Classify Who Ranks: Forums, Brands, Niche Sites, Aggregators

Full transcript

1,015 words · script draft, under review.

Intro

Look at two results pages. On the first, the top ten is all big brands and marketplaces. On the second, half the top ten is forum threads and small hobby sites.

Both keywords have the same volume. But they are completely different fights. On the first page you're up against companies with huge budgets. On the second, Google is basically telling you it couldn't find a good dedicated page.

In episode twenty-six we wrote a short label for each result by hand. That works for ten keywords. For four hundred, you need a fixed set of site types, a way to label domains quickly, and a number that summarizes each results page.

In this episode: a domain-type taxonomy, how to label a whole keyword set in an afternoon, and how to measure the share of the top ten per type.

A fixed set of site types

First, a fixed label set. Seven types cover almost everything I see.

Brands and retailers: companies selling their own products or a large range, under their own name.

Marketplaces and aggregators: sites that list other people's products, services or data. Big shopping platforms, directories, comparison and review aggregators.

Forums and user-generated content: discussion sites, question-and-answer sites, community threads. Sites like Reddit and Quora belong here.

Niche sites: independent sites focused on one topic. Often small, often specialists.

Publishers: magazines, newspapers and media sites with broad coverage.

Reference and institutions: encyclopedias like Wikipedia, government sites, universities, official bodies.

Video and social: YouTube and similar platforms.

And one more: unknown. Anything you haven't labeled yet goes there, so it's visible instead of hidden.

The exact list matters less than keeping it fixed. If you add a new label every week, you can't compare this month with last month.

Labeling fast: frequency first

You don't label every domain. You label the ones that matter.

Start from the snapshots stored in episode twenty-seven. Count how many top-ten slots each domain takes across your whole keyword set. Sort from most to least.

What you'll almost always see is a steep curve. A few dozen domains take a large share of all slots. Then there's a long tail of domains that appear once or twice.

Label the top of the list by hand. Open the site, look for a minute, choose a label. Thirty or forty domains take maybe half an hour, and they usually cover most of your slots.

Then add simple rules for the tail. Government and university domains go to reference. URLs with forum, threads, community or questions in the path go to forums. A short list of known video platforms goes to video. Everything else stays unknown until you have time, or until it climbs the frequency list.

You can also let a language model suggest labels from the domain, title and description, with the same fixed label set. Treat its output as a draft. Check a sample by hand before you trust it.

Label the page, not only the domain

One complication. Some large sites are several types at once. A big retailer may also host a community forum. A publisher may have a shopping section.

For those, label the result, not just the domain. The rule order is: URL pattern first, then the domain label. So a forum thread on a retailer's site counts as a forum, because that's what's ranking.

Also decide how to handle subdomains. I treat a subdomain as its own entry when it's clearly a different kind of site, like a forum subdomain, and as part of the main domain otherwise. Write the rule down, and apply it the same way every time.

Share of top 10 per type

Now the summary number. For each keyword, count how many of the top ten results belong to each type. Four brands, three marketplaces, two forums, one niche site. Divide by ten, and you have the share of the top ten per type.

Here are made-up numbers for illustration, for twelve keywords on example-shop.test. Some bars are almost all brands and marketplaces. Some have two or three forum threads. A few have several niche sites.

Sort by the share of forums and niche sites. Those are the results pages where an independent site has already proved it can get in. That's not the same as easy. But it's where you start looking.

Then summarize across all keywords. Which types own your market overall? If marketplaces take most of the slots everywhere, that's a strategic fact, not a detail. It shapes what kind of pages you can realistically build.

It's also worth splitting the share by position. Count types in the top three separately from positions four to ten. A forum at position nine is a hint. A forum at position two is a strong signal that Google wants a real answer and can't find one. In my experience, the top three and the rest of page one often tell slightly different stories, and the difference is where the opportunities hide.

Store the result per keyword and per date. When you rerun the classification next quarter, you'll see whether forums are being pushed out, or whether marketplaces are taking more of the page.

Pitfalls

Three pitfalls.

Labels that drift. If two people label domains, or you label in two sessions months apart, the same kind of site ends up in different boxes. Keep one table of domain labels and reuse it for every analysis.

Hiding the unknowns. If a quarter of your slots are unknown, any share you report is shaky. Show the unknown share next to every chart.

And mistaking type for strength. A forum in the top ten suggests a gap, but some forums are very strong pages. A niche site might be weak or might be the best page on the internet. Type tells you what kind of competitor it is. How hard it is to beat is episode thirty-one.

Wrap-up

So: a fixed set of site types, label the frequent domains by hand, use rules for the tail, and report the share of the top ten per type. The taxonomy and labeling rules are linked in the description. Next episode: we read search intent directly from the results page, instead of guessing it from the keyword. See you there.