Internal Links and SEO: What 21 of Them Actually Did

How I tested internal links for SEO on a real site

A client of mine runs a business-to-business site of about 66 pages. Good pages, written years ago, most of them sitting on page two or three of Google and staying there.

Between 22 June and 6 July I added 21 internal links to that site. Real sentences inside the body copy, each one approved by the client before it went in, deployed in three batches and verified live on the page afterwards. That was the entire change. No new pages. No rewritten titles. No new listings or directory submissions. The content queue was deliberately parked for the duration.

Twenty-one links landed on 14 destination pages. The site’s other fifty-odd pages received nothing.

That last sentence is the part that makes this worth reading. Fourteen treated pages, fifty-two untreated ones, on the same domain, in the same weeks, subject to the same Google updates and the same summer trade. A control group, which almost no SEO case study has, and which I did not set out to build.

What the rank tracker said about the internal links

Two weeks after the last batch went in, the rank tracker delivered a result that made the whole thing look settled.

On the 20 July reading, one term had gone from 19 to 9 and another from 19 to 8. Page one, on both, a fortnight after a change that cost nothing but an afternoon. That is a story I could have sold, and to be honest it is a story I nearly wrote.

The next pull was on 2 August. The first term was back at 18 and the second at 24.

Look at that first line again with the whole series in view: 23, 18, 18, 16, 16, 19, 9, 18. The 9 is one reading standing outside a band the term had been sitting in for two months and returned to immediately. It is not a result. It is a wobble that happened to be pointing the right way on the day somebody looked.

That came very close to being published as fact, and the mechanism is the ordinary one. The tracker ran on its own schedule. It produced a number I was hoping for. I had a client to report to and a genuinely good piece of work to report. Nobody had to be dishonest for that to become “a definitive improvement we can directly attribute”.

What Google’s own record said about the same internal links

So I went to the source that does not have a marketing department: Search Console, which is Google telling you what Google did.

I pulled every page on the site, every day, from early May to the start of September, and split it into the 14 pages the links point into and everything else.

The lines cross. Before the links the treated pages were the weaker half of the site and behaving like it. Afterwards they sit above their own average in every week but two, while the rest of the site sits below its own in every week but one.

And there is a bump in that orange line at W25 that I will come back to, because it is the best argument against everything I am about to say.

Here is the same finding without the indexing, because indexed charts can flatter and you should always be able to see the raw shape.

Clicks on the linked-to pages went from an average of 4.9 a week to 8.1. Over the same weeks, clicks on the untreated pages went from 45.7 to 35.4. Average position on the treated pages improved by 1.7 places while the untreated pages lost 2.2. Click-through rate on the treated pages rose 74% against the rest of the site’s 6%.

Three measures, all moving the same way at once, with an untreated control group going the other way. The weekly click counts on the treated pages separate cleanly too: 6, 6, 3, 4, 4, 3, 8 before, and 10, 8, 8, 8, 6, 6, 11 after.

Why I still will not tell you internal linking is proven

Four things stand between that and a claim I would defend without qualification, and I would rather put them in front of you than have you find them.

The control group is not clean. Impressions on the untreated pages fell 27% over the same period while the treated pages held level. Something else was happening to the rest of the site independently of anything I did. A control that is itself in motion weakens the comparison in both directions.

The numbers are small. Three to eleven clicks a week. That difference is real money to a business with a long sales cycle and a high order value, and it is fragile as evidence. A fortnight of ordinary variation could produce a good chunk of it.

The seasons are not matched. The before window is May and June. The after window is July and August, and August is quiet for business-to-business trade in a way that would depress the busier pages more than the quiet ones.

The treated pages started worse. Position 24.9 against the control’s 20.4. They had further to climb and more room to regress towards the average, and I cannot separate that from the treatment.

And the lift starts a week early. This is the one that bothers me. Look again at the second chart: the treated pages jump in W25, which ran from 15 to 21 June. The first link went in on 22 June. That week they took eight clicks, which is exactly their average for the seven weeks after the links. So the strongest single data point against my own reading is sitting in my own chart, a week before the thing I am crediting. It could be ordinary variation on numbers this small. It could be that the improvement was already under way for reasons I never identified and the links joined a trend rather than starting one. I do not know which, and neither does anyone who shows you a chart like this without pointing at that week.

How to do internal linking on your own site

The practical shape of it is simple, and none of it needs a specialist.

Find the pages that already get read. In Search Console, sort your pages by impressions. The top few are the pages Google already trusts enough to show people.

Find the pages that make you money. Usually a service page, a product category, or the page with the enquiry form on it. On most small sites these are not the same pages as the ones above, and that gap is the whole opportunity.

Write the sentence that joins them. Not a “related links” block in the footer, and not the words “click here”. A sentence in the body copy that a person would read anyway, with the link sitting on words that describe what is at the other end. On this site 21 of those took an afternoon to write and three short sessions to deploy.

Then measure it like this, not like the tracker. Pick the pages you are treating and write the list down before you start. Change nothing else for a fortnight, which is the hard part. Then compare the treated pages against the untreated ones over the same weeks, rather than comparing the treated pages against their own past. Everything on the internet moved between May and August; only the difference between your two groups tells you anything.

If you take one thing from this, take the shape of the mistake rather than the result. I had a tracker telling me one thing and Google’s own record telling me something smaller. Both were describing the same fortnight on the same site. The number that arrives with a marketing story attached is the one to check first, and it is usually the one nobody checks at all.

That is what 21 links did. If you want a second opinion on what your own pages are doing, send me what you are seeing and I will tell you what I make of it.

Related: Impressions But No Clicks: The Ten-Minute Check · Improve Your Website Before You Rebuild It · Keyword Cannibalisation