The state of landing pages, measured

How landing pages score · N=334

Score distribution across all 334 completed audits. The median page scores 6.7 out of 10; 96% of audited pages carry at least one failed condition.

Run an audit to see where your page sits on this distribution.

Condition failure base rates

The share of audited pages that fail each condition, as defined in the published specification. If your page fails one of these, this is how common that failure is.

Each condition links to its reference page: definition, decision rule, and pass/fail examples from the published specification.

Failures that travel together

Conditional failure rates: when a page fails the first condition, how often it also fails the second. Useful for knowing what else is probably wrong.

How fixable are the failures

Every finding is classified by severity and implementation effort. The split across all findings in the dataset:

The AI systems, what they read, and how deep they reach

Search crawlers are counted separately: Googlebot has reached 4 of 188 sitemap paths, Bingbot 27. SEO tools and unclassified agents are excluded from every number on this page.

The content AI retrieves most

Content paths only. robots.txt, sitemaps, and API endpoints are excluded.

Does reading turn into citing? Measured answer outcomes

Reads measure retrieval. Citations measure whether an answer engine actually uses Nebula when a buyer asks. We query live answer engines on a fixed set of buyer-intent questions and record verbatim responses.

Latest run: 2026-09-01 · 14 live responses across 7 buyer-intent queries · Raw responses are hashed and archived. Wins and gaps are both published; a gap stays a gap until a measured run says otherwise.

Repair verification: measured, not promised

After a Repair Sprint, Nebula re-observes the same versioned condition on the same page and records whether it moved from FAIL to PASS. That is a condition-state change, not a conversion claim. We publish this rate only from real re-observations.

Diagnostic reliability: how often the instrument can determine

Every determination is stamped with observation integrity. When the engine cannot establish an observation, the finding says INDETERMINATE with a reason code instead of guessing. We publish the determinable rate once the measurement gate is met.

How these numbers are counted

• AI reads count page requests from user agents classified as AI assistants or their crawlers: PerplexityBot, OpenAI (OAI-SearchBot, ChatGPT-User), GPTBot, ClaudeBot, Applebot. Static assets are excluded.
• Search reads (Googlebot, Bingbot, YandexBot) are reported separately. SEO tools (Ahrefs, Semrush), monitoring agents, and unclassified traffic are excluded from both.
• Citation outcomes come from live answer-engine API queries on a fixed buyer-intent query set, run on a schedule. Raw responses are hashed and archived before summarization. A citation is a link to nebulacomponents.com in the engine's cited sources; a mention is the brand name in the answer text.
• The measurement window starts 2026-09-01 and accumulates daily. We publish the exact window rather than implying a longer history.
• Dataset statistics (score distribution, base rates, co-occurrence) aggregate all completed audits with internal traffic excluded. Any cell with fewer than 30 observations is suppressed rather than published. Conditional rates state their own denominator (n).
• Repair verification counts condition-state changes (FAIL to PASS on the same versioned condition ID) from independent re-audits. Reliability counts audits whose observations were all determinable. Both publish only after their stated evidence gates are met; before that, the gate state itself is shown.
• Audit dataset figures come from completed audits only. Counts are facts about the dataset. They are not conversion claims, and reads are not endorsements.