Search Console Data Inside Your Assistant, Without Extra Servers

Google Search Console answers a question your analytics cannot: what did people search for before they saw your site? Analytics answers the one Search Console cannot: what did they do afterward? Joining the two is where most of the SEO value is, and also where most of the tedium is.

This post covers how OnyxMetric brings Search Console data into the same place as your traffic, how an AI assistant can query it, and what to ask.

Two data sets, one gap

Search Console reports on the part of the journey that happens on Google: impressions (how often your page appeared), clicks, average position, and the queries that triggered them. It stops at the click.

Your analytics starts at the visit. It knows the landing page, the session, and what happened next, but "organic search" is mostly all it can say about the query, since search engines stopped passing keywords along years ago.

Because of this gap, people end up with two tabs open and a spreadsheet to reconcile them. The page that gets the most clicks may not be the one that produces signups. The query with the most impressions may land on a page that doesn't match what the searcher wanted.

Connecting Search Console in OnyxMetric

On the Cloud product you connect a Google Search Console account to a site in the site settings, under Integrations. Once connected, search performance sits next to your regular reports. You can group the data by query, page, country or device.

A few facts worth knowing:

• The connection uses a Google account, so the connecting user needs access to the Search Console property.
• If a site is transferred to another organization, the Search Console connection is removed, because it relies on the sender's Google account. The new owner connects their own.
• There's no Google Cloud project for you to create. The integration is part of the product.

Asking an assistant instead of clicking

OnyxMetric's hosted MCP server includes two tools for this: one reports whether Search Console is connected for a site and which property it uses, and another returns the data. The data tool can group results by up to three dimensions at once, using query, page, country and device. You do not need a separate Search Console MCP server, and nothing extra to install.

Set up your assistant with the endpoint https://app.onyxmetric.com/api/mcp, authorize through OAuth or provide an API key, and the Search Console tools appear alongside the traffic, funnel and event tools. If you ask for search data on a site where the integration isn't connected, the call fails with a "not connected" error. The fix is to connect it in the site settings and retry.

Because the assistant can use both sets of tools in one conversation, the join I described above becomes a single question.

Questions worth asking

Finding quick wins. "Show queries where my pages rank between positions 8 and 20 and have more than 100 impressions." Pages just off the first page are the cheapest to improve: better title, clearer answer near the top, a few internal links.

Spotting low click-through. "Which pages have high impressions but a click-through rate below two percent?" Usually a title or description that doesn't match the intent of the query. Rewriting it costs little.

Checking search intent against behavior. "For my top five organic landing pages, show clicks from Search Console and the goal completion rate from analytics." A page with many clicks and no conversions is attracting the wrong audience, or leading them nowhere.

Device gaps. "Compare average position and click-through on mobile versus desktop for the pricing page." A mobile position well below desktop can point to a mobile usability or speed problem. Check Web Vitals for that page by device next.

Country gaps. "Which countries produce impressions but almost no clicks?" That can mean you rank in a market where your snippet or language doesn't fit.

Cannibalization. "Do several of my pages appear for the same query?" Two pages competing for one term split attention. Merge or differentiate them.

Read the numbers with care

Search Console data has quirks, and an assistant won't always warn you.

• Average position is an average. A page can rank first for one query and twentieth for another. Look at position per query, not per site.
• Anonymized queries. Google hides rare queries for privacy, so totals by query don't always add up to totals by page.
• Delay. Recent days are incomplete. Don't compare the last two days to anything.
• Clicks differ from sessions. A click in Search Console and a session in analytics will not match exactly. Blockers, redirects and bounces before the script loads all create differences. Compare trends, not totals.
• Small numbers swing. Ten impressions at position 3 means very little. Set a floor before you trust a ratio.

When asking an assistant, request the date range, the dimension used and the row count along with the answer. That makes mistakes easy to see.

A monthly routine

Fifteen minutes is enough:

1. Ask for the top 20 queries and pages by clicks, compared with the previous month. Note the winners and losers.
2. Ask for pages with high impressions and low click-through. Pick two to rewrite.
3. Ask for pages in positions 8 to 20 with decent impressions. Pick two to improve.
4. For your top landing pages, check goal completions and Web Vitals. A page that ranks well but converts poorly or loads slowly is worth fixing first.
5. Write down the changes and check the same view next month.

Limits

The Search Console tools are read-only. They help you find and prioritize; the writing and the fixing remain yours. They also only see what Google reports about your property, so they say nothing about other search engines.

Still, having the query data in the same conversation as your behavior data removes the tab-juggling. That alone changes how often people actually do the analysis.