What is Google Search Console, and how to read it without fooling yourself

How to use Google Search Console: what it does, and what it doesn't

Search Console tells you what Google crawled, what it indexed, what it refused to index and what people typed before they clicked. It doesn't tell you what visitors did once they arrived: that's GA4, and the two disagree on purpose.

It also doesn't rank your site, and nothing you press in it makes Google index a page. The URL Inspection tool lets you request indexing, and Google is explicit that no test can guarantee a page will be indexed.

Here's how it works: Google records what it does with your site as it crawls, indexes and shows it, and Search Console turns that into reports. You prove the site is yours, and data starts showing up within a few days.

Google's own documentation calls it a free service that helps you monitor, maintain and troubleshoot your site's presence in Google Search results.

Do you build websites? Here's how to set up Google Search Console

Setup is six steps, and only the third can take a while. All you need is a Google Account and access to the panel where the domain was bought, because verification uses a DNS record. If the site belongs to a client, the property should be in the client's name, with you added as a user.

• Sign in to Search Console with your Google account and click “Add property”.

• Choose “Domain” and type only your domain, without https or www (for example, your-site.com). Why Domain and not URL prefix is in the next section.

• Copy the TXT record Google gives you and add it to your domain's DNS: in your registrar's panel, or in Cloudflare if you use it. Google warns that your provider can take up to two or three days to start serving it.

• Go back to Search Console and click “Verify”. If it fails, wait a day and try again before changing anything.

• Submit your sitemap in the Sitemaps report. How to read its answer is further down.

• If someone else needs to see it, go to Settings > Users and permissions, click “Add user” and pick Full or Restricted. Only an owner can do it.

Google starts collecting a property's data as soon as it's added, even before it's verified, but it takes a few days to show up.

Having more than one person get the alerts isn't a formality. In 2018, Search Engine Land, the SEO industry's reference publication, vanished from Google for more than 12 hours over a false positive on Google's side. They found out from their readers on social media before Search Console's alert reached them.

Which property should you pick? Domain, not URL prefix

When you add a site, Google offers two property types. A Domain property covers every subdomain and both protocols at once. A URL-prefix property covers exactly one, so https and http, www and bare, are four different properties.

On a domain migration in September 2026 I verified the Domain property through DNS, and the home page showed as indexed the same day. The API said “Crawled - currently not indexed” and the interface said “URL is on Google”. That's two instruments and two readings, and I don't pick one until I check again.

If you already have both, always say which one each number comes from, and don't mix their series: they can start on different dates. On a family member's site, the URL-prefix property's “last 3 months” start on August 22, 2026. So its total doesn't compare with the other view's.

On the 27 days they share, clicks and impressions match day by day (I checked it when saving the Oct 2, 2026 export).

Which report should you open first? Page Indexing

Before Performance, before anything: the Page Indexing report tells you how many of your pages Google decided to keep. A page that isn't indexed can't rank, and no amount of content work changes that.

Two statuses are worth learning by heart. “Discovered - currently not indexed” means Google knows the URL exists and hasn't fetched it. “Crawled - currently not indexed” means it fetched it, read it and chose not to keep it.

The first is usually a crawl-budget or a linking problem. The second is a content problem, and it's the harder one. The full list is in the seven causes of a page not appearing in Google.

Not everything listed as “not indexed” is a problem. This is how I read the statuses people ask me about most, based on Google's help on the report:

On a site I reviewed, the “three pages with redirect” that worried everyone were the home page with “www”, with “http” and with both. That was expected.

If someone alarmed you with a list of “not indexed” pages to sell you a fix, ask them for the exact status of each one. Several of those rows can be expected, like the home page redirects I described above. If you want a second reading before you pay, the free diagnostic gives you one on what's public, in writing, with no sales call.

What does URL Inspection tell you? Two versions you shouldn't mix

Mix up the two versions it shows and you'll either think you're indexed when you aren't, or panic over nothing. “View crawled page” is what Google stored the last time it came by, with its date. “Test live URL” followed by “View tested page” is a visit made today, and it says nothing about what's stored.

In a review I was running on a site built with Base44, the AI agent mixed up the two. It concluded Google couldn't see the site.

Once the crawled page, the tested page and the page source were compared, each with its date, the pattern was different. Pages crawled weeks earlier were missing the content, and recent ones had it. Since then I don't call anything a blocker without those three readings.

Three limits Google states in its help:

• The live test follows redirects without indicating that it did, or showing the final URL. It can't confirm a redirect.

