Which SEO Numbers to Track and Which to Ignore

The argument in most reviews is not about performance. It is about meaning. One person reads "position" as the rank they see in a private browser window, another reads it as the average across every query a page appeared for. Both are looking at different things and neither is wrong.

Google describes an impression as counting when a link to your site appears in the current set of results, with extra rules for widgets that need scrolling or expanding before the item is seen (Search Console help). A click is a click from Google to your site. Click-through rate is clicks divided by impressions. Write those sentences into your dashboard next to the chart and the argument disappears.

Average position needs the most care. Search Console reports the topmost position your result occupied for a query, then averages that across searches in the period, and the value is only shown for Google Search results (Search Console help). That means one page can hold position 2 for its main query and still show an average of 14 because it also appears for a long tail of unrelated queries.

A worked example shows why page level position misleads. Suppose a page takes 1,200 clicks in a four week window, then you rewrite the intro. In the next four weeks it takes 1,400 clicks, but it now ranks for 900 queries instead of 600, most of them deep. If clicks rise, the rewrite probably worked even if average position looks worse, because the average is diluted by a wider set of queries. That is the same trap documented in explanations of the average position metric (SEOTesting). The decision rule: read clicks and conversions first, then use position to explain them, never the reverse.

A review rhythm has two jobs: catch breakage quickly, and stop people from declaring victory or failure too early. Split them.

Every two weeks, look for breakage only. Has a page dropped out of the index, lost most of its impressions, or stopped converting? This is a health check, not a performance review. Google itself warns that crawling and reindexing can take days to weeks after a change, so a page that looks flat two days later is not evidence of anything (Ask Google to recrawl your URLs). If you have edited a page and want Google to see it sooner, request indexing for a small number of URLs through URL Inspection, or resubmit a sitemap when the change is broad (same guide).

Every four weeks, judge the change. Compare like with like: same four week window length, same device filter, same country filter, and the query tab rather than the page average.

A simple log keeps this honest. One row per change, with the date it shipped and the date it becomes judgeable.

change_log.csv
page_url, change_type, shipped_on, judge_on, baseline_clicks_28d, target_note
/blog/shipping-rates/, rewrite, 2025-03-03, 2025-03-31, 340, intent match on "how much is shipping"
/pricing/, add faq, 2025-03-05, 2025-04-02, 512, answer "does price include vat"

Four weeks is not a magic number handed down by anyone. It is a floor that reflects how long recrawling and reevaluation can take, and it stops the common failure where a team ships five changes in a week and then cannot attribute anything. If you ship several changes to one page, log them as one bundle, because you will never separate them afterward.

Two more habits that keep the cadence useful:

• Freeze the definition of each metric for a quarter. If you change how conversions are counted, you have reset the baseline and cannot compare.
• Write the expected outcome before shipping. "Clicks up, position unchanged" is a testable claim. "Improve SEO" is not.

Some metrics are popular because they are easy to produce, not because they help a decision. Google's guidance is blunt about the direction of travel: build for people first, and treat SEO practices as a way to bring searchers additional value, not as the goal itself (Creating helpful, reliable, people-first content). Google's own announcement of its helpful content work made the same point, warning that content created primarily for search engine traffic correlates with content searchers find unsatisfying (Search Central blog). Metrics that reward volume over usefulness push a team in exactly the wrong direction.

Domain level average position. It moves for reasons unrelated to your work. If your site starts ranking for many more long tail queries, the average can get worse while traffic grows, and if you lose rankings for a set of queries you never wanted, the average can improve. Practitioners who test SEO changes see this regularly and recommend reading it only alongside clicks and impressions (SEOTesting).

Keyword difficulty scores. These are third party estimates of how hard a term is to rank for, useful for planning and useless as a success measure. A difficulty score moving does not mean your page moved.

Any single day spike. A one day jump in impressions is usually a crawl event, a news mention or a reporting artifact. Only counts that hold across the review window belong in a decision.

Vanity totals generally. A widely shared argument in the industry is that teams optimize toward domain authority, link counts, impressions and session totals while the thing that matters, revenue from organic traffic, sits unexamined (The MarTech Summit). You do not have to accept every claim in that piece to accept the core point: a number that never changes a decision is costing you attention.

You may add a fifth number, but only if you define it yourself and keep it stable. The test is whether the measure would change what you do. "Bounce rate" usually fails that test because it means different things in different tools and depends on how events fire. A better measure is specific and tied to the page's job: the share of organic visitors who reach the pricing section, or who submit the contact form, or who watch past the first minute of a video.

Pick one, write the exact event definition, and record the tool that reports it. Then compare it only against itself, over the same window length, filtered to organic search and to a single URL.

The reason to be strict is that engagement is where definitions drift fastest. If your analytics tool changes how it counts an interaction, your baseline disappears. Google's page experience guidance is about the real experience of a visitor, not about a single vendor's score (Creating helpful, reliable, people-first content), so the measure you choose should describe something a person actually did.

A tracking setup earns its keep when it produces decisions. The pattern that works:

1. A page's clicks fall while impressions hold. The page is still being seen but is being chosen less. Look at the title and snippet, then at whether the query mix changed.
2. Impressions fall while clicks hold. You may have lost visibility on queries that never converted. Check conversions before reacting.
3. Both hold, conversions fall. The problem is on the page or after the click, not in search.
4. Everything rises. Log what you changed, but do not assume you caused it. Check whether a seasonal pattern or a wider query set explains it.

You can see the same logic in Google's guidance on how search works: crawling, indexing and serving are separate stages, and not every page makes it through each one (How Google Search works). A metric that does not tell you which stage failed is not doing its job. When a page loses impressions, that is a discovery or indexing question. When it keeps impressions but loses clicks, that is a relevance and presentation question. When it keeps clicks but loses conversions, that is a page question.

Search intent belongs in this loop. The four common intent types (informational, commercial, navigational, transactional) exist to help you check whether your page matches what the searcher wanted (Semrush), and a mismatch shows up as impressions without clicks. Keyword research tells you what people type; intent tells you why, and the second is what your page has to satisfy (Backlinko).

A workable setup for a small team:

• One dashboard, five tiles at most, each captioned with its tool and filter.
• A two week health check that only asks "is anything broken?"
• A four week judgement that only asks "did the logged change do what we expected?"
• A change log with a judge date on every row.
• A quarterly review of the definitions themselves, so drift gets caught deliberately rather than during an argument.

The rest is discipline. Most teams already have the data. What they lack is a rule for when a number is allowed to change a decision, and that rule is the whole system.

If your review keeps surfacing pages that need rewriting, new comparison content, or a rebuild of the template that holds them, that work is where the numbers lead. Quanta takes a described site, app or document and has a specialist build it with you: https://quantaintelligence.ai/.