Brandon Lazovic

Pulse · August 22, 2026

Observed

Search Console's Crawl Stats report is missing two days of data, and Google has not fixed it

Search Engine Roundtable reported August 21, 2026 that Google Search Console's Crawl Stats report has a data gap covering August 15 and 16, with the chart's August 15 bar showing data actually dated August 18. Multiple site owners flagged the gap on X, some tagging Google's John Mueller directly, and Glenn Gabe said he would raise it with Google. Google had not corrected the display, and has had similar Crawl Stats gaps before, including in 2021, 2022, and 2025.

In plain termsCrawl Stats is the Search Console report that charts how many times Googlebot actually requested pages from your site each day; when a day's bar goes missing, that is usually the dashboard failing to display data Google already collected, not proof Googlebot skipped your site.

Why it mattersA blank window in Crawl Stats can look identical to a genuine drop in Googlebot activity, and the difference matters before anyone spends time debugging a fetch problem that may not exist.

Our takeI flagged the same pattern in Google Wants a 304, where the Page Indexing report that would confirm a crawl-budget fix worked runs on its own broken schedule, a mismatch I called the Reconciliation Trap. This is that same trap on a different report: the log that would tell you whether Googlebot really slowed down is the one currently showing gaps. Treat the two missing days as a display bug until Google backfills them; if the gap widens instead of closing, that is the point to escalate it as a real data-loss incident.

Confirmed

llm-openrouter 0.7 adds built-in Shell, WebFetch, and WebSearch tools to Simon Willison's LLM plugin

Simon Willison released llm-openrouter 0.7 on August 21, 2026, updating the plugin that connects his LLM command-line tool to models hosted on OpenRouter. The release integrates OpenRouter's Responses API, adds compatibility with LLM 0.32, and ships three server-side tools, Shell, WebFetch, and WebSearch, that a model can invoke directly during a session via flags like -T WebSearch. Full release notes are published on the project's GitHub releases page.

In plain termsLLM is a widely used command-line tool for talking to AI models, and OpenRouter is a service that routes those requests to whichever model a person picks; this update lets the model itself decide mid-conversation to search the web or pull a specific page, instead of a person feeding it search results by hand.

Why it mattersAnother popular agent tool now fetches and searches the live web as a built-in capability rather than a bolted-on integration, adding to the traffic hitting server logs that looks like neither a browser nor a declared crawler.

Our takeEvery agent framework that ships web search and page-fetching as a first-class, model-triggered tool adds another path pulling site traffic that never opens in a browser and may not identify itself the way a declared crawler does. Worth checking server logs for fetch patterns that look like an ad hoc request from a script rather than a named bot, since that gap is where these visits currently go unmeasured.