• It doesn't check manual actions or security issues: a clean result doesn't rule them out.

• “Request indexing” has a daily limit and guarantees nothing, and asking several times won't get it crawled faster. For many pages, use the sitemap.

Also check two lines: “User-declared canonical” and “Google-selected canonical”. If they differ, Google picked another version of your page.

Should you submit a sitemap? Yes, but don't trust the first status

Submitting a sitemap tells Google which URLs you consider canonical. It's a hint, not an instruction: Google indexes URLs that aren't in it and ignores URLs that are.

The day this site went live (Sep 15, 2026), the report said it couldn't read the sitemap, though the server answered it with a 200. Four days later, on Sep 19, 2026, the server was still answering with a 200.

Don't believe the first status: open the sitemap URL in a browser, and if it renders, read the report again before changing anything.

How to do it with a Worker in front of the site is in the guide to SEO for single page applications.

This is the content of my sitemap: [paste the text of your-site.com/sitemap.xml].
This is what Search Console says in the Page Indexing report: [indexed pages, not indexed pages and each reason].
List and number the addresses in the sitemap’s <loc> tags, give the total and compare it with the indexed ones. If I didn't give you the list of non-indexed pages, don't guess which they are.

Rules for this review:
1. Quote the exact text you found. Don't paraphrase it.
2. If something can't be seen from the browser, write “not verifiable from the browser.” Don't assume.
3. Give me a table: element · what you found · fine / review / missing · why it matters in one sentence.
4. Don't make up traffic, visits or rankings. Don't promise anything will rank higher on Google.
5. If you can't open an address yourself, say so in the first line and work only with what I pasted. Don't describe a page you didn't see.

Want to check it yourself?

Three things you can verify in five minutes, without waiting for any report to refresh:

• Open Page Indexing and write down the number of indexed pages. Compare it with the number of URLs in your sitemap. A large gap is your first real finding.

• Pick your most important page, run it through URL Inspection and read “Crawled as”. If it says a smartphone user agent and the rendered HTML has no body text, your problem is rendering, not content.

• Filter Performance by that page and sort by impressions, never by position.

Do you create content on Instagram, YouTube or TikTok? What Search Console Insights shows you

Search Console doesn't index Instagram, TikTok or YouTube profiles, and you can't add them to it. What does exist is a Search Console Insights experiment: it shows how the channels Google automatically links to your website perform in Google Search. For now, it reaches a limited set of sites.

Google announced it on December 8, 2025. It shows clicks and impressions from Search, top content, the queries that lead to those channels and your audience's countries. The help page names YouTube, Instagram and TikTok as examples, and says that for now you can't choose or add channels.

If you don't see it, there's nothing to switch on. What is in your hands is measuring which of your content brings people to your website:

• In GA4, with UTM links in your bio and in each campaign. They tell you which network and which post sent you visits and leads. How to build them is in do I need a website if I already sell on Instagram?

• In Search Console, with searches for your name. If your content works, you'd expect more people to look you up on Google. Compare the same number of days before and after a series of posts.

• In Search Console, with your website's queries. They're your audience's exact questions, and the ones already earning impressions are topics people already search for, for your next video or post.

What is Search Console Insights?

It's Search Console's summary report, built for people who aren't data experts. Since June 30, 2025, it lives inside Search Console. It shows clicks and impressions against the previous period, and which pages and queries are rising or falling.

Running a business and spending money on your site? Take a baseline snapshot first

Before you pay for ads, redesign or commission content, save how your site stands in Search Console today. Without that snapshot, three months from now nobody can tell you whether the money moved anything, not even me. It takes a moment and costs nothing.

The snapshot is three numbers from the last 28 days: impressions, clicks and CTR, with today's date and the view they come from. If your site is new, compare counts, not percentages: a percentage over zero means nothing.

A real example: on September 28, 2026, I saved the baseline snapshot of a family member's site, in Search Console and in Google Analytics. I keep it in the project log, with its date.

That's how I work: I set the date of the next reading before the data exists. In the measurable SEO and paid search plan, for example, the reading is in late November 2026.

Today is [date]. These are my site's numbers in Search Console, Performance, Search results, last 28 days:
- Impressions: [ ] · Clicks: [ ] · CTR: [ ]
- Without searches that include my business name: [the same three, or “I didn't filter it”]
Build me a one-row table with the date, the tool, the period, the view (“chart totals”), the three numbers and the three without my business name, ready to paste into a document.

Rules: use only my numbers. Don't calculate percentages or projections. If a number is missing, write “pending” in its cell instead of estimating it. Don't fill anything in with data that isn't in this message.

