AI insights

AI insights

MetriXs ships two AI-focused tools. They answer the two questions that matter now that assistants like ChatGPT, Perplexity, and Gemini send real traffic:

1. AI referrals (all plans) — how many visitors arrive via links in AI answers?
2. GEO audit (Team and Pro) — how easily can AI systems find, crawl, and read your pages?

Both live on the AI tab of the dashboard. Neither uses any LLM API: the referral report works from data MetriXs already stores, and the audit is a deterministic scanner that reads your site’s public files. Nothing is sent to an AI provider.

AI referrals

The AI referrals panel counts visits that arrive with the referrer of an AI assistant: someone asked ChatGPT a question, clicked a link in the answer, and landed on your site. ChatGPT, Perplexity, Gemini, Copilot, Claude, and several other platforms are recognized.

• Each row shows the sending platform with visits, bounce rate, and (on commerce-enabled sites) the revenue these visitors generated.
• Referred visits are classified as the ai traffic medium, so they also appear in the Medium tab of the Sources panel and can be used as a filter.
• The panel respects the period picker like every other card.

Two honest limitations:

• AI crawlers (GPTBot and friends) never appear in any JavaScript-based analytics pipeline — they do not execute scripts. The panel counts human clicks out of AI answers, not crawler hits.
• Mobile AI apps sometimes send no referrer at all, so traffic from those apps is undercounted.

GEO audit

The GEO audit scores how machine-readable your site is. It checks the homepage and your most-visited pages of the last 30 days (picked from your own analytics, so the audit reflects the pages that actually matter).

Three categories:

Category	Checks
Crawler access	Does robots.txt block known AI crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended, CCBot, and more)? Is an llms.txt present? Is a sitemap findable?
Content	Is there real text visible without JavaScript (AI crawlers do not run scripts)? Title length, meta description, a single H1.
Structure	JSON-LD structured data, canonical URL, hreflang language alternatives.

Each check passes, warns, or fails; every page gets a score, and the overall score combines crawler access with the average page score. A dedicated matrix shows per AI crawler whether robots.txt allows it.

• The audit re-scans at most once per hour; running it again sooner returns the stored result.
• The last 12 audits are kept, so you can track your score over time.
• Availability: Team and Pro plans (trials included). The card shows an upgrade prompt on other plans.

What the audit does not do

• It does not ask an LLM whether your brand is mentioned in answers. Measuring actual AI answers requires sending prompts through an AI provider, which conflicts with the MetriXs privacy model.
• It does not automatically fix issues. The check details tell you what to change (for example: unblock GPTBot in robots.txt, add an llms.txt, or add structured data), and each check links its finding to a concrete detail.