SEO Skill

seo data

Raw rows for questions the report does not answer. Search Console by default, or landing pages from your analytics.

Use it when you need a cut of the data the report does not show, such as clicks by country for one folder, or which landing pages got visits from search.

Search Console rows

Ask your agent for the rows you want:

Show Search Console clicks by country for pages containing /blog/ on example.com

You can group by up to four of query, page, date, country, device and searchAppearance, and filter by page or query text. Each row comes with the previous period and the change.

In a terminal:

seo data --site example.com --dimensions page,country --page /blog/
seo data --site example.com --dimensions query --query pricing

Image, video, news and Discover

Search Console counts web search apart from image search, video search, the News tab, Discover and Google News, and the report and these rows show web search unless you ask. Ask for another surface by name:

Which pages on example.com get image search impressions but no clicks?
seo data --site example.com --search-type image --dimensions page
seo data --site example.com --search-type discover

--search-type (search_type in MCP) takes web, image, video, news, discover or google_news. Image, video and news are the Images, Videos and News tabs in Google Search. google_news is the Google News app and news.google.com, not the News tab.

Discover and Google News show pages without a search, so they have no queries and no average position. Rows for them default to pages, and seo data refuses a query cut with an error rather than return an empty list that looks like a zero.

Rows from each of these surfaces come with a line for your agent on what the numbers mean, and a reminder to ask whether that traffic matters to you before it suggests work.

Filter with regular expressions

You don’t need a filter to find questions, comparisons or AI-style searches: the report and seo page label them for you. For any other cut, write a regular expression in the RE2 syntax Search Console uses.

CLIMCP (seo_data)What it does
--query-regexquery_regexKeeps queries that match
--exclude-query-regexexclude_query_regexLeaves out queries that match, such as your brand
--page-regexpage_regexKeeps pages whose URL matches

A row has to match every filter you pass, including plain --query and --page text.

# Pricing questions without your brand
seo data --site example.com --query-regex 'price|pricing|cost' --exclude-query-regex 'example|exmaple'

# Queries for pages in two folders, by country
seo data --site example.com --dimensions query,country --page-regex '/(blog|compare)/'

Through MCP, ask in plain words, such as “pricing queries for pages under /blog/ on example.com, without our brand”, and your agent fills in the options.

Matching ignores case. RE2 has no lookarounds such as (?!...) and no backreferences such as \1. Search Console answers those with no rows, which would look like a real zero, so seo data rejects them with an error; use the exclude option to leave queries out. Each expression can be up to 4,096 characters.

Sort the rows

Rows come with the most clicks first. To rank them another way, ask your agent, such as “queries on example.com with the most impressions” or “pages with the lowest click-through rate and at least 100 impressions”.

seo data --site example.com --sort impressions
seo data --site example.com --sort ctr_low --min-impressions 100
seo data --site example.com --dimensions page --sort impressions_loss

--sort (sort in MCP) takes clicks, impressions, ctr (highest first), ctr_low (lowest first), position (best first), clicks_gain, clicks_loss, impressions_gain or impressions_loss. --min-impressions (min_impressions) leaves out rows seen too rarely to judge. Each row also shows its impressions in the previous period and the change.

Search Console hands rows over with the most clicks first. On a very large site, the rows with the fewest clicks may not make the list, and seo data says so; add a page or query filter to see them.

Add --start-date and --end-date for other dates, and --detail rows with --offset to page through long results. The SEO keyword report page shows how to read the query and page rows as a keyword report.

Queries that look like spam or tool noise get a spam_hint label: host names on top-level domains common in referral spam (such as 07.seo-anomaly-traffic.xyz), rank tracker test queries, and spreadsheet rows pasted into Google. It’s a guess from the query text alone. The rows stay in, with their numbers, so totals still match Search Console.

Analytics landing pages

In the hosted MCP, seo_data also reads Google Analytics landing pages for visits from search, with your Google sign-in, Clicky landing pages for a Clicky site saved on your account, and Simple Analytics landing pages with the API key saved there. Each tool keeps its own metric names, such as sessions, visits or visitors, so compare them with Search Console clicks with care. The analytics guide covers each one.

In the CLI, seo data reads Search Console rows, Bing rows with --source bing, or Clicky and Simple Analytics landing pages with --source clicky or --source simple_analytics. For Google Analytics, use seo analytics.

Bing rows

With Bing connected in the CLI or the local MCP server, ask for Bing as the source to get Bing’s query or page rows with the previous period:

seo data --site example.com --source bing --dimensions page

Bing lists only its top rows, by week, so a row missing from Bing is unknown, not zero. The hosted MCP has no Bing for now, so it tells your agent to run the CLI for these rows.

For the usual questions, start with the SEO report. It already sorts the rows that matter most.