Internal Linking Opportunities: Find Them in Search Console

What are internal linking opportunities?

Internal linking opportunities are places where a link from one of your pages to another would help readers, and help Google find and understand the target page, but no link exists yet. The data-led way to find them is with Google Search Console: pick target pages that already get impressions at average positions of roughly 5–20, pick source pages that earn clicks for related queries, and write anchor text from the queries the target actually ranks for. Then add the links, confirm Google has recrawled the source pages, and compare the target’s clicks after four to eight weeks.

Most internal linking tools find opportunities by text matching, which produces hundreds of possible links and no sense of which ones matter. Search Console adds demand: which pages Google is already testing near the top, which pages people land on, and the words they typed.

We make GSC MCP Server, a read-only Google Search Console MCP server that lets Claude, ChatGPT, Cursor and other AI assistants query your Search Console and Bing data, so the prompts below use its tool names. Every step also works by hand in the Search Console interface and a spreadsheet; it just takes longer.

What Google says about internal links and anchor text

Google’s link best practices page (last updated 10 December 2025) is short and specific. These are the points that shape an internal linking strategy:

• Links must be crawlable. Google can generally only follow <a> elements with an href attribute. Links inserted with JavaScript are fine if they end up as the same markup; <span href>, <a routerLink> and links that only work through an onclick handler are not recommended.
• Every page you care about needs at least one link from another page on your site, placed in context next to related content.
• Anchor text should be descriptive and concise, and relevant to both the page it sits on and the page it points to. Google names “click here” and “read more” as anchors to avoid, discourages overly long ones, and suggests reading the anchor on its own to check it still makes sense.
• No keyword stuffing. It breaks the spam policies. The words around a link count too, so don’t chain links one after another.
• Image links use the alt text as the anchor. There is no ideal number of links per page, though a page can have too many.

Google’s SEO Starter Guide adds that Google finds most new pages through links from pages it has already crawled, and that site structure mostly matters on sites with more than a few thousand URLs. Internal links are a discovery and context signal, not a dial you can keep turning.

What the Search Console Links report can and can’t tell you

Search Console has a Links report (Links in the left-hand menu). Only one of its tables is about internal links, but two others are useful for this workflow.

Mind Google’s caveats: tables stop at 1,000 rows, the data is a sample, pages are grouped by canonical URL, removed links can linger, and unindexed pages may be missing. A page absent from the internal table isn’t proof that nothing links to it.

There is no API for the Links report

The Search Console API has four parts: Search Analytics, Sitemaps, Sites and URL Inspection. None returns links, so no Search Console MCP server, ours included, can read Google’s link data. On the Bing side, get_bing_crawl_stats returns a daily inbound-link count for the whole site: a trend line, not a list of pages.

How to use the export instead

1. Open Links, then More under Internal links → Top linked pages, and use the table’s download button (up to 1,000 rows, as a CSV or Google Sheet).
2. For external links, the main page’s export offers Latest links and More sample links, each up to 100,000 rows.
3. Attach the internal export to your AI assistant’s chat.

Step 1: pick target pages from Search Console data

A target is a page you want to perform better. The best ones are pages Google already considers relevant, so a clearer signal has something to work with. Check three groups.

Pages with impressions at positions 5–20

These pages already appear at the bottom of page one or on page two. The window is wider than the 8–20 in our striking distance keywords playbook, because positions 5–7 are still worth moving and a link is cheaper than a rewrite. Work at page-and-query level: a page averaging position 9.0 might rank third for one valuable query and fortieth for fifty others.

At 1,000 rows per call, that’s three tool calls. Grouping by query leaves out anonymised queries, so per-query rows won’t add up exactly to a page’s total.

Pages that lost clicks

compare_search_periods with dimension set to page and compareTo set to previous_period (or previous_year for seasonal sites) returns your biggest losing pages by clicks. A page that slid off page one usually needs a refresh as well as links; the content decay audit covers how to tell decay from seasonality.

New pages Google hasn’t indexed yet

New pages have no performance data, so check them with inspect_url. A coverage state of “URL is unknown to Google” or “Discovered – currently not indexed”, with no referring page listed, suggests Google needs better paths to the page; links from pages it already crawls provide them. If the page was crawled and still not indexed, links alone rarely fix it; see how to fix “Crawled – currently not indexed”.

Pages to leave out

• Pages where inspect_url shows a Google-selected canonical that differs from yours. Fix the duplicate first.
• Pages that redirect, return errors or carry a noindex rule.
• The weaker of two pages competing for the same query. Run find_keyword_cannibalization first and settle which page should rank, using our keyword cannibalization guide.

Step 2: find source pages that rank for related queries

A source is the page that carries the new link. If a page earns clicks for queries related to the target’s topic, Google already associates it with that topic and real readers arrive on it, so the link is likely to be seen and followed. Use query_search_analytics with the page and query dimensions and a query filter: contains for one word, or includingRegex for a group of related terms (Search Console uses RE2 syntax; the Search Console regex cheat sheet has tested patterns). Add a notEquals page filter to drop the target itself. Up to 10 filters can be combined.

Before a candidate goes into the plan, check three things:

• Does it already link to the target? Our server doesn’t crawl your site, so check the source file in your repo or CMS, or select the target in the Links report’s internal table to see a sample of the pages linking to it.
• Is there a natural place for the link? If you’d have to write a new paragraph just to hold it, choose another source.
• How many new links is it carrying? Google gives no number. We keep to two or three per source per round, so each has its own sentence of context.

Step 3: choose anchor text from the queries a page ranks for

The queries a page already earns impressions for are the vocabulary searchers and Google associate with it. Pull them, then write anchors that use that vocabulary naturally, and differently each time.

