Long or short content? Word count is the wrong question

Set aside anyone who answers this question with a number. Google has already answered it, and the answer isn't a number.

Google's Search Liaison, Danny Sullivan, wrote in 2023 that the best word count for succeeding in Search is not a thing — it doesn't exist. Search Advocate John Mueller has said the same repeatedly: word count is not a ranking factor.

And yet figures like "the ideal blog post is 1,890 words" keep circulating. It's worth looking at where they come from, because the mistake sits in exactly one place and it's an instructive one.

A round-up of those statements with dates · Google's people-first content guidance

Where the correlation comes from

The studies producing those numbers are built like this: collect the pages on the first page of search results, measure their word counts, take the average. The resulting number is real. What it says is not "write longer and you'll rank."

Because the same data fits at least two other explanations. A piece that genuinely covers a topic ends up long — length is the consequence of coverage. And the page types that attract links (guides, comparisons, research round-ups) are long to begin with; the links do the ranking and the length rides along.

So the measurement is right and the inference is backwards. A page that ranks because it's long and a page that's long because it's the ranking kind look identical in that data.

In Google's own framing, what matters is topical coverage: actually addressing the relevant concepts, facts and subtopics. Coverage usually brings words with it. Words don't bring coverage.

I don't have data. I have numbers.

Here's where I have to be honest. There are seven published articles on this site and none of them has traffic data — the site is new and measurement was only just connected. So I have nothing that lets me say "my longer pieces do better," and I'm not going to say it without that.

What I can show is the distribution. Word counts across the seven: 976, 832, 718, 706, 592, 546, 544. The longest is 1.8 times the shortest.

That spread isn't random. The longest piece walks through six separate experiments with their sources, including one replication that collapsed; making it shorter would mean throwing something out. The shortest covers five questions and the incident sitting under each one; making it longer would mean padding something.

I didn't choose the length. The amount of material chose it.

Changing the question

The version of "long or short" that actually helps is this: is there anything in here I could remove?

That question doesn't give you a number, but it gives you a direction. If a paragraph can come out, the piece is too long, whatever the word count says. When nothing is left to cut, length stops being a problem and becomes just a measurement.

The reverse is equally true and said less often: cutting a piece doesn't make it good. Strip out the source, the number and the counterexample and what's left isn't short, it's an assertion. Brevity is not a virtue on its own either.

What I don't know

I don't know which length works better for my audience, and I won't know for months. Finding out requires readership data, and that only exists once there are readers.

When it does exist I'll have to be careful with it. A piece getting read more could be its length, or its subject, or where it was shared, or something else entirely that happened that week. What I said above about other people's data will apply to mine: correlation isn't cause.