Keyword Ranking Analysis: A Step-by-Step Diagnostic Workflow for Explaining Rank Drops
A practical, repeatable process for turning rank data into a diagnosis - how to isolate whether a position change came from a technical regression, cannibalisation, intent shift, or lost click-through.

Why Ranking Analysis Is a Diagnostic Job, Not a Forecasting One
Most rank content online is about prediction. This is the opposite: a retrospective workflow for explaining a movement that has already happened. Take a Cape Town building-materials retailer whose category queries slipped from positions 4-6 to 11-14 in a month after a template refresh. The question is never "where will this keyword go next?" but "what changed, and was it on our side or Google's?" Keyword ranking analysis is forensic work: gather evidence, eliminate suspects, arrive at a cause you can act on.
Diagnosis needs different inputs than forecasting. It leans on a change log, segmented data, and access to everything you edited in the suspect window. If your team cannot answer "what did we ship in the last 30 days?", no rank tooling will help.
Diagnose a known drop in days; do not let it sit in a spreadsheet for a quarter. Position losses at the top of page one compound into click losses quickly.
What You Need in Place Before You Analyse Anything
You cannot reconstruct a baseline you never saved. Search Console retains 16 months of data but only compares against a rolling period, so export your query-level baseline on the first of every month - CSV, clicks and impressions, segmented by country - and store it versioned, not in a shared inbox.
Second, keep a dated change log covering template deploys, migrations, robots.txt and canonical edits, internal linking updates, and content rewrites. Google's own documentation on how Search works makes clear that indexing and serving pass through many stages, so a same-week edit is always a candidate cause.
Why Search Console Average Position Misleads Most First-Time Analysts
"Average position" is not a rank. It is the impression-weighted mean position of the best result Google surfaced per query, blended across devices and locations. A keyword ranking 3 in Johannesburg desktop and 14 in Durban mobile reads as a single number that looks like a rank. Analysts routinely blame "an algorithm hit" for a number that shifted because mobile impressions grew.
Treat average position as a directional signal only, and segment it before you believe it. Country filters are non-negotiable for South African sites, because Google serves localised results and a global blend mixes unrelated SERPs into one figure.
Step 1: De-Blend the Data by Country and Device
Open the Performance report filtered to South Africa and split desktop against mobile. Watch for desktop holding position while mobile collapses - a template, viewport, or Core Web Vitals symptom, not an authority problem. Only when segments look consistent do you escalate.
| Segment | Typical symptom | Likely cause |
|---|---|---|
| Mobile only drops | Desktop stable, mobile slides 5+ places | Template, layout, speed regression |
| All segments drop equally | Uniform loss across devices | Cannibalisation, content rewrite, SERP feature |
| Impressions up, clicks flat | Average position improves but CTR falls | SERP feature absorbing the click |
Step 2: Build a Rolling Position Band Instead of Chasing Daily Snapshots
Daily rank-tracker spikes are noise. Compare 7-day or 28-day bands against the previous equivalent band and record the delta. Use tools for what they do well - capturing AI Overview presence, local pack occupation, and featured snippets - rather than treating an absolute position as gospel.
Step 3: Separate a Ranking Change from a Click-Through Change
A keyword can hold position and still lose half its clicks. Before rewriting anything, overlay impressions against clicks: stable impressions plus falling clicks points to a title, meta, or SERP-feature problem, not a ranking problem. Straider's platform tracks AI Overview and ChatGPT visibility alongside classic rank data, which is where much of this click loss now originates.


