Web content optimization: the repeatable pass on pages you have already published

Web content optimization is the repeatable improvement of pages you have already published, so each one serves its single job better. It is a different activity from writing a new page and it returns faster, because the hard part is done: Google has the page, has decided roughly where it belongs, and has given you data about it. The work is three steps on a loop. Pick the pages where a small change moves a real number, usually those ranking between position five and twenty or collecting impressions without clicks. Change one class of thing. Then measure over a fixed window before touching that page again.

This page covers the optimization pass on existing pages. For the techniques to apply when you are writing a page from scratch, see on page SEO techniques, and for deciding which pages should survive at all, see a site content audit.

Step one: pick the pages, in this order

Four buckets, pulled from the Performance report in Google Search Console over the last three months against the previous three. Work them in this sequence, because the order is the order of return per hour.

• Striking distance pages. Average position between five and twenty with meaningful impressions. Google already treats the page as a candidate; the gap is usually coverage, structure or the match between the page and the query people are actually typing.
• High impressions, low click through. The page is being shown and skipped. This is the fastest work on the list because the change is in the title and description rather than in the body.
• Declining pages. Clicks down against the previous period while impressions hold. Something changed around the page: a competitor published, the results page gained a feature, or the facts on your page went stale.
• Pages that convert but get little traffic. Low volume, high value. These are worth more attention than their traffic suggests, and they are where an internal linking change pays immediately.

Pick ten pages, not a hundred. Ten pages optimised properly and measured beats a site wide sweep whose effect nobody can read.

Step two: diagnose the one thing that is wrong

Read the page against the query it is meant to win, then match the symptom to the change. The discipline that makes this work is changing one class of thing at a time, so the measurement afterwards means something.

Symptom	The change	What to watch
Shown often, rarely clicked	Rewrite the title to lead with the query in natural word order and add the thing that separates you: a number, a year, a format word. Rewrite the description as ad copy.	Click through rate for that page, 28 days
Stuck at positions eight to fifteen	Add the sub questions the page does not answer, each as its own heading with a short direct answer underneath.	Average position, and the count of distinct queries the page earns impressions for
Ranks for the wrong queries	Narrow the page to one job. Move the second audience into a second page and link the two.	Query mix for the URL
Answer buried below a warm up	Put the direct answer in the first forty to sixty words.	Click through rate, and whether the page starts appearing in answer boxes
Facts, prices or screenshots are out of date	Update the facts, date them in the text, and remove anything you cannot source.	Clicks, and the trend against the previous period
Nothing on the site links to it	Add links from the three strongest related pages, with descriptive anchor text.	Impressions, and crawl frequency in the server log
Slow, or the layout jumps while loading	Resize and compress the images, remove unused embeds and scripts, set width and height on images.	PageSpeed Insights field data for the URL
Serves the query but produces nothing	Add the next step the reader needs at the point they need it, in the body rather than only at the end.	Conversion rate for the page

Step three: make the page easier to read and easier to quote

Two audiences read a page: a person skimming on a phone, and a machine extracting a short answer from it. Both reward the same structure, which is why this is not a trade off.

• One idea per paragraph, three sentences at most. Long paragraphs lose the skimmer and dilute the extractable answer.
• Headings that state the point, not the topic. "What this costs to run monthly" is useful. "Costs" is not.
• A short direct answer under every heading, before any elaboration. This is the single formatting habit that gets pages quoted.
• Tables for anything with three or more dimensions, lists for sequences, prose for reasoning.
• Plain words. Write the term your market uses, then explain it once. Vocabulary that proves expertise to a colleague loses a buyer.
• Specifics instead of adjectives. A number, a named integration, a stated timeframe or a dated source beats any amount of description.

Step four: the measurement window

This is the part most teams skip, and skipping it means every future argument about content is a matter of opinion.

• Record the date of every change, per URL, in one sheet or as an annotation in your analytics. Without it you are comparing two numbers with no idea what happened between them.
• Wait 28 days before judging, and compare the 28 days after against the 28 days before for that specific URL. Shorter windows pick up weekday and seasonal noise.
• Compare like for like. Same device split, same country, same query set. A change in the mix explains more apparent movement than most content changes do.
• Judge on the right number. A title change is judged on click through rate. A coverage change is judged on position and on the number of distinct queries. A conversion change is judged on conversions, not on traffic.
• Keep a decision log. One line per page: what you changed, the date, the before number, the after number. After two quarters this log tells you which kinds of change work on your site, which is knowledge no general guide can give you.

Build it into a calendar instead of a project

Content goes out of date continuously, so the pass should be scheduled rather than organised as an annual push.

• Monthly: ten pages from the four buckets. Two hours of selection, a day of work, dates logged.
• Quarterly: every page that carries revenue gets a facts check. Prices, screenshots, named integrations, regulations, team names, anything with a year in it.
• Annually: the full inventory, which is the audit rather than the optimization pass, and which ends in keep, improve, merge, redirect or remove.

What to leave alone

• Pages that are working. Top three position, healthy click through, current facts. Refresh the facts and move on.
• URLs. Changing one means a redirect and a recovery period. Change a URL only when the existing one is actively misleading.
• Anything you changed in the last 28 days. Overlapping changes destroy the measurement for both.
• Length for its own sake. Adding a thousand words that answer nothing makes a page worse. Add coverage, not volume.

How is content optimization different from a content audit?

An audit looks at every URL on the site and decides its fate: keep, improve, merge, redirect or remove. Optimization is the improve branch of that decision, run as a monthly loop on a small set of pages. Audit annually, optimise monthly.

Do content scoring tools help?

They are useful as a coverage checklist: they show terms and subtopics the pages currently ranking cover that yours does not, which saves reading ten results by hand. Use the list as prompts for sections a knowledgeable writer would have included anyway, and ignore the target score itself, because writing to a number produces padded copy.

How long before an optimised page moves?

Get the page recrawled, then read the 28 day window. Title and description changes show up in click through rate first, often within two weeks. Coverage and structure changes move position over four to eight weeks as the page is reassessed against more queries.

Should I consolidate several similar pages into one?

When two or more pages serve the same question, yes: merge into the strongest URL and redirect the others. Splitting attention across near duplicates is the most common self inflicted content problem there is.

What about pages with no search demand at all?

Judge them on the job they do rather than on traffic. A pricing page, a careers page or a support article can be essential and never rank for anything. Optimise those for clarity and conversion, and keep them out of the search driven selection buckets.

One more reader to optimise for

A page is now also read by ChatGPT, Gemini, Perplexity, Claude and Google's AI Overviews, which pull a short answer out of it and name a handful of companies in reply to a buyer's question. The structure that gets a page quoted there is the structure that serves a skimming reader: the direct answer first, headings that state the point, specifics with dates, and facts that can be checked. Being quoted in one of those answers is invisible in your analytics, so test it directly: ask your main queries in two or three assistants, leave your own brand name out of the prompt so the answer is not steered, and note which of your pages get cited. AI Knows Us runs that check across the five assistants and reports which sources each answer was built from, and the first scan needs no account.