Skip to content
Signal network online · 14 chains Request Demo
SEO与数字营销

What Each of 9 SEO Metrics Counts Before You Edit Pages

PBy PFN DaiPFN Dai

For data analysts briefing management: what each of the 9 SEO metric groups counts, how to group high-impression low-click cases, and when to edit the page versus repair tracking and attribution.

Brand promotion: this note explains Guangsuan's SEO metric definitions to help an analytics reader prepare a management update.

Short answer with limits: lock one comparable view first — same property, same search type, one finished period, same query set, same address, same country and same device — then verify recording, then confirm eligibility, then read presentation and enquiry handling. Each metric group counts a separate stage, so a rise in one total proves only that stage.

For method detail, see Guangsuan's guide to 9 SEO metrics. For full-site scope across technical work, content and links, see the Guangsuan Google SEO service.

Freeze the comparison before reading any total

An analyst often brings one export to a review. That export is comparable only when every line shares property, search type and finished dates. Leave the unfinished latest day out. Write language, country, device, brand against non-brand intent, and landing address on each export. Keep brand queries apart from non-brand queries and read handset apart from desktop. Leave absent rows marked as absent rather than filling them with another source.

Once scope is fixed, find which addresses, countries and devices moved. If Console looks normal while site analytics records almost nothing, restore capture before judging demand. Check tags, permission handling, streams and filters. Then name the moved pages.

How the nine groups divide the work

Seen in search versus observed after arrival

Search Console describes shows and selections inside Google Search. GA4 records observable activity after arrival on the site. A Console click is therefore not an analytics session. Zone offsets, coverage, consent, tag firing and canonical assignment can pull Search Console and analytics totals apart. Do not force the two totals to agree.

Query cover and mean position

Mean position is a show-weighted summary for the selected view, not one live rank seen by every searcher. Outside keyword counts rest on their own lists, markets and collection dates. Group queries by product family, use and buying question. Record language, country, device, brand split and landing address.

For a sample query, open the live result and note which job it serves. Where results favour a product, guide, video or local answer, ask whether your address serves that same job. When the query asks for model, delivery or compatibility but the ranked address only explains ideas, the wrong page type is ranked and needs product or comparison material before wording changes.

Selection share inside narrow groups

Read clicks divided by shows only inside matched intent, position band, device, market and page. For example, 12 clicks from 600 impressions equals 2% as arithmetic. That illustration shows the maths and is not a goal for every intent. Brand, broad informational and buying queries have different settings. Added low-position shows can pull a blended share down while total clicks still rise.

State what the page answers and describe only genuine page material. Do not insert an unverified certificate, test statement or performance figure to win selections.

Contacts and ratios with both sides named

A chat tap, finished form, file download, basket move and paid order are separate outcomes. Test that the live event fires, then keep trials, repeats, spam and business-confirmed contacts on separate lines. Name the ratio you use, because sessions containing the event divided by sessions differs from events divided by sessions when one visit holds several events. Review landing-page sessions, sessions with the intended event, and business-verified leads together. Check handset forms, contact routes and repeated events. Query lines as a rule cannot be tied one by one to a single order.

Bounce read under GA4 rules

In GA4, bounce is the share of sessions without engagement, the reverse of engagement share. It is not the old one-page idea. Usual engagement conditions involve duration beyond a short threshold, a chosen event, or at least two views, but confirm the actual property setting. One answered question on a single page can still satisfy the reader, while repeated hits from a setup fault can inflate engagement. Verify capture first, then compare the snippet promise with the delivered answer and next step.

Focus time and depth judged against the task

GA4 focus time records time with the page in front of the reader. Longer time can signal careful study or confusion. Finding a tolerance on one page is not worse than needing four pages. Read time, later pages and task completion together, then supply the missing specification with units, applicable condition, worked check or next page.

Eligibility from a list you intend to index

Keep an explicit list of canonical addresses for indexing and handle important ones first. Sign-in areas, parameter copies and material held out by choice stay outside that list. Group the indexing report by reason and open samples. The inspection tool separates the stored copy from the live test. Asking for indexing starts a review and does not force entry. A crawled but currently not indexed note alone does not prove thin copy. Use canonical only for duplicate or highly similar material, after checking the exact robots rule path and user-agent where access is in doubt.

Outside links: delivery kept apart from later movement

Total links and referring-site totals are different counts. Coverage, refresh timing and de-duplication vary by crawler, so one absent row does not prove a live link vanished. A links report alone does not show whether each source page is currently indexed. Where outside help is used, match source address, target address, anchor, attribute, availability and check time against the delivery list. Going live, entering the index and later ranking better remain three separate stages.

Vendor scores kept inside one scale

Scores from different vendors use different models and cannot be swapped. They are vendor predictions, not grades assigned by Google. Compare a relevant peer set inside the same tool and time window while still reading real pages, real links and search outcomes.

Shows up, picks flat: grouping order

Flat shows mean the page is seldom offered for the query. Shows up with few added picks means the page is offered more often but picked rarely, or offered for new queries in lower positions.

  1. On one finished export, open Pages to locate the moved address, then open its query list with shows, picks and position together.
  2. Tag new, continuing and off-task queries separately rather than sorting by rate alone.
  3. For a sample query, read the title and summary actually shown for that query, page, country and device, and ask whether your page serves that same job.
  4. Adjust the title or snippet-relevant passage only when the same narrow slice shows many shows but few picks and the shown title misstates the page answer.

A blended share fall can follow a wider mix, not weaker titles. Check whether new low-position queries widened the mix, then compare narrow slices before rewriting.

Repair tracking or edit the page first

Work in this order: capture, then eligibility, then presentation, then contact path, then outside resources.

SituationVerify firstDo next
Console normal, analytics near emptyTags, consent, streams, filtersRestore reliable capture before judging demand
Shows up, blended share down, picks upQuery and position mixCompare narrow slices before rewriting titles
Picks up, confirmed contacts flatIntent, market, forms, qualificationWalk the product path and read actual enquiry records
Key canonical outside the indexAccess, signals, material, discovery pathFix the proven block and log date, filters, owner and recheck

For indexing, log exact address, inspection date, robots path with user-agent, canonical target, stored copy against live test, what changed, who owns it and what stays open. Fix only the block proven on that path, such as accidental noindex, blocking, wrong canonical target, unavailable page or missing rendered material.

For people-first checks, see Google guidance on creating helpful, reliable, people-first content.

Bring facts another person can repeat to the next review: site address, target country and language stated separately, affected addresses, one finished Console export with filters, what changed and when, who can approve code or copy, and how your team qualifies an enquiry. This paragraph is hypothetical teaching material, not a client case.

FAQ

Do flat shows by themselves call for new copy? No. Flat shows mean the page is seldom offered for the query. Confirm eligibility, entry links and whether the page type fits what results favour for that task first.

Does a lower blended share by itself call for a new title? Not alone. Check whether new low-position queries widened the mix, then compare narrow slices.

Does longer reading time prove the page worked? No. Longer time can mean useful study or confusion. Add the missing condition or next step and check whether the task was completed.

Allocate capital with intent.

PFN Dai fuses transformer-based on-chain analytics with proprietary DeFi signal models — engineered for funds that don't guess yield.

Request a Platform DemoView Audited Performance