside its files rather than overwriting anything.

{
  "url": "https://example.com",
  "audited_at": "YYYY-MM-DD",
  "band": "Blocked|Invisible|Readable|Citable",
  "crawler_access": {
    "robots_readable": true,
    "blocking_issues": [],
    "crawlers": [
      {"crawler": "OAI-SearchBot", "role": "search", "engine": "ChatGPT",
       "robots_allowed": true, "edge_status": 200, "edge_blocked": false}
    ]
  },
  "entity": {
    "consistent": true,
    "surfaces": {"title": "", "og:site_name": "", "jsonld:Organization": "", "llms.txt": ""},
    "mismatches": []
  },
  "citability": {
    "extracted_words": 0,
    "longest_paragraph_words": 0,
    "blocks_over_40_words": 0,
    "percentages": 0,
    "dollar_figures": 0,
    "question_form_headings": "0/0",
    "snippet_eligible": true
  },
  "findings": [
    {"title": "", "severity": "Critical|High|Medium|Low|Info",
     "description": "", "recommendation": ""}
  ]
}

Both bundled scripts accept --json, so capture their output directly rather than retyping their results:

"${CLAUDE_PLUGIN_ROOT}/scripts/claude-seo" run aeo_crawler_check.py <url> --json
"${CLAUDE_PLUGIN_ROOT}/scripts/claude-seo" run aeo_entity_check.py <url> --json

After writing the files, tell the user where they are. If audit-data.json from a seo-audit run is present in the same directory, offer to merge the AEO findings into it so they appear in the generated PDF or HTML report.

Reporting

Lead with what is broken and free to fix (crawler access, entity splits), then what is earned and slow (mentions, coverage). For every recommendation state how you would know it failed. If a finding rests on a correlation rather than a controlled study, say so in the report.

If asked to add llms.txt or bolt on FAQ schema "for AI", explain the evidence, note that it is cheap and harmless, and put it below the levers that actually move. Do not lead a deliverable with it.