Anchor text ratios and what actually matters

Two anchor text guides, two browser tabs. One says keep exact match under 1% of your profile. The other says 10% is the safe ceiling.

Same confident tone. Same tidy table. Neither one says where the number came from.

That is a tenfold disagreement about a quantity that both of them describe as the thing standing between you and a penalty.

The trail always goes cold

I have tried to source these percentages properly, more than once, and I have never got anywhere. Blog post cites blog post cites a conference deck from an agency that has since been absorbed into something else. Somewhere around 2014 the trail stops.

No sample size published. No word on how many domains anyone looked at. No definition of how an anchor got sorted into partial match instead of generic, which is a judgement call that would shift every number in the table.

I do not think anyone invented this in bad faith. Somebody reasonable looked at the profiles they had access to, wrote down roughly what they saw, and the observation got rounded into a rule and then hardened into a target. The engineering leaked out somewhere between the second and third retelling. What survived was a set of numbers that add up to 100 and answer to nobody.

The thing underneath it is real

Here is where I have to be fair to the folklore, because it is gesturing at something true.

Links that people write on their own initiative do not agree on wording. Someone drops your site into a forum reply as “here”. Someone pastes the bare domain because they copied it out of the address bar. Someone types your company name and spells it wrong. Someone links the words “this article” because that is the way their sentence happened to run.

Fifty people writing independently produce fifty phrasings, and most of them are worthless as keywords, because none of those fifty people was thinking about your keywords.

So a profile where a big share of the links carry the same commercial phrase had a hand behind it. You can see that without a statistical test. It is the same as noticing that forty letters arrived in the same handwriting.

I believe that part, and I believe it from the sell side rather than from a chart. Fourteen people have asked to buy a link from me. I also keep a ledger of every placement sitting on our sites for somebody else, and the anchor column of that file is the whole argument in one place. Almost every requested anchor is a commercial phrase with nothing around it. Nobody has ever asked me to link them with the word “here”, because nobody pays for “here”. Which is exactly why “here” is everywhere in an earned profile and almost nowhere in a bought one.

My own brand name sits in two buckets at once

This is the smaller problem that finished the argument for me.

The buckets are not defined well enough to count with.

My site is called The SEO Desk. A link reading “The SEO Desk” is branded. A link reading “SEO desk” is branded. A link reading “SEO” is exact match. The distance between the second and the third is one word out of three, and the classification flips completely.

Anyone selling proxies, hosting or VPNs has it worse, because the brand contains the category. A link carrying your product name is branded and commercial at the same moment, and which column it lands in depends entirely on who is doing the sorting. Two tools will classify the same profile differently. I have watched that happen on one domain inside the same week.

So before anyone argues about whether exact match belongs at 5% or 8%, somebody has to agree what exact match means. Nobody has.

A percentage needs a denominator worth having

Most people watching these ratios do not have enough links for a percentage to carry information.

At 40 referring links, buying one placement moves your exact match share by two and a half points. The dashboard draws a trend line. What actually happened is that you bought one link.

It gets worse. The denominator is only the set of links one crawler managed to find. Tools disagree about what is in a backlink profile by wide margins, so the numerator and the denominator are both estimates with their own coverage gaps, and the ratio inherits every one of them. You are computing a precise number out of two fuzzy ones and then reading it to one decimal place.

Optimising the ratio shapes the ratio

Say you take the targets seriously and manage to them.

You have optimised the number. The thing the number was standing in for is untouched.

It was standing in for whether a human chose the wording or you did. Choosing more varied wording does not fix that, because you are still the one choosing. You have moved from one hand writing forty identical anchors to one hand writing forty deliberately different ones.

And a profile engineered to land on a target distribution is itself a pattern. Real profiles are lumpy and ugly. One site links you four times in a month with the identical phrase because that is what its template does. A forum thread hands you nine links that all say “here”. A directory gives you your company name in title case because the form field was labelled Company Name. Nobody’s honest profile looks like a clean pie chart.

I am not claiming anyone detects manufactured variety. I am saying that if you believe uniformity is a signal, you should be at least as suspicious of engineered diversity, and the people selling ratio management never are.

Authorship is the variable

The reason bought links cluster has nothing to do with percentages.

When you buy a placement, you choose the anchor. The seller sends a form or an email asking what it should say, and you type the phrase you want to rank for, because that is what you think you are paying for. Typing anything else feels like throwing money away.

When you earn a link, someone else chooses, and they choose badly from your point of view. They choose whatever fits their sentence.

So what separates a bought profile from an earned one is authorship. One was written by a single hand, the other by dozens, and every anchor you specified yourself has your intent sitting inside it. A percentage is one crude way to look at that fingerprint. Not a good one.

What I do when I am buying

Send the destination URL and one sentence about what the page covers, then let the writer phrase it. You get worse keywords and better placements.

Vary because the pages are different, not because a spreadsheet said so. Your home page gets your brand, because that is what a person would call it. A pricing page gets whatever that page is actually about.

Do not buy ten placements inside a fortnight with the same phrase from ten different sellers. Timing is the other half of the pattern, and it is the half people forget while they are busy with the first half.

If a placement only works with an anchor that makes the sentence read badly, buy a different placement.

Stop asking for your hardest commercial term on a paid link in the first month of a site’s life. A site with nine links total, four of which say the same three words, is doing something no real site has ever done.

The test that actually applies

Does the link make sense in the sentence it sits in.

That is it, and it is irritating precisely because you cannot put it on a dashboard. Read the sentence with the link removed. If it still tells a reader something they needed, the anchor is fine whatever words it uses. If the sentence exists in order to hold the link, everyone can tell, and no ratio rescues it.

I have pulled paid insertions out of my own articles over this. The anchor was never the problem. The sentence around it was visibly bolted on, and it made the page read like rented shelf space. A sentence written to carry a link is obvious to a human editor in about two seconds, and a human editor is who actually deletes it.

Counting to two decimal places

If you are tracking your anchor distribution to two decimal places, you have found a way to feel productive without doing anything.

Ratio management is deeply satisfying. There is a spreadsheet. There are targets. There is a number that responds when you act. It has every texture of work except the outcome. Meanwhile the variables that move something are which sites you are on and whether your paragraph belongs on the page, and neither of those fits in a chart or changes inside a week.

I ran an anchor spreadsheet myself for about a year. Colour coded, maintained properly, purchases adjusted to keep the exact match column inside a range I had read somewhere. I cannot point at one thing that changed because of it.

What I cannot prove

I have never run the test that would settle this, and neither has anyone else I know of. You would need two near identical sites on the same timeline, and a willingness to deliberately over optimise one of them and watch it burn. Nobody with money on a site is doing that.

What I have is the sell side view, a set of ledgers, and the fact that every percentage ever quoted at me evaporates the moment I ask where it came from. That is enough to stop me managing to a ratio. It is not enough to tell you the ratio does not matter, and anyone stating either version with certainty is doing the same thing the spreadsheet people are doing, just aimed the other way.

My own domain vets at 48% spam on a vendor tool while it sits indexed and keeps selling placements. So what I think looks unnatural and what that tool thinks looks unnatural are already two different opinions. Adding a third one with a decimal point on it does not obviously help.

What I check before money moves, and the wording I put in the email, is here.