I was sceptical going in. Internal linking is one of those things everyone agrees matters in theory and almost nobody tests properly, because the honest version of the test is boring: build hundreds of links by hand, wait a month, and compare. So we did the boring version. Across eight bookmaker groups on a large sports betting affiliate, contextual links to operator pages lifted clicks per day by 79.7%, impressions by 297% and average position by 19%, with statistical significance on every group. This is the method, the numbers, and the caveat I would want to read if I were you.
In this article
The situation
The site was a major sports betting affiliate, a category where every page competes hard for commercial keywords and the silo structure genuinely matters. The bookmaker section had the usual shape: a main review per operator, then sub-articles for the queries people actually type (the app, the welcome bonus, live betting, payment methods). A loose content strategy existed. Internal linking existed too, but almost entirely in “related articles” blocks at the bottom of pages.
Contextual links, the ones that sit inside a paragraph and carry real semantic signal through their anchor text and surrounding sentence, were sparse. The pages ranked for roughly the right things. The hunch was that they were leaving most of the section’s PageRank on the table.
The hypothesis, stated before we touched anything: if we built a proper topic-cluster model, main review as hub, sub-articles as spokes, with contextual links that use descriptive anchor text in both directions, we would see meaningful ranking changes on the operator pages. “Meaningful” was defined as a statistically significant lift in clicks and impressions in Search Console over a 28-day window.
Method
Finding the opportunities
The opportunity-finding step was built on Craig Campbell’s approach using Screaming Frog’s custom extraction. The idea is simple and scales well:
- Crawl the section and extract the body text of every page (custom extraction on the article container, text only).
- List the target pages and the keyword or phrase each one should be linked with (operator name, “app”, “bonus code”, and so on).
- In Google Sheets, flag every page whose body text contains a target phrase that is not already a link to the target page. A plain
=ISNUMBER(SEARCH(phrase, body))does the first pass. - Review every match by hand.
Step four is where the value is and where most teams cut corners. Not every string match deserves a link. You need to check whether the sentence actually supports it, whether the anchor reads as a description of the target page rather than a keyword stuffed into a clause, and whether the referring page already links to that target from somewhere sensible. For the bookmaker section alone the manual review left us with just over 300 in-content linking opportunities, organised in a sheet with referring page, target page, proposed anchor and a placement screenshot for the editor.
Measuring the effect
Once the links were live we set up group tests in SEOtesting.com, one per bookmaker group. The tool takes a URL group, compares a before period against an after period on Search Console data, and calculates a p-value so you can distinguish a lift from noise. We used 28 days on each side. New-user numbers came from the GA landing-page report for the same windows.
Before/after designs on the same URLs are quasi-experiments, not randomised trials. Seasonality and concurrent site changes are the two threats to validity that matter here. We addressed seasonality by checking that the comparison windows did not straddle a major fixture cluster; we could not fully address concurrent changes, which is discussed below.
Results
| Metric (all eight groups, aggregated) | Change |
|---|---|
| Clicks per day | +79.7% |
| Impressions per day | +297% |
| Average position | +19% (improved) |
| Groups reaching statistical significance | 8 of 8 |
Two things in the detail deserve comment. The first is one operator’s main review, which recorded a 200% increase in new users. That number is dramatic because the page had been essentially invisible before: zero clicks in the control window according to Search Console. The links gave Google enough reason to start showing it at all. That is not the typical case, but it illustrates the floor-to-ceiling range: a genuinely good page that is not surfaced can go from nothing to something on internal signal alone.
The impressions figure matters more than the clicks. A near-tripling of impressions means the pages did not just rank higher for the queries they already had. They started appearing for queries they had not been considered for before.
That is the second point and the one that surprised me. You expect clicks and position to move together after a linking change. But impressions tripling suggests something else: the pages were being understood differently. Descriptive anchors in a coherent cluster gave the crawler enough semantic context to associate the operator pages with a wider set of related queries. Linking did not just pass authority. It passed meaning.
The caveat
We could not fully isolate internal links as the only variable. Some pages had minor content edits in the same window: a comparison table added here, a section refreshed there. On a commercially important section with an active editorial team, freezing all changes for two months is a hard negotiation, and we did not win it. The scale and consistency of the lift across eight independent groups makes it very hard to attribute to those edits, but “very hard” is not “impossible”, and I would rather say so than pretend.
If I ran it again I would do two things differently. First, push harder for a content freeze during the test period, or at minimum log every other change so it can be checked against the group results. Second, keep a true control group: a comparable set of operator pages in the same section that received no new links. SEOtesting supports control groups; we simply did not use one here. With both in place, the design moves from “strongly suggestive” to “defensible in front of a sceptical engineering lead”, which is the standard I want.
How to run this on your own site
- Pick one section with commercial intent pages that should rank better. Do not audit the whole site. A cluster of 20 to 60 pages is enough to see an effect and small enough to review by hand.
- Write the hypothesis and the success metric down before you start. Clicks, impressions and position from Search Console, over fixed windows, with a significance threshold.
- Extract body text with Screaming Frog and match against target phrases in a sheet. Then review every match. Reject anything where the anchor is not a fair description of the target page.
- Link in both directions. Hub to spokes and spokes back to the hub, plus spoke to spoke where the context is real.
- Measure with group tests and a control group. Eyeballing a Search Console graph is not measurement.
- Log every other change to the section during the window. You will need it when someone asks what else moved.
The broader point
Internal linking is boring to talk about and genuinely impactful when done systematically. The failure mode is treating it as a one-off audit task rather than a repeatable process with a measurement step at the end. What made this project work was not a clever insight. It was a scalable way to find opportunities, a human review of every one of them, and a proper test instead of a vibe check.
If your site has a cluster of commercial pages that ought to be ranking better, this is one of the highest-leverage things you can do without writing a new article or asking engineering for a single change.