Why don't the Performance rows ever add up?

The Performance report shows four numbers. Impressions are how often Google showed your page, and clicks how often someone came in. CTR is the share of viewers who came in, and then there's average position.

The first thing to know about it is that the totals don't equal the sum of the rows, and that's by design, not a bug.

Two reasons. Google anonymizes queries that very few people type, so they exist in the total and in no row. And the chart counts by property: two of your pages in one search are one impression in the total and one in each page row.

Google's Daniel Waisberg defines them this way: “Anonymized queries are those that aren't issued by more than a few dozen users over a two-to-three month period.” On a small site they can be most of the clicks, as in the example below.

I checked it on a URL-prefix property on Sep 28, 2026, with a Search Console connector, for Aug 29 to Sep 28. The total by date said 14 clicks, the sum by page came to 19 and the visible queries added up to 3. Whoever reports “the sum of the table” is reporting a fraction without knowing it.

Average position is the other trap. Google defines it as an average of the topmost position your site reached on each impression. So a page seen once at position 3 and never again beats a page seen four hundred times at position 11.

Read it next to impressions or don't read it at all.

Before explaining a rise or a drop, check the data anomalies list Google publishes. For example, Google says it didn't count impressions correctly from May 13, 2025 to April 27, 2026. A year-over-year comparison that crosses those dates compares two different ways of counting.

One question these reports can't answer is what Google would get from your page if it looked today. The live test in URL Inspection answers it for one page at a time. Without a terminal, View page source on the published page is the other way in: it shows what the server sends your browser.

curl -s -A "Googlebot" https://your-site.com/page | wc -w
→ how many words the crawler actually receives, right now
The reports tell you what Google did the last time it crawled.
This tells you what it would get if it looked today.

What do you decide from the Performance report? Three readings that end in a task

A number only helps if it ends in a task. Open Performance for the last three months and sort the Pages tab by impressions, highest first. Then open the queries for each page and compare them with what the page says.

• Lots of impressions and hardly any clicks: Google shows it and nobody clicks it. The title and description don't promise what that person was looking for; it's the first thing to rewrite, and it doesn't touch the content.

• A query with impressions that your page never mentions: Google already connects it to you, but the page doesn't answer it. Answer it in a subheading on that same page, using those same words, not on a new page.

• The same query split across two of your pages: they're competing with each other. Pick the one that answers it and link to it from the other.

Neil Patel puts it plainly in his guide to keyword cannibalization: “Because Google doesn't know which one to prioritize, they both lose ground.” The way to prevent it from the start is a keyword map, which I cover in how to build a website with AI.

Before deciding, take your brand out. In the query filter, choose “Queries not containing” and type the part of your business name that all its variants share. Searches from people who already know you aren't SEO's doing.

If your property shows the branded queries filter Google introduced in November 2025, it does the same job.

A search that doesn't show up isn't proof that nobody makes it: the query view hides rare searches. What does show up is the floor of demand, not its ceiling.

Selling with data: queries are your clients' own words

Your website's queries are what your clients type when they have the problem, unfiltered. Use them as they are in your offer's headline, in your ads and in the replies you send on WhatsApp. If a query with impressions asks for something you don't offer, write it down: it may be your next service, or a new page.

Rewriting the title and description of a page nobody clicks

When the task is a title, this prompt gives you options tied to that page's real queries, not to what sounds good:

Page: [URL]. Current title: [ ]. Current description: [ or “none”]. H1: [ ].
Search Console queries for this page, last three months: [query · impressions · clicks; filter the report by this page before copying them].
Suggest three titles and three descriptions. The title starts with the main search and says what the page really offers; the description says what the reader will find and why here.
As a guide, not a rule: titles of 50 to 60 characters and descriptions of 120 to 155.

Rules: only use what the page says: no “best”, no “guaranteed”, no “top spots”, and no numbers I didn't give you. In the “query it answers” column, put the query in quotes; don't put it inside the title or description text. Write the way the page speaks: if it's one person's, use “I”, not “we”. Give me a table: suggestion · characters · query it answers · what changes compared with the current one. If the queries aren't enough to decide, say so. If you can't open the page, say so in the first line and work only with what I pasted.

Google sets no character limit and may rewrite the title and the description. Change one page at a time, write down the date and check the report again after 30 days, comparing the same number of days.

Reading your Search Console with you, page by page, is part of the technical SEO audit.

How do you analyze Search Console with AI?

