How AI Search Engines Score Your Site Differently
Run any website through a modern AI visibility audit and you will get five different scores — one for ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews — even though nothing about the page changed between measurements. That is not a bug. Each engine reads the web through a different lens, so the same site can be loud on one platform and quiet on another.
Why a single score is misleading
An overall "AI visibility" number averages everything the crawlers care about. But engines disagree on what matters. A site that nails /llms.txt and content structure will score high on ChatGPT and Claude, yet get near-zero on Google AI Overviews if it has no structured data and no E-E-A-T signals. The overall score hides that gap.
Per-platform scores expose it. They re-weight the same underlying checks using what each engine publicly weights most.
What each engine prioritizes
Based on public specifications, crawler behavior, and the quality guidelines each parent company has published, this is how LLCrawler weights the six signal categories per platform:
| Signal | ChatGPT | Claude | Perplexity | Gemini | AI Overviews |
|---|---|---|---|---|---|
| AI readiness (llms.txt, bots) | High | High | Med | Low | Low |
| Content structure | Med | High | High | Med | Med |
| Structured data (JSON-LD) | Med | Med | Med | High | High |
| E-E-A-T signals | Med | Med | High | Med | High |
| Brand presence | Low | Low | Med | Low | Med |
| Meta / social | Low | Med | Low | Low | Med |
| Performance | Low | Low | Low | Med | Med |
| Sitemap | Low | Low | Low | Med | Low |
The short version:
- ChatGPT and Claude lean on the AI-native specs — a clean
/llms.txt, open robots.txt, readable content without JS. - Perplexity is a citation engine. It rewards E-E-A-T signals (author, date, credentials) because its answers literally cite sources.
- Gemini and Google AI Overviews are Google-shaped. They pull heavily from Schema.org, sitemap quality, and E-E-A-T — the same inputs Google has always valued.
What to do with per-platform scores
Treat them as a diagnostic, not five separate targets. If ChatGPT is at 95 but Google AI Overviews is at 70, you do not need a different site — you need to fix the category the low platform weights heaviest. That will usually be structured data, E-E-A-T, or performance.
Three rules that hold across engines:
- A valid
/llms.txtnever hurts and boosts three of the five platforms. - JSON-LD on the homepage — Organization, WebSite, and BreadcrumbList at minimum — lifts every Google-family engine.
- An author byline and updated date on every page you want cited lifts Perplexity and AI Overviews simultaneously.
Not every signal applies to every site
A local car dealer does not need a Wikipedia page to rank on Perplexity, and a solo Substack writer does not need a Crunchbase profile. The brand-presence checks are a menu, not a mandate: pick the ones your industry uses.
- Local business: Google Business Profile, Yelp, industry directories (Avvo for lawyers, Houzz for builders, Zillow for real estate).
- B2B SaaS: G2, Capterra, Product Hunt, LinkedIn company page, GitHub.
- Creator / media: YouTube, Substack, Twitter/X, a tagged
rel="me"link to your personal site. - Ecommerce: Trustpilot, Product Hunt, a Reddit community, Instagram shop.
The signal that does apply everywhere: an author byline with a date. That costs ten minutes and moves every engine.
See your per-platform breakdown
Paste your URL into LLCrawler and the report shows your score for each of the five engines, which signals are dragging the low ones down, and the exact fixes to close the gap. If you have already created an llms.txt and added JSON-LD, you are already ahead of most sites on ChatGPT — the per-platform view tells you where to spend the next hour of work.
Sources
- llmstxt.org — the official llms.txt specification
- schema.org — vocabulary Gemini and AI Overviews lean on
- platform.openai.com/docs/gptbot — OpenAI's published crawler behavior
- robotstxt.org — the base spec every bot still honors