How Machine Learning Helps SEO

Machine learning helps SEO by taking away the one thing this trade used to run on. A rule you could read.

That sounds like a loss. It is closer to the best thing to happen to a small business in this trade.

The Three Answers Everybody Gives

Ask the question and three answers come back. Machine learning sits inside Google’s ranking systems. It sits inside the tools you already pay for. Then there is the third. You could point a model at your own data.

All three are correct, and covered elsewhere on this site.

Still, each says where the machine learning is. None says what it did to the work.

A Rule Somebody Wrote, and a Rule a Machine Settled Into

Two kinds of rule can decide where your page ranks. They are not the same kind of object.

The first kind gets written. Somebody sits down and types it out. Count the links pointing at this page. Every line has an author. Ask the author what he meant and the rule reads back in a sentence.

The second kind never gets typed at all.

A model is shown an enormous pile of examples. Searches that ended well. Others that ended with somebody hitting the back button. Then it gets nudged, over and over, until it reproduces the pattern. What it arrives at is a shape. Millions of numbers, each meaningless alone.

Nobody typed that. So nobody reads it back.

Nobody Is Keeping This Secret

The usual line is that nobody outside Google knows the search ranking rules. Which quietly promises that somebody inside does. Get hold of the right document and the sentence would be there.

Documents have leaked. Thousands of named features came out of them, and the industry read every line. What nobody found was the sentence.

A list of the things a model was shown is not the rule it settled into.

So a secret needs somebody keeping it. The judging has nobody.

People who train a model can tell you what it was shown. They cannot hand you a rule you could read back, because nobody ever wrote one. Whole research careers go into prying explanations out of trained models, which tells you how easily those explanations come.

None of that is a claim about Google. It is a claim about machine learning. Point the technology at loan applications and the property holds.

Two Things Happened at Once and Nobody Separated Them

Machine learning arrived in search and did two things at once.

The judging got better. Worth saying plainly. A written rule counts things. Words. Links pointing in. What it cannot do is tell a page written for a reader from one assembled for a crawler. A learned model has a better chance at it.

Then the judging stopped being sayable. Same afternoon, same system.

Those were one event. The industry mostly reported the first half.

Why SEO Advice Got Vaguer Every Year

Go back far enough and the advice arrived with numbers in it. Keyword density, chased to a decimal point. Whether those numbers ever worked is settled elsewhere.

Read the good advice today. Answer the question better than the pages above you. Be worth linking to.

It reads like a platitude. Or like somebody dodging.

It is neither. Instead it is the only truthful thing left to say about a judgment nobody wrote down. The advice went vague at precisely the rate the machine got good. Those are one fact, seen from opposite ends.

Anybody still selling you a number for the merit of a page is selling you a rule that stopped existing. Worth knowing what the RankBrain system does before you buy another number. The knowable parts still get measured, mostly by crawlers like Screaming Frog SEO Spider.

What Machine Learning Actually Took Away

Machine learning did not hand you a technique. It took everybody else’s away.

A written rule has been read by somebody. That person knows a thing you do not, and the knowing is worth money. Much of this industry was exactly that. A number, a threshold, a loophole, priced by whoever found it first.

Meanwhile a learned rule has no reading to sell. No sentence anybody could have got hold of. The operator with the inside line on the algorithm has nothing to be clever about.

So the work stopped rewarding whoever knew the trick. It started rewarding whoever had something to say. That is the help. It arrived looking like a loss of control, which is why hardly anybody filed it under help.

The Half Machine Learning Never Touched

Machine learning came for the judging. It mostly left the machinery alone.

Whether a crawler can reach your page is no judgment. It is a fact you can check this morning. Page speed is a number. A broken link is broken. That half stays as knowable as it ever was. Which is why a technical audit comes with a checklist while nothing about your content ever will. That is what a website audit is for.

Every number you can still trust is a number about the machinery. Never about the merit.

It Was Fitted to People. You Have People.

Nobody is going to read that model back to you. Agreed. Yet the model was never a mystery about what it wanted.

It was fitted to people. Whatever sat in that pile of examples, people put it there. What they clicked. The pages they read to the end. What sent them straight back to the search results to try somebody else.

The model is a copy of your customer, blurred by the scale it was taken at. Meanwhile you have the original. Yours answers the phone.

So try this. Take the query that matters most to your business, then open the page sitting at the top of it. Read it the way a customer would. Now ask it the one question the model was fitted to answer.

Having read that, would you still need to look anywhere else?

Maybe not, and then you know what you are up against. Or you would, and you have just found the gap. No study of ranking factors was going to hand you that.

How To-The-TOP! Works Under a Rule Nobody Can State

SEO Company To-The-TOP! has been doing search engine optimization in Calgary since 2007.

The method never changed much, because it was never built on knowing the rule. We fix the machinery, since the machinery stays knowable. Then we chase the thing the model was fitted to find. A person, getting an answer, not needing a second one. The Calgary SEO work at To-The-TOP! lives in that second half.

No guarantees. Nor any numbers to chase. Three to six months before anything meaningful moves. The work sits in the SEO portfolio.

Need traffic while the slow half runs? Buy it by the click through Google Ads management. It stops the day you stop paying, and that is the trade. Ask about how AIO in SEO works if the answer boxes are eating your clicks.

Common Questions About Machine Learning and SEO

How does machine learning help SEO?

Not by handing you a new tactic. It helps by removing the advantage held by whoever knew the old ones. A written rule can be read, and whoever reads it first sells the reading. Not so a learned rule.

Why has SEO advice become so vague?

The rule it points at was never written in words. Old advice could quote figures because the things it described were counted. Today the judging is a fitted model, and little of it reduces to a figure.

Can any tool tell me exactly why a page ranks?

No tool has one, and be wary of the one that says otherwise. Tools measure things that move alongside ranking. Correlation is not the rule, and a hundred correlations stacked up still do not add to one. Deciding what to do stays a human job.

Has machine learning made SEO harder for a small business?

Shortcuts got harder, which is a different thing from the work getting harder. A business that genuinely answers what somebody typed is no longer up against a rulebook it never had.

What is still worth measuring?

Anything mechanical. Whether a crawler reaches your pages. How fast they load. Which links have broken. Those are facts, so they answer to numbers. Whether your page was worth somebody’s afternoon does not. Measure the machinery. Then judge the merit the way a reader would.