The simple way is to export the Performance report and paste it into Claude, ChatGPT or Gemini, with a prompt that forbids guessing. The direct way is to connect Claude to your Search Console account with read-only access. Then it queries the data itself and tells you which content works so you can repeat it.

I use the direct way. The tools I work with, which I specified and generated with Claude Code, read Search Console through Google's official API and my own account. They have read-only access: they can query data, but they can't change anything in the property.

Without connecting anything: from the export to the tasks

In Performance, press Export and download the file. Open it, copy the Queries, Pages and Chart tabs (Chart is the one with the dates), and paste them into your assistant with this text:

Here is the Search Console Performance export for my site.
- Period: [dates]
- My brand is written as: [name and variants]
- Queries tab: [paste the rows]
- Pages tab: [paste the rows]
- Chart tab (the one with the dates): [paste the rows]
- Queries for a specific page, if you want to review it (filter the report by that page and copy the Queries tab): [paste the rows, or write “I didn't”]

Do this, in this order:
1. Separate the brand queries and work only with the rest.
2. For each page, list its queries with the most impressions and tell me whether the page mentions them.
3. Flag pages with lots of impressions and hardly any clicks: candidates for a rewritten title and description.
4. Flag queries that show up on two of my pages.
Give me a table: page · what you saw in my data · proposed task · how I check in a month whether it worked.

Rules: use only my numbers and say which tab each one comes from. For the total use the Chart tab, never the sum of the rows: rows leave out anonymized searches. A query that doesn't appear doesn't mean nobody searches for it. If a page has few impressions, say so and draw no conclusions from it. Don't invent search volumes or promise rankings. If you can't open one of my pages, write “I couldn't open [address]” in its row and work only with what I pasted.

With Claude connected: three ways, from less to more control

• A connector hosted by a third party. In Claude, under Settings > Connectors, you add the address the provider gives you and sign in with Google. It's the fastest, but your data goes through that provider.

• An open-source MCP server on your computer. You add it to Claude Code with the claude mcp add command, which Anthropic's documentation explains.

• Your own server on the official API, which is how I run it. It takes more work up front, and in exchange you decide which permission it asks for and where your credentials live.

The steps for the last two, roughly:

• Create a project in Google Cloud and enable the Search Console API.

• Create a desktop-type OAuth client and keep its file outside any folder that gets shared or pushed to GitHub.

• Install the server and add it to Claude Code.

• The first time, sign in with your Google Account and read which permissions the screen asks for before you accept.

• Ask Claude to list your properties. If yours isn't there, stop.

The risks, before you connect

• Credentials. The OAuth client file and the token give access to your data. Don't share them, don't push them to GitHub and don't paste them into a chat.

• Permissions. Ask for read-only whenever you can. If the server ships with tools to add or delete sites and sitemaps, leave them switched off.

• Third-party servers. Anthropic warns that you should trust each server before connecting it, because servers that fetch outside content can slip instructions to Claude.

• The wrong property. If Claude reads another site, it will answer with the same confidence: have it always tell you which property and which period it read.

• The same limits as the report. The API also leaves out anonymized queries, so the total comes from the by-date numbers, not from the sum of the rows.

Once it's connected, this prompt tells you which content works and what to repeat:

You have read-only access to my Search Console. Before anything else, tell me which property you're going to read and the exact period. If it isn't [your-domain.com], stop and tell me.

Do this, in this order:
1. Total clicks and impressions for the last 90 days against the previous 90, taken from the by-date numbers.
2. Separate the queries that include my brand: [name and variants]. Work only with the rest.
3. Group the queries by topic and tell me which pages each topic appears on.
4. Flag the topics growing against the previous period, and the pages with lots of impressions and hardly any clicks.
Give me a table: topic · pages · what my data says · what to repeat or go deeper on · how I check it in 30 days.

Rules: use only the data you read from Search Console and say which query each number comes from. If a topic has fewer than [100] impressions, say so and draw no conclusions. A query that doesn't appear doesn't mean nobody searches for it. Don't invent search volumes or promise rankings. Don't change anything in the account: read only.

Watch it explained, by the people who built it

Three short videos from Google's own channels. They're the same source the links above point to, in video form:

• Intro to Google Search Console · Google Search Central, 3 minutes. Episode 1 of the official training series.

• 7 ways to verify site ownership · Google Search Central, 14 minutes. Every verification method, including the DNS one that a Domain property needs.

• Performance reports in Search Console · Google Search Central, 8 minutes. The report whose rows don't add up, explained by the person who owns it.

Frequently asked questions