• Vary across sources. Different sources, different real queries. The same exact phrase on thirty pages reads like a template, because it is one.
• One phrase, one target. Don’t point the same anchor at two different pages; mixed signals from your own links feed cannibalisation.
• Navigation is different. Short labels such as “Pricing” are normal in menus. Descriptive anchors matter most in body copy.

Step 4: build the link plan

Put everything in one table before anyone edits a page. It keeps the reasoning visible and gives you something to measure against. Here is an example for a fictional site, example.com, which sells invoicing software to freelancers. The numbers are illustrative.

Start with targets that have the most impressions in the 5–20 window and matter to the business. If the query data shows a theme where you have no page at all, that’s a content gap rather than a linking opportunity; our guide to keyword research with Search Console covers those.

Find orphan pages with sitemaps and URL Inspection

An orphan page has no internal links pointing to it. Google can still find it through a sitemap or a link from another site, but Google’s sitemap documentation says a sitemap doesn’t guarantee crawling or indexing. Orphans usually come from removed navigation, retired category pages, campaign landing pages and posts published without a link from anywhere.

The three-list method

1. URLs you want indexed: your XML sitemap, a CMS export or, for a static site, the routes in your repo. list_sitemaps confirms which sitemaps Google and Bing have, when they were last downloaded and their error counts, but it doesn’t list the URLs inside them or how many are indexed.
2. URLs reachable by links: a crawl from your home page with a crawler such as Screaming Frog or Sitebulb, with sitemap URLs kept out of the crawl (our server doesn’t crawl sites). In a repo, an AI code editor can approximate this by collecting every internal href in your content and templates.
3. URLs Google shows: get_search_pages or query_search_analytics with the page dimension for the last 90 days.

Pages in the first or third list but not the second are orphan candidates. Pages only in the third list are URLs Google shows that you don’t declare, such as old paths or parameter versions.

Confirm each candidate with URL Inspection

inspect_url returns Google’s verdict, coverage state, last crawl time and both canonicals. The underlying API result also lists sitemaps that include the URL and referring URLs, as far as Google knows them. Google’s URL Inspection help says the referring page list isn’t exhaustive, so an empty one is a hint, not proof. Google allows 2,000 inspections per property per day, which is plenty for an orphan check.

Each inspection inside a multicall counts as one tool call, so that prompt uses 25. Then fix each orphan in one of three ways:

• Still useful: link to it from its hub and from one or two related pages chosen with step 2.
• Obsolete: 301 redirect it to a close equivalent, or return a 404 or 410 if there isn’t one, and remove it from the sitemap.
• Deliberately unlinked, such as paid campaign landing pages: keep it out of the sitemap, and add a noindex rule if it shouldn’t appear in search at all.

Add the links with an AI code editor

If your site lives in a repo (Next.js, Astro, Hugo or MDX content), Claude Code, Cursor or Codex can apply the plan with Search Console data and your files in the same session. Our Claude Code SEO guide covers the setup and the wider editor workflow; for Cursor and VS Code, see technical SEO in your code editor. Three rules for the edit itself:

• Ship real links. The rendered HTML must contain <a href>. In Next.js, the <Link> component renders one; a button that calls router.push() doesn’t.
• Edit existing sentences. Ask for the smallest change that makes the link read naturally, and review every diff.
• Treat template links separately. Related-post blocks, breadcrumbs and hub navigation change hundreds of pages at once, so give them their own commit and review.

After you deploy, view the page source to confirm the links are in the HTML. If a component inserts links with JavaScript, use Test live URL in Search Console’s URL Inspection tool and check the HTML under View tested page. Our read-only inspect_url can’t run live tests or request indexing; do both in the Search Console interface.

Measure the results after 4–8 weeks

1. Mark the date. In the Performance report, right-click the chart and add an annotation such as “Internal links round 3: 14 links, 6 targets” (120 characters at most).
2. Confirm the recrawl. Google has to recrawl the source page before it sees the new link. Check the last crawl date of each source with inspect_url.
3. Wait four to eight weeks, then compare an equal period before and after, clicks first, then position for the queries you used as anchors.
4. Use a control group of similar pages you didn’t link. If they rose as much as your targets, something else, such as a core update or seasonality, probably caused the gain.

Both of those periods fall after 27 April 2026, the end of the impressions logging error listed on Google’s Search Console data anomalies page, so impressions and position are comparable as well as clicks. Be honest about the result. If a target hasn’t moved after eight weeks, links weren’t its bottleneck: look at content, intent or competition from your own pages next.

From one-off fixes to an internal linking strategy

Hub-and-spoke topic clusters

A topic cluster is a hub page that covers a subject broadly, linked to and from spoke pages that each cover one part in depth. Search Console lets you draw clusters from data: run a query regex for the topic, list the pages that get impressions for it, and the page with the widest spread of queries is your natural hub. The hub links to every spoke; each spoke links back with a descriptive anchor, and to sibling spokes only where a reader would plausibly move between them. Keep it in proportion: on a 50-page site, a few well-placed contextual links matter more than a formal cluster diagram.

A quarterly routine

1. Refresh the target list (step 1).
2. Export the Links report’s internal table and file it.
3. Run the orphan check on anything published since last time.
4. Add 10–20 links from a new link plan.
5. Measure the previous round against its control group.

By our count, a round for ten targets is about 35 tool calls: three to five to find targets, then one per target each for sources, anchors and an inspection. The Free plan’s 30 calls a month covers a round of seven or eight targets; regular rounds across several sites fit easily in the Pro plan’s 10,000 (see pricing). For more ready-made prompts, see our 40 Search Console prompts for Claude and ChatGPT.

Frequently asked questions