What 9 SEO Metrics Actually Mean Before You Change a Page
For analysts briefing management: what each of the 9 SEO metric groups counts, which page pile to check for CTR, ranking and indexing issues, and when to bring a defined export to Guangsuan instead of only editing copy.
Brand promotion: this article introduces Guangsuan's SEO measurement guidance and services.
You have one report, three piles of pages, and a management meeting that wants a decision. This briefing note is written for that moment. It keeps the Friday-report question narrow: which pile moves first, and what check proves it.
Keep one scope frozen for every comparison in the note below: same property, same search type, one finished date range, one query group, one page address, one country, one device split. An unfinished yesterday next to a finished month will mislead the room even when every formula is correct.
Briefing pack to bring into the room
Write the audience and the candidates on one page. Audience here is management. Candidates are product and enquiry pages that carry specifications and contact entries, supporting guides that answer a buying task, and canonical addresses that may carry an access or eligibility block. State target country and language as two separate lines. Keep brand queries apart from non-brand queries. Read phone apart from desktop. Leave an empty cell marked as missing instead of filling it with another tool benchmark.
Nine groups, what each fraction counts, which window to lock
The table below is the denominator check management needs before any rewrite. Read it left to right: what is counted on top, what is counted below, and which window makes the number comparable.
| Group | Numerator and denominator to record | Window and split to lock |
|---|---|---|
| 1. Arrival from Search versus observed on site | Search Console counts shows and selections inside Search. GA4 counts observable sessions after arrival. No shared denominator links them. | Same property, same search type, same finished range. Note zone offset, reporting scope, consent choice, tag firing and canonical assignment that can pull the two totals apart. |
| 2. Query cover and mean position | Mean position is a show-weighted summary for the view selected. Outside keyword totals depend on that vendor list and collection date. | Same dates, pages, country, device, brand split and landing page. Group by product family, use and buying question. |
| 3. Selection share | Selections divided by shows, times 100. As a hypothetical teaching illustration only, 20 selections from 1,000 shows is arithmetic for 2 percent, not a goal for every intent. | Compare only inside close intent, position band, device, market and page. Added low-position shows can drag a blended share down while total selections rise. |
| 4. Contacts and ratio | A chat tap, finished form, file download, basket move and paid order are separate outcomes. Sessions holding the chosen event divided by sessions differs from events divided by sessions when one visit holds several events. | Same landing pages and range. Test the live event, then separate trials, repeats, spam and business-confirmed contacts in your own log. Query lines as a rule cannot be tied one by one to a single order. |
| 5. Bounce under GA4 rules | Sessions without engagement divided by sessions. Reverse of engagement share for the same reporting slice. | Confirm the actual property setting. Usual engagement conditions involve duration past a short threshold, a chosen event, or at least two views. One answered question on a single page can still satisfy the reader. |
| 6. Focus time and depth | Focus time records time with the page in front or the app ahead. Different averages use different bases, so read the label before comparing. | Same page task and device. Longer time can signal careful study or confusion. Read time with later pages and task completion together. |
| 7. Eligibility to appear | Count against an explicit list of canonical addresses scheduled for indexing, with important entries ranked first. Login pages, parameter copies and material kept out of Search stay outside that list. | Group indexing report entries by reason and open samples. Inspection separates the stored copy from the live test. Asking for indexing starts a process and does not force entry. |
| 8. Outside links | Total links and referring sites are different counts. Coverage, refresh timing and de-duplication differ across crawlers. | Same observation date. Match source address, target address, anchor, attribute, availability and check time against the delivery list. Going live, becoming indexed and later movement are separate stages. |
| 9. Vendor scores | Scores from different vendors come from different models and cannot be swapped with each other. | Compare close peers inside the same tool and same time window while also reading actual pages and outcomes. |
Selection share slips: which pages to open first
Input: one finished export with filters, the address that moved, its query list with shows, selections and position together. Operation: keep intent, band, device and market fixed, then read the title and summary actually shown for that query, page, country and device. Decision rule: edit the title or the snippet-relevant passage only when the same narrow segment shows many shows but few selections and the title misstates the page answer. State what the page answers and describe only genuine page material. Do not add an unverified certificate, test claim or performance figure to win selections. If the query asks for model, delivery or compatibility but the ranking address only explains ideas, the wrong page type is ranking and needs product or comparison material before wording changes.
Ranking questions: which pages to open first
Input: representative queries that slipped or show an opening, with landing pages attached. Operation: open the live result and note which page type is favoured there, then ask whether your address performs that same job. Decision rule: when results favour a product, guide, video or local answer and your address does another job, change page type or supporting material first. Product and guide piles come before title tweaks. Do not copy a markup pattern across pages only because it appears in source.
Eligibility questions: which pages to open first
Input: exact address, inspection date, robots rule path checked with user-agent, canonical target checked. Operation: group the indexing report by reason, open representative addresses, compare stored copy against live test. Decision rule: repair only the block proven on the path you checked, such as accidental noindex, blocking, wrong canonical target, unavailable page or absent rendered material, then log a recheck with date, filters, owner and what stays open. A crawled but currently not indexed note alone does not prove thin copy. A correctly merged duplicate is not by itself a removal job. Use canonical only for duplicate or highly similar material.
Global device figure apart from your own split
For September 2026 worldwide, its Desktop versus Mobile report shows Mobile 60.04 percent and Desktop 39.96 percent. That is a share of web traffic by device, not a statement about your sector, country or buyer. It varies widely by country and it measures general web use, while search conduct and conversion differ by sector. The only figure that should drive page work is your own Search Console device and country report on identical property and dates. Inspect top queries and landing pages for handheld and desktop separately and report shows and selections separately. Where handheld share is small, aim for a competent handheld experience rather than moving the whole plan away from desktop.
When copy edits are not enough
When the block touches templates, canonical targets, a robots rule for an exact path and user-agent, consent breaking capture, or section-wide eligibility, address-level wording will not resolve it. That is the point to discuss technical repair, content supply and outside resources together. For method detail see Guangsuan 9 SEO metrics guide. For whole-site scope across technical work, content and links see the Guangsuan Google SEO service. Bring 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 a contact. For people-first checks see Google guidance on creating helpful, reliable, people-first content.
FAQ
Do flat shows require a rewrite? 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 falling blended share force a title change? Not alone. Check whether new low-position queries widened the mix, then compare narrow segments before editing.
Does longer 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.
Get the weekly brief.
One long-read, every Friday. The dashboards other analysts won't show you, the post-mortems the vendors hope you skip.