Search & Gaps

Search & gaps (/know/search, under Content in the Know sub-nav) shows what readers and their agents looked for, and what to write next. It is on every plan.

It takes the date window with Compare with prior period, and the page type, readers, agents and account filters. A question that led to no read has no page type, so when you pick a page type those questions stay listed, marked No page type.

The Search & gaps view: Site searches, Agent questions and Searched, then nothing tiles above the What to write next list, each item with its evidence chips and Evidence, Draft a page and Dismiss buttons

The headline figures

FigureWhat it counts
Site searchesSearches typed into your site’s search box, and how many readers searched
Agent questionsQuestions agents asked your Docs MCP and agent surface, with a link to the Agents query log
Searched, then nothingThe share of questions that led to no read, with its change in points against the prior period

Site search rates can only be worked out for searches recorded with a session. When the window reaches back before that, the figures say Counted from and the date of the first such search.

What to write next

The main list ranks the items most worth writing for. It has three tabs, Open, In progress and Dismissed, each with its count. Ten items show at a time, with Show more below.

There are three kinds of item, and each carries its evidence as chips:

KindEvidence
An unanswered questionN questions · N% no read, Reworded N times when readers rephrased it, N people · N accounts, and the nearest page with its status, or No page covers this
Traffic rising, reading fallingViews before → after, Reading depth before → after, and Last revised N days ago
Shipped, unreadShipped in the release on its date · 0 agent reads, and Nobody was told when no email or share named the page

People and accounts are counted only inside your plan’s contact limit, and nobody is named. The Search & gaps badge in the sub-nav is the Open count; it is left out at zero.

Evidence and Release

Each item has a button that opens what it is based on:

  • an unanswered question opens Agents › Query log on that question;
  • a rising page opens its page detail;
  • a shipped page (Release) opens its release’s scorecard on the Unread tab, under agent readers.

Draft a page

Draft a page, on an open question or rising page, asks for a page type, creates a draft titled from the item, links it to the item (which moves to In progress) and opens it in the editor.

If your role carries ai:write, the button reads Create and draft. The item’s evidence, read from the server as get_gap reads it (every spelling of the question, the site search terms, how readers reworded it, the pages they read after asking, the capped reader counts and the nearest pages), and the nearest page’s body go to the editor’s AI Writing Assistant. It writes a first version as tracked suggestions while a Drafting banner shows, with Retry if it fails. The suggestions record the Leed assistant as their source. A person accepts or rejects each one, and publishes the page. Without ai:write, the button reads Create draft and the page opens empty.

Dismiss and Reopen

Dismiss opens a small Why? (optional) box for a reason of up to 280 characters, then hides the item until its count doubles. Leaving the box empty still dismisses it in one click. The reason shows under the item on the Dismissed tab.

Reopen, on the In progress and Dismissed tabs, undoes a dismissal or a link. A linked page is left as it is. Moving an item to In progress, or reopening it, clears its reason.

An item closes on its own once its linked page is published.

Site search terms and Rising questions

Site search terms lists what visitors typed into your site’s search box: searches, how often a page was viewed next, how often the search was followed by a conversion, and the page viewed next. Rising questions lists the questions growing fastest against the prior period.

Drafting from an agent

The same list is open to an agent over the Operator MCP and to the leed know CLI, and the decisions it records are the same ones the screen makes. The leed:analytics plugin skill runs this flow:

flowchart TD
  A["get_know_view searchGaps<br/>pick an open item"] --> B["get_gap<br/>the item and its evidence"]
  B --> C{"New page or<br/>revision?"}
  C -->|"New page"| D["create_page_from_markdown<br/>with gap"]
  C -->|"Revision"| E["Put the page IN_REVISION<br/>update_page_body in suggest mode<br/>with sourceId"]
  E --> F["mark_gap_in_progress"]
  D --> G["A person reviews<br/>and accepts or rejects"]
  F --> G
  G --> H["Published<br/>the item closes"]

Docs pages publish with the release that carries their source id. An agent never publishes. To set an item aside instead, dismiss_gap takes the item’s count and an optional reason, and reopen_gap undoes it.

From a terminal, the same steps are leed know gap show, leed know gap link, leed know gap dismiss --reason and leed know gap reopen, and leed page create --gap creates and links a new page in one command. See leed know commands.

ESC