Cliff, Slide, or Step: Reading the Shape of a Drop

A keyword that fell from 4 to 19 tells you almost nothing. The same fall spread over one check, over six weeks, or in two flat plateaus tells you quite a lot — because each shape is compatible with a different set of causes and rules the others out.

Your rank history already contains the shape. Most people read the endpoints and skip it.

Cliff

One check to the next, large move, and it stays there. Nothing gradual about it.

A cliff is caused