Skip to content

MarketEson · Field Notes

SEO Metrics First: Which Fix Decides the Next Change

For an in-house B2B SaaS team before a revamp: what nine SEO numbers each mean, how to split them by device and country, and which page to fix first.

Brand promotion: this article introduces Guangsuan's Google SEO guides and services. It is not an independent review.

Your team plans to rebuild templates next sprint. Search Console shows movement, analytics looks flat, and everyone points to a different number. Before you rewrite anything, sort the nine readings into separate jobs. Each one answers a narrow question about being found, being chosen, loading, reading, or turning into a real enquiry.

Start by freezing how you look

Pull all numbers from one property, one search type, and one finished date range. Leave the last incomplete day out. Note language, nation, device, whether the query mentions your brand, and which landing page received the visit. Group queries by product line, use case, and buying question.

Search Console clicks and GA4 sessions come from separate systems and will not always match. One records appearances and picks inside Google Search. The other records what could be observed after arrival. Tag firing, consent choices, coverage, time zones, and canonical credit can pull them apart, so do not force them to agree.

If Console is steady while analytics is almost empty, repair tracking first. Check tags, consent handling, data streams, and report filters before you judge demand.

1. Arrival: clicks are not visits

Treat discovery and arrival as two layers. Discovery asks whether searches surfaced you and whether people picked you. Arrival asks whether they reached the site in a measurable way. Find which landing pages, nations, and devices moved, then confirm recording before you blame season, stock, or copy. A smaller organic share can also mean paid spend grew.

2. Queries and positions need steady settings

Average position is a weighted summary for the dates, pages, nations, and devices you selected. It is not the slot every searcher saw. Outside tallies of ranking keywords rest on their own lists, markets, and collection dates, so two vendors can disagree without either being broken. Open a falling or promising query, see whether live results favour a product page, guide, video, or local result, and ask whether your page serves that same task.

3. Pick rate: keep comparisons narrow

CTR is clicks divided by impressions, multiplied by 100. As a teaching illustration only with made-up numbers, 20 picks from 1,000 shows is 2%. That arithmetic is not a target for every intent.

Adding many low-position impressions can pull the overall CTR down even when total clicks grow. Compare only queries with similar intent, similar position bands, the same device, and the same market. State plainly what the page answers and describe only what is really on the page. Do not insert an unverified certificate or result claim to win picks.

4. Enquiries: name the outcome before you divide

A chat tap, finished form, download, basket move, and paid order are separate outcomes. Test that the live event fires, then separate internal tests, repeats, spam, and genuine business enquiries in your own log. Name which ratio you use, because sessions containing the event divided by sessions is different from events divided by sessions when one visit holds several events. As a rule, queries cannot be linked one by one to a single order. Read landing-page visits, visits with the wanted event, and business-confirmed contacts side by side, and check the mobile form and reply path.

5. Bounce and engagement in GA4 terms

In GA4, bounce share is the inverse of engagement share, not the old one-page view idea. The default engaged case can involve staying past a short time limit, completing the chosen event, or viewing at least two screens. Confirm how your property is set, because defaults can be changed. A single-page stay can fully satisfy the reader, while duplicated view hits from a setup fault can inflate engagement. Verify recording first, then compare what the snippet promised with what the page delivers and what useful step comes next.

6. Time and depth only make sense with the task

Focus time captures time with the page in front of the reader. Four pages to locate a spec is not better than one page that answers it. A long stay can mean careful study or confusion. Read time, later pages, and whether the job was done together. Fill the gap that blocks a decision: specs with units, applicable conditions, worked checks, or relevant next pages, rather than adding headings by quota.

7. Eligibility: begin from pages you want found

Trying to index every discovered URL is not a sensible target. Login areas, parameter copies, and pages held out by choice remain in crawl space. Keep a short list of canonical pages you actually want in search and work through the important ones first. Group the indexing report by reason and open samples. Asking Google to index a URL starts a review and does not guarantee inclusion. Repair proven blocks such as unintended noindex, blocking, wrong canonical signals, missing pages, or matter that does not render, then log date, filters, edits, owner, and what still needs rechecking. A notice that a page was crawled but is currently not indexed does not by itself prove the text was too short.

8. Links: keep delivery apart from later movement

Total link counts and referring-site counts describe different things. A missing row in one crawler report does not prove a live link is gone. Crawlers vary in coverage, refresh rhythm, and how repeats are handled, and a links report is not a live check that 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, getting indexed, and later ranking better are three separate stages.

9. Vendor scores: stay inside one scale

Scores from Moz, Ahrefs and other vendors use different models and cannot be swapped. They are vendor predictions, not grades handed out by Google. Compare a relevant peer set inside the same tool and same period, while still reading real pages, real links, and search outcomes. Do not tune only for the third-party number while neglecting page matter, technical access, and enquiry handling.

Split by device and nation before you choose

Results shift with time, place, device, and searcher history. Apply the device filter, then the nation filter, on the same property and dates. Look at top pages and queries for handheld versus desktop inside the same country. Note whether handheld queries seek fast specs while desktop queries seek comparison, and whether the title truncates or the spec table is hard to tap on a phone. Report shows and picks separately, because a row with very few shows can swing on one pick.

Which repair comes first

Use one finished range and the same filters throughout:

  • Console steady, analytics near empty: inspect tags, consent, and filters. Restore reliable capture before judging need.
  • Shows up, blended pick rate down, picks up: inspect whether new low-position queries widened the mix. Compare close segments before editing titles.
  • Picks up, confirmed contacts flat: inspect intent, market, offer detail, form ease, and qualification rules. Walk the product path.
  • Key canonical outside the index: inspect access, signals, matter, and discovery path. Fix what evidence supports and note rechecks.

Write down dates, filters, what changed, who made it, and what remains open. Keep absent data marked absent rather than filling it with another benchmark.

Where Guangsuan fits

Once targets are set, Guangsuan describes its Google SEO service as technical work, page content, and outside resources that support each other. For method definitions, see Guangsuan's 9 SEO metrics guide.

Before you ask any provider, bring the live address, target nation and language stated separately, one finished Console export with filters, and how you define a qualified contact.

FAQ

Should flat shows trigger a rewrite?

No. Flat shows mean the page is seldom offered for the query. Confirm eligibility, entry links, and whether the page type matches what results favour first.

Should a falling blended rate trigger a title change?

Not alone. Check whether new low-position queries changed the mix, then compare close segments.

Does longer time prove success?

No. Add the missing condition or next step and check whether the job was done.

Turn this read into ranking content.

Get a free, no-credit-card audit of your site — keyword gaps, competitor intel, and a 90-day content roadmap in 11 minutes.

Get Your Free SEO Audit
2,140+
Paying teams
4.2B
Keywords tracked
63%
Faster time-to-rank