Which pages on my website need SEO improvement?

Most SEO work happens on whichever page a stakeholder mentioned most recently. That is not a prioritisation method, and it explains a great deal of why content programmes underdeliver.

Almost every site has a handful of pages where two hours of work produces more than a quarter of new publishing. Finding them is a spreadsheet exercise rather than a judgement call.

The one export you need

Open Search Console, go to Performance, set the date range to the last three months, and export queries and pages with impressions, clicks, CTR and average position. Then do it again for the same three months a year earlier.

Those two exports contain every bucket below. You do not need a paid tool for this, and the exercise takes about an hour.

Bucket 1: striking distance

Filter: average position 8–20, impressions above 100 a month.

These pages have proven relevance — Google already considers them a credible answer — and real demand, but sit far enough down that they receive almost nothing. They are the closest thing to guaranteed return in SEO, because the page exists and the gap is depth rather than existence.

The work: list the H2s of the top three ranking results and add the subtopics you are missing. Move the direct answer to the first two sentences of each section. Refresh anything factual that has aged. Route internal links from your strongest related pages with descriptive anchor text.

Expected return: high. Moving a page from 12 to 6 can double its traffic, because the click-through curve is steepest exactly there.

Bucket 2: high impressions, low CTR

Filter: impressions in your top decile, CTR materially below typical rates for the position.

These pages already rank. Thousands of people see them and choose someone else. Nothing is wrong with the SEO; what is failing is the title and description.

The work: rewrite the title using the searcher’s phrasing rather than your internal vocabulary. Be specific rather than clever. Give a visible reason this result answers the question. Write the description as a promise rather than a summary.

Expected return: highest return per hour of anything in SEO, and the fastest — CTR changes as soon as the new snippet is served, typically within one to three weeks.

Bucket 3: decaying pages

Filter: compare the two exports. Pages with materially fewer clicks than the same period last year, where position has slipped.

Decay is normal and it is reversible. It usually means a competitor published something more current, or your information aged, or the query’s intent shifted. Recovering a decayed page is considerably cheaper than building a new one, because the links, the age and the internal linking already exist.

The work: read what now outranks you and identify what changed. Usually it is currency — dates, figures, examples, or advice that has aged badly. Substantive updates move pages; changing the date alone does not and is easy to detect.

Expected return: good, and often faster than striking-distance work because the page has history.

Bucket 4: cannibalising pages

Filter: group the export by query. Any query where several of your URLs appear is a candidate.

When multiple pages target one query, relevance and link signals split between them and none accumulates enough to rank well. Google also has to guess which page you meant, and it frequently picks the weakest.

The work: pick the strongest page, fold the genuinely useful content from the others into it, and 301 the rest. Then check internal links: they should now all point at the survivor.

Expected return: often dramatic and always counterintuitive. Sites gain positions from deletion, which is why this bucket is worth checking even when nothing looks obviously wrong.

Bucket 5: dead weight

Filter: pages with negligible impressions over twelve months, no meaningful backlinks, and no strategic purpose.

Thin pages are not neutral. They compete with your good pages, dilute site-level quality signals, and consume crawl attention that should go somewhere useful. A large crawled-not-indexed bucket in Search Console is Google telling you this directly.

The work: merge before deleting. Fold anything useful into a stronger related page and redirect. Delete outright only when there is genuinely nothing worth keeping and nothing links to it.

Expected return: indirect but real. This is site-level hygiene rather than page-level gain, and it makes everything else work better.

The order to work them

1. Bucket 2 — title and description rewrites. Fastest payback, lowest effort, works on demand you have already earned.
2. Bucket 4 — consolidation. Often unlocks pages that were competing with themselves for months.
3. Bucket 1 — striking distance. The largest sustained gain, and the work that compounds.
4. Bucket 3 — decay recovery. Cheap wins on pages with existing history.
5. Bucket 5 — dead weight. Hygiene, best done as a batch once or twice a year.

Ten to twenty pages a quarter is a realistic scope for most teams. Fewer and the effect will not be visible in the numbers; more and the work becomes too shallow to move anything.

The sixth bucket

There is now a bucket that no Search Console export can produce, because nothing in it registers an impression.

Pages that answer a question assistants get asked, and are never cited for it. These do not appear in any of the five filters above — they may rank perfectly well — but when a buyer asks ChatGPT the same question, a competitor is named instead.

Finding them requires running your category’s real questions against the assistants and comparing who is named against which of your pages should have been the answer. The usual causes are mechanical: the answer only exists after JavaScript runs, the page answers six questions rather than one, or third-party sources describe competitors consistently and you not at all.

It is worth checking before you spend a quarter on the other five buckets, because a page that is perfect on all conventional measures and unreadable by GPTBot is a different problem with a much cheaper fix. Our guide to why content does not appear in AI answers covers the diagnosis.

A worked triage

A 400-page site with flat traffic. The export produces:

• 28 pages in striking distance with meaningful impressions.
• 11 pages in the top impression decile with CTR well below par for their position.
• 17 pages down more than 40% year over year.
• 4 query groups where three or more URLs compete.
• 210 pages with fewer than 20 impressions in twelve months.

The last number is the one that stops the room. More than half the site produces nothing measurable, which is simultaneously the least urgent finding and the most informative one — it explains the crawled-not-indexed bucket, and it explains why the good pages are working harder than they should have to.

The quarter that follows: rewrite 11 titles in week one. Consolidate the 4 cannibalising groups in weeks two and three, which removes about 9 URLs. Deepen 15 of the 28 striking-distance pages across weeks four to ten. Refresh the 8 highest-value decayed pages in weeks eleven and twelve. Schedule the 210-page cleanup as a separate batch project, merging rather than deleting where anything is salvageable.

Zero new pages. That is usually the correct answer for a site with 400 pages and flat traffic, and it is almost never the answer anyone expects to hear.

How to know the triage worked

Set the check at ninety days and measure three things against the export you started from: the number of queries in the top ten, total clicks from non-branded queries, and site-wide CTR. All three should move if the triage was right.

If clicks rose but top-ten query count did not, the gain came from bucket two and the ranking work has not landed yet — normal at ninety days. If top-ten count rose but clicks did not, check AI Overview presence on those queries, because the positions may be real and the clicks may be going into the answer above them.

What not to prioritise by

Not by how much you dislike the page. Internal embarrassment correlates poorly with traffic opportunity. Pages that make the team wince are frequently ranking fine.

Not by how recently someone complained about it. The loudest stakeholder and the largest opportunity are unrelated variables.

Not by strategic importance alone. A page can matter commercially and have no realistic path to ranking. Those pages need a different channel, not more SEO effort.

Not by how easy the fix is. Easy and valuable overlap only sometimes, and organising work around effort rather than return is how a quarter fills with completed tasks and unchanged numbers.

Sources

• Google Search Console — the only first-party source of query, impression, click and position data.
• Vercel and MERJ — analysis of 500M+ GPTBot fetches finding zero JavaScript execution; same behaviour for ClaudeBot and PerplexityBot.
• Ahrefs — CTR impact of AI Overviews; approximately 58% decline for top-ranking pages (February 2026).

Frequently asked questions