AI Search Visibility Auditor

Can ChatGPT, Claude, Perplexity, Gemini and Copilot find, read and cite your website? And when customers ask AI for recommendations, do they hear your name, or your competitor's?

This Actor audits any website for AI search (also called GEO / AEO / LLM SEO) and, optionally, measures your brand's share of voice in real AI answers. You get a score, a prioritised fix list, a ready-to-upload llms.txt, and a trend over time.

What it checks

1. Can AI crawlers get in?

• robots.txt per AI bot: OAI-SearchBot, ChatGPT-User, GPTBot, Claude-SearchBot, Claude-User, ClaudeBot, PerplexityBot, Perplexity-User, Googlebot, Google-Extended, Bingbot, Applebot-Extended, Meta-ExternalAgent, CCBot, Amazonbot, Bytespider. Parsed per RFC 9309 (longest match, wildcards, $).
• Search bots vs training bots: blocking search bots removes you from AI answers (critical). Blocking training bots is a business choice (reported, low severity).
• Live firewall/CDN test: fetches your home page with each bot's user agent. This catches the very common case where robots.txt allows AI bots but Cloudflare/WAF "Block AI bots" or Bot Fight Mode returns 403 or a challenge.

2. Can they read the page?

• JavaScript rendering: most AI crawlers do not run JavaScript. Pages that are an empty React/Vue/Angular shell are flagged critical.
• noindex, nosnippet, max-snippet:0, noai, X-Robots-Tag headers, canonical pointing elsewhere, HTTP errors, slow responses, huge HTML.

3. Can they understand and quote it?

• Title, meta description, single H1, subheadings, lang.
• Structured data (JSON-LD + microdata): Organization/LocalBusiness on the home page with sameAs, plus Article, Product, Service, FAQPage. Invalid JSON-LD is flagged.
• Citability signals: FAQ / question-style headings, concrete figures (prices, %, years, customers), lists and tables, author, last-updated date and content age, outbound references.
• Site files: llms.txt (validated against llmstxt.org), llms-full.txt, XML sitemap and <lastmod>, HTTPS.

4. (Optional) Are you in the answers?

Ask AI models the questions your customers ask (e.g. "Best solar installer in Nairobi?"). For every answer the Actor records whether your brand is mentioned, its position, which competitors appear, other brands AI recommends that you are not tracking, and whether your domain is cited. It summarises:

• brand mention rate, average position, top-3 rate
• share of voice vs your competitors
• per-question wins and gaps
• change since last run (same Monitor name): questions gained and lost

Works with any OpenAI-compatible API using your own key: OpenAI, or OpenRouter to query Claude, Gemini, Llama, Mistral and Perplexity models through one key, plus Groq, Together, Azure or a local model.

Output

Where	What
Dataset, view Site scores	One row per site: score 0–100, grade A–F, blocked AI search/training bots, pages needing JS, top 15 fixes with how-to
Dataset, view Page audits	One row per page: score, words, JS needed, indexable, schema types, FAQ, facts, freshness, issues + fixes
Dataset, view AI answers	One row per answer: question, model, brand mentioned/position, competitors, other brands, sources, answer text
REPORT	Shareable HTML report
LLMS_TXT_<domain>	Generated llms.txt to upload to your site root
AI_VISIBILITY	Share-of-voice summary and changes
SUMMARY	Compact JSON for automations

Example site row (shortened):

{
  "record_type": "site", "site": "https://www.example.co.ke", "ai_visibility_score": 58, "grade": "D",
  "blocked_ai_search_bots": ["PerplexityBot"], "blocked_ai_training_bots": ["GPTBot"],
  "pages_audited": 20, "pages_needing_javascript": 6, "llms_txt": { "found": false },
  "top_fixes": [{ "severity": "critical", "check": "javascript", "example": "Only 12 words visible without JavaScript…", "fix": "Use server-side rendering or pre-rendering…", "pages": 6 }]
}

How the score works

Each page starts at 100 and loses points per issue (critical 30, high 15, medium 8, low 3), with small bonuses for FAQ, rich schema, concrete facts and authorship. The site checks score works the same way. The site score = 50 % site checks + 50 % average page score. The AI answer score = 70 × mention rate + up to 30 for a high average position.

Use cases

• Agencies: sell a "Can ChatGPT find you?" audit; run it on a prospect and three competitors and send the HTML report.
• Marketing teams: schedule weekly brand tracking with a webhook to Slack and catch drops early.
• Developers: verify a Cloudflare or WAF change didn't lock out AI search bots; check SSR after a frontend migration.

Tips

• Start with 20 pages per site. Raise it to 100+ for a full content audit.
• Use 3–5 samples per question for stable share of voice. AI answers vary.
• Ask questions the way customers ask them, including location and budget.
• The live bot test uses the bots' user agents from Apify's IP addresses. Some firewalls verify real bot IPs, so a site can pass here and still block real bots (and vice versa). Treat a block as a strong signal to check your CDN settings.

Limits & responsible use

• Only public pages are fetched, at a gentle rate, and robots.txt is respected by default.
• AI answers are a sample in time, not a guarantee of what every user sees.
• This Actor does not send your API key anywhere except the base URL you set.

Pricing

Pay per event: $0.02 per website, $0.004 per page, $0.005 per AI answer (see PRICING.md). Your AI token costs are billed by your provider.