One sentence was ranking at position nine, so we wrote the other two thousand

Our own side project, and the only site here where we can show every decision without a client to protect. A single passing mention of one word was ranking at position 9 — that is not a fluke, it is an instruction.

Visit Nat20 Labs →

The numbers

Whose site this is, and why we are showing you

Nat20 Labs is not a client. It is a side project belonging to Karl, who runs Leodis Digital — a tabletop gaming site with a verifiable dice roller and a library of rules guides, published under his middle name.

We are including it here for one reason: it is the only site we can show you in full. On client work we can quote a figure and describe an approach, but not publish the search data, the mistakes, or the arguments about what was worth doing. Here we can, because the only person who could object is us.

So treat this as working shown rather than a result sold. Everything below is real, including the part where we told ourselves the honest answer was to stop.

The signal almost nobody looks for

Nat20 Labs publishes tabletop roleplaying rules guides. Clicks were up 250% and average position had moved from 17.0 to 13.8, so the site was already working — it just sat on page two for everything.

Then there was one line in the data that did not fit. The site ranked at position 9.2 for the search “pf2e retch”. We checked the post. The word appeared exactly once, in a sentence about a troll, as an aside.

One passing mention, page one. That is not luck. It is Google saying there is demand here and almost no supply — and it is the single most actionable thing that can appear in a Search Console report.

So we wrote the section that should have been there: what the action actually does, which DC you roll against, the restriction that decides more fights than the numeric penalty does, and the five arguments that come up at real tables. Plus the quick-reference table people are really searching for when they type a rules query.

The advice we got wrong

We had previously concluded the click-through rate was suffering from weak page titles. That was wrong, and it is worth showing because we would have sold that same diagnosis to a client.

We had read the URL slugs rather than the actual title tags, which were mostly fine. A 0.3–0.6% click rate at average position 11–16 is simply what page two looks like. The problem was position, not presentation — a completely different piece of work, and the version we would have sold first would not have helped.

We are also clear-eyed about the ceiling. “dnd 2024 stealth” drew 16 searches in 90 days. Ranking first for everything the site currently ranks for might produce a few hundred clicks a year. Growing it fivefold is very achievable and it will still be small, so the real route is the free dice tool, the community, and something to sell — not more blog posts. Telling a client their niche has a low ceiling is not a comfortable conversation. It is easier to reach that conclusion on your own site, which is exactly why running one is useful — we hit the wall here first, and now we recognise it.

The thing nobody had noticed

While working through the library we checked the publish metadata, and 17 of 27 posts carried a “last updated” date that fell before their own publication date.

It was a bulk-set artefact from an earlier migration — every one of them read the same day. The effect was that the site's structured data was telling Google each article had been modified before it existed. That is not a signal worth sending, and it is exactly the kind of fault that survives for years because it is invisible on the page.

We removed the invalid dates rather than inventing replacements, and added an opt-in review-date field instead, shown on the page as “Rules checked —” and claimed only where a human has actually done it. In a niche where errata land regularly, a genuine review date is both a reader trust signal and a freshness signal. A fabricated one is neither.

Services used

← All case studies