How entries are written

Every entry here follows the same process. It is written down so you can judge the output against it.

How topics were chosen

The entry list is not a guess about what people should want to know. It was built from search demand: roughly 1,900 parent topics from Ahrefs, expanded with related-keyword and question data from DataForSEO, covering about 21,000 distinct phrasings.

That list was then cut hard. Non-English queries were removed because this is an English-language reference and a bad translation helps nobody. City and "hire an agency" queries were removed because they are shopping questions, not reference questions, and this site does not sell anything. Topics whose surrounding search data showed they were not really about search at all were removed. What survived became the entry list.

How duplicates were removed

Two entries that answer the same question are worse than one. They split the explanation, they contradict each other eventually, and they force the reader to decide which page to trust.

So before anything was written, every topic was compared against every other on four measures: overlap in the search results they compete in, whether one phrasing is simply a subset of another, whether each appears in the other's related-term list, and whether the difference between them changes the intent rather than just the wording. Where two topics were the same question in different words, they became one entry, and the alternative phrasings became that entry's secondary terms and its questions section.

That merged 294 near-duplicates into 192 entries and removed 209 more outright. It is why there is one entry on canonical tags rather than five.

The shape of an entry

Fixed, deliberately. The first sentence is the answer, written so that a reader who stops there is still right. Then a short bulleted summary and a facts box, all of which fit on one phone screen. Then the detail in short sections, a comparison table where the comparison is genuinely tabular, and the questions people actually ask about the term, taken from real search data rather than invented.

Entries have a word ceiling — 1,300 for the broadest topics, 900 for most, 600 for narrow ones. An entry that is complete in 340 words is published at 340 words. Padding is the exact failure this site was built to avoid.

What counts as a source

Primary documentation first: Google Search Central, Bing Webmaster documentation, the W3C, schema.org, IETF RFCs, and the official docs of whatever product an entry is about. Where a claim comes from published research or a large-scale study, the study is cited, not a blog post summarising it.

Sources are linked with nofollow and open in a new tab. That is not a slight on them; it means no site can gain anything by getting itself cited here, which keeps the citation list about accuracy rather than about favours.

Where something is genuinely unknown outside Google — how a signal is weighted, whether a patent is actually in use — the entry says so. "Google has not documented this" is a legitimate and frequently correct answer.

How AI is used, and how it is not

Being straight about this matters more than the answer looks like it should.

Drafting is machine-assisted. Each entry starts from a research brief assembled from the sources above, and a language model produces a first draft against a fixed structure and a fixed style guide. Every draft is then checked automatically for things a machine can check — that it does not exceed its length, that its sources are real URLs from the approved list and still resolve, that it does not repeat a neighbouring entry, that it contains none of the marketing filler this site exists to avoid — and it is read by me before it is published.

What a model is never allowed to do here: invent a statistic, invent a Google statement, invent a source, or invent a link. Sources come from the research stage and are verified independently. Internal links come from a pre-computed map of the site, which is why no link on Rankpedia points at a page that does not exist.

If you find an entry that reads like it was generated and not checked, that is a failure of my process, not an accepted cost of it. Tell me and I will fix it and log it.

How entries are kept current

Search changes, and a reference site that quietly rots is worse than no reference site. Each entry shows the date I last read it end to end and confirmed it was still true. That date is not bumped by a rebuild that changed nothing — a site-wide refresh with no actual change is a lie told to a crawler.

Entries covering things that move quickly are re-read more often than entries covering things that do not. When an entry changes materially, the change is listed on the changelog.

Known limits

One person, so coverage is uneven and the deepest entries are in the areas I have worked in most. The demand data behind the topic list is US-weighted, so a handful of regional practices are underrepresented. Some entries cover commercial tools; those describe what the tool does and are not endorsements, and no vendor has been consulted, paid, or given a preview.

And a general one worth stating plainly: nobody outside the search engines can verify how their ranking systems weigh anything. Where this site describes a mechanism, it is describing documented behaviour and observed behaviour, and it tries hard to say which is which.