AI bot crawls (top-200 sample): 735,076 hits — 18.1% are non-editorial (WordPress JSON + assets); editorial-only: 602,089. Domain-wide total for /blog: ~1.9M editorial AI-bot hits.
AI assistant visits: 5,005 sessions across 6 named sources. ChatGPT = 71% (3,529 visits).
Google Search clicks: 17,396 clicks across the top-500 /blog pages.
Overlap (all 3 channels, editorial only):87 qualifying URLs.
Heaviest editorial AI crawler: ChatGPT (ChatGPT-User: 1.09M hits).
ChatGPT dominates end-to-end — 70% of LLM visits, 1.26M raw bot hits, ~106:1 editorial crawl→visit ratio.
Gemini crawls aggressively, converts poorly. Top-crawled URLs are .js/.css. Most volume is asset scraping, not content ingestion.
Perplexity is the most efficient converter at ~17:1 — meaningful editorial hits, real article URLs, decent visit yield.
Meta-ExternalAgent (170K hits) and Applebot (103K) crawl heavily but drive zero attributable LLM visits — pure training scrapers, not live assistants.
Copilot has zero classified AI bot (Bing uses bingbot, which isn't classed as AI) — visit data lands but crawl side is invisible. Don't try to compute a ratio.
AI-meta content owns the AI Priority top.ai-overviews-reduce-clicks-update, ai-overview-citations-top-10, what-is-llms-txt, geo-generative-engine-optimization all rank in the top 15 by AI Priority — content about AI search is what AI assistants cite most.
GSC top is dominated by evergreens.google-advanced-search-operators, google-keyword-planner, keyword-research all have strong Google ranking + moderate AI engagement. These are refresh candidates where Google ranks the page but AI synthesises the answer without citing.
Strong Bot, weak Web = AI sees but doesn't pick. Diagnose answer quality and direct-answer structure (TL;DR up top, clear definitions).
Low Bot, high Web = cited from training. Freshness updates compound here.
Balanced rows = cleanest signal of active AI engagement; prioritise updates here first.
AI-meta + commercial pages are compounding (what-is-llms-txt, ahrefs-pricing, google-keyword-planner, google-web-guide) — all three channels up.
Early-2026 AI-topic pages are losing freshness:mcp-use-cases, how-to-track-ai-overviews, ai-visibility all 2–3 channels down. These were timely a quarter ago and the conversation has moved on — refresh candidates.
New entrants from zero:academy, most-searched-people, on-page-seo-checklist, how-to-increase-website-authority — fundamentals being re-cited.
Bot caveat: the top-200 bot sample covers ~735K of 1.9M editorial /blog hits. Bot Δ% on lower-ranked pages can read "down" simply because the URL dropped out of the top-200 in one window. Trust Web + GSC more on borderline rows; trust 3↑/3↓ rows fully.
Priority 1 — refresh balanced top-20: pages in the Top 25 overlap with non-trivial values in all three channels. Improvements compound — these are already being seen, cited, and ranked.
Priority 2 — fix the AI-sees-doesn't-pick gap: pages with high Bot, low Web (e.g. top-google-searches, hreflang-tags). Add direct-answer summaries, statistics boxes, and explicit definitions near the top.
Priority 3 — GSC-strong, AI-weak refresh: top GSC pages with low Web Visits (google-advanced-search-operators ranks #2 GSC but only 65 LLM visits). Modernise for 2026 AI citation patterns.
Priority 4 — refresh the decline cluster (mcp-use-cases, how-to-track-ai-overviews, ai-visibility) — these still have audience signal but are degrading on all sides.
Don't waste cycles on Meta/Apple traffic. They crawl 270K+ pages a month but drive zero visits. Make sure they're not crashing infrastructure — that's the only meaningful action.