Brand promotion: this article introduces Guangsuan's SEO measurement guidance and Google SEO service. It is a vendor introduction, not an independent review.
Short answer for a growth owner: do not approve a landing-page rewrite or an outsourcing brief from blended totals. Write down one property, one search type, one date range, one query group, one page set, one country and one device, then work through capture, eligibility, selection and enquiry in that order. Each metric group below answers a separate question, so movement in one line never proves the whole funnel improved.
Fix the comparison before you fix the page
Decide what counts as traffic in this review. Search Console shows how your property surfaced in Google Search and which results were chosen. Analytics tools record what could be observed after arrival on your site. The two systems use different clocks, scopes, consent handling, tag firing and attribution rules, so their totals for the same week can differ without either being broken.
For this decision, record on every export: property, dates, search type, page or page group, query filter, country and device. For session-based acquisition, keep the session-based organic grouping apart from first-visit groupings. When Console shows activity while analytics shows almost none, repair tagging, consent, data streams and filters first. Send only that repaired baseline to a vendor. For method definitions, see Guangsuan's guide to the nine SEO metric groups.
What each group is really asking
1. Arrival versus appearance. Treat appearance and arrival as two lines. Note which landing pages, countries and devices moved, and confirm whether measurement changed before you invoke seasonality, availability or content quality. A change in paid spend can shift the organic share without any change in organic demand.
2. Queries and mean position with filters held still. Note language, country, device, brand versus non-brand intent, and the pages tied to each query cluster. A mean position is a weighted summary for the selected view, not a rank every searcher sees. External keyword counts come from other databases, markets and crawl dates. Sort queries by product, use case and buying question, then open one slipped or promising example and check whether results now favor a product page, tutorial, video or local option and whether your page serves that job.
3. Selection rate only inside similar sets. Define the rate as selections divided by appearances. As a hypothetical teaching illustration with invented figures, not a target: 20 selections from 1,000 appearances is arithmetic of 2%. That number sets no pass mark. Brand, informational and purchase queries behave differently, and new low-position appearances can pull a blended rate down while total selections rise. Compare matched query sets, positions, devices and markets before judging a title. When you revise, state the page answer accurately in the title and describe only real page content in the description.
4. Enquiry with both sides written down. A chat tap, form submit, download, basket action and paid order are separate outcomes. Verify how the event fires, then separate tests, duplicates, spam and sales-accepted enquiries in your own records. State which formula you used, because sessions containing the event divided by sessions is different from events divided by sessions when one session produces several events. Review organic sessions to the page, sessions with the intended event, and business-verified leads together, including mobile forms, contact paths and repeated events. Queries in Console normally cannot be joined one by one to an individual order.
5. Bounce in the GA4 sense. In GA4, bounce is the share of sessions that were not engaged, the inverse of engagement rate. Default engagement includes a stay beyond 10 seconds, a key event, or at least two views, subject to the property setting. A one-page visit can still satisfy the reader, while duplicated view events can inflate engagement. Check implementation first, then compare the search promise with the delivered answer, readability and next steps.
6. Time and depth against the task. GA4 engagement time counts time with the page in focus or the app in the foreground. Different averages use different denominators, and none equals the gap between leaving Search and returning to it. Finding a specification on one page is not worse than finding it after four pages. More time can mean study or confusion. Read time, further pages and task completion together, then add the missing specification, condition, example or next step. Do not add headings or images to hit a quota.
7. Eligibility starting from canonicals you want. Login screens, parameter duplicates and intentionally excluded URLs should never be expected to appear. Keep an explicit list of canonical URLs intended for search and prioritize inside that list. Group indexing-report entries by reason and inspect examples. The inspection tool separates the indexed copy from the live test. Asking for indexing does not compel acceptance. Fix the specific block you found on the checked path and user-agent, such as an accidental noindex, access block, wrong canonical target, unavailable page or missing rendered content. One status line alone does not name the cause, and a correctly consolidated duplicate is not a cleanup task. Reserve canonical for duplicate or near-duplicate material. Log URL, inspection date, indexed versus live result, robots path checked, canonical target checked, change, owner and what stays unverified.
8. Link delivery apart from later movement. Link totals and referring-domain totals are separate dimensions. Discovery coverage, refresh timing and de-duplication differ by crawler, so absence in one report does not prove a live link disappeared. A links report is not a fresh indexing test for each source page. When external help is used, reconcile source URL, target URL, anchor, attribute, availability and observation time against the delivery list. Publication, indexing of the linking page and movement of the target page are separate milestones with separate dates.
9. Vendor scores only inside one scale. One vendor score and another vendor score come from different models and are not interchangeable units. Compare a relevant peer set inside the same tool and period while looking at real links, target pages and search results. Do not optimize the third-party number apart from technical work, content and qualified enquiries.
Use device and country to pick one page
Do not steer by a worldwide device share. Your own Console device and country reports decide. Filter to the same property, search type and window, then view handheld, desktop and tablet separately for top queries and landing pages for that country.
Practical order: compare the whole property first, then the same page group and nearby queries, to see whether a gap comes from a shift in query mix. Small samples swing on one selection, so keep appearances and selections visible together. Note whether you are asking where you were shown or where people acted. Where device work fits in continuing work is described on the Google SEO service page.
What to hand to a vendor
- Console steady, analytics near zero: restore capture and do not rewrite yet.
- Appearances up, blended selection rate down, selections up: inspect query and position mix on matched segments before touching titles.
- Selections up, verified leads flat: inspect intent, market, forms and qualification, then walk the product-page path with real enquiry records.
- Important canonical missing from results: inspect access, signals, content and discovery path, fix the evidenced block, and note the recheck.
Keep missing data marked as missing rather than filling it with another site benchmark. Bring the live address, target country and language stated separately, one complete export with filters, the enquiry pages, what changed and when, who can approve code or copy, and how your team qualifies a contact.
FAQ
Do flat appearances force a rewrite?
No. Seldom being offered for the query points first to eligibility, entry links and whether the page type matches what results favor.
Does a lower blended rate force a title change?
Not alone. Check whether new low-position queries widened the mix, then compare narrow segments.
Does longer dwell prove success?
No. Confirm whether the task was completed and whether a missing condition or next step remains.