How to Write a Page That Holds Its Ranking

Getting a page to rank is a different skill from getting it to stay. Plenty of pages hit position six in month two and are gone by month nine, and the post-mortem usually blames an algorithm update when the real cause was that the page was never better than its neighbours — it was just newer.

Holding a position means being the page a searcher would still choose after they’ve seen the alternatives. That’s a higher bar than being thorough, and it’s mostly decided before you write the first sentence.

What’s the difference between ranking once and holding a ranking?

Ranking once means you cleared the relevance bar and got a trial. Holding means readers who clicked found what they came for and didn’t go back for something better. The first is a content and links problem; the second is entirely about whether the page actually satisfies.

New pages often get a short positional trial — a window where search engines test how people respond. If the response is weak, the page drifts down over the following weeks and most site owners never connect the two events. They see “we ranked and then we dropped” and assume volatility. It’s usually feedback.

The practical implication is that a page’s ceiling is set by how completely it answers, not by how much authority you point at it. Links get you the trial. The writing decides whether you keep the seat. That’s also why buying links for a mediocre page tends to produce a brief lift followed by a slide back — the mechanism worked, the page didn’t.

Why is word count a bad target?

Because length is a byproduct of covering a question properly, not a cause of ranking. Writing to a number produces padding, and padding is exactly what makes a page worse than a shorter competitor that answers cleanly. Aim at complete coverage and let the length land wherever it lands.

The “match the average length of the top ten” advice is where this goes wrong. That average is a description of what happened to win, filtered through whatever those particular writers decided to include. Copying it gives you a page the same size as the incumbents with none of the reasons they earned their positions.

A better target: list every sub-question a reader would still have after reading your draft. If the list is empty, you’re done, whether that took 900 words or 2,600. If the list has five items, you’re not finished regardless of how long the page already is. Coverage is measured in questions answered, not in characters.

There’s a discipline in stopping, too. Pages that keep going past the point of usefulness — the “related considerations” section nobody asked for, the history of the topic nobody needed — dilute the parts that matter and push the good material below the fold. Cutting a strong page down is usually a bigger improvement than adding to it.

What page structure survives over time?

Question-shaped headings that map to real sub-questions, a direct two-or-three-sentence answer immediately under each one, and elaboration after that. This structure survives because it works for the skimmer, the careful reader, and the systems that extract answers — all without any of them needing to hunt.

Write headings as the searcher would phrase the question, not as an internal label. “How much does installation cost?” holds up better than “Pricing considerations”, because it matches how people actually ask and it forces the paragraph underneath to be an answer rather than a discussion. If a heading can’t be turned into a question, it’s usually a sign the section has no clear job.

Put the answer first in every section. The instinct to build up to a conclusion is a habit from essay writing and it costs you readers, because most people won’t wait. Answer, then explain why, then cover the exceptions. Anyone who wanted the short version already has it; anyone who wanted depth keeps reading.

The other thing that survives is genuine scannability — headings that make sense read alone, tables where the data is comparative, short paragraphs where the material is dense. Not decoration. A page covered in bold text and callouts that don’t add information reads as anxious rather than confident, and readers can tell.

What actually needs updating, and what doesn’t?

Update when a fact has changed, a recommendation is no longer the one you’d give, or the SERP has shifted format. Don’t update to change a date, shuffle paragraphs, or add a “2026” to the title. Freshness signals track substantive change, and cosmetic edits mostly waste your time.

That last row catches people out. Links disappear quietly — an editor trims outbound links during a redesign, a host page gets rewritten, a site changes hands. If a page’s decline tracks with a falling referring-domain count rather than with competitor activity, the content isn’t the problem. Our piece on what happens to links over the following years covers the patterns worth watching.

How do you demonstrate first-hand experience concretely?

Include the specifics only someone who did the thing would know — the exact tool version, the number that surprised you, the step that failed, the workaround. Generic competence is easy to write and impossible to distinguish. Specific detail is the only durable moat a page has.

Compare two sentences. “Response rates for outreach vary depending on your approach.” Versus: “In our experience, cold outreach for link placements lands somewhere around one reply per five to eight sites contacted, and the replies cluster on the sites where the pitch named a specific article rather than the site in general.” The second one is hedged, honest, and could only be written by someone who’s sent the emails.

You don’t need proprietary data to do this. You need to write down what you actually observed, including the parts that were inconvenient. Admit what you’re unsure about. Name the trade-off you made and why. Say what didn’t work. Pages that only report successes read as marketing, and readers discount them accordingly — which shows up as short sessions and returns to the SERP, which is the mechanism behind the slow slide in row one of that table.

Screenshots of real workflows, actual numbers from real processes, named failure modes, and honest limits all do the same job. They make the page unclonable by someone who hasn’t done the work.

Why does AI-generated filler decay so fast?

Because it’s built from the average of what already exists, so it has no information that isn’t already ranking somewhere else. It reads competently and offers nothing to choose it for, which means it survives only while it’s novel to the index.

Here’s the opinion, and it’s a mild one: the problem with AI-written filler isn’t that a model wrote it. It’s that nobody added anything. A page assembled from a summary of the top ten results is a compression of the SERP, and a compression is by definition less useful than the sources — so the moment a competitor publishes something with an original observation in it, the compressed page has no argument for its position.

Used differently, the same tools are fine. Drafting structure, tightening prose, generating the list of sub-questions you might have missed — all reasonable. What doesn’t work is skipping the part where a person who knows the subject adds the specifics. That’s the part that holds the ranking, and it’s the part that can’t be automated because it depends on having done something.

The tell is easy to spot once you know it. Read a page and ask what you learned that you couldn’t have guessed. If the answer is nothing, the page is filler regardless of who or what produced it.

What should you check before publishing?

Run the page against a short list that tests coverage, structure, and honesty rather than surface polish. Most of these take under a minute each, and skipping them is why teams end up rewriting the same page twice a year.

• The primary question is answered in the first hundred words, before any setup
• Every H2 is a question a real person would type or ask aloud
• Each section answers in two or three sentences before elaborating
• At least three specifics no competitor could have written without doing the work
• At least one honest limitation, trade-off, or thing that didn’t work
• No section exists purely to add length — cut anything you’d skip as a reader
• Format matches the dominant format on the live SERP for the target query
• Internal links point to the pages a reader would genuinely want next
• Title reflects what the page delivers — no promise the body doesn’t keep
• A review date is set — six months for volatile topics, twelve for stable ones

That last one is the item most often skipped and the one that prevents the most damage. Pages decay because nobody’s looking, not because decay is inevitable. A calendar entry costs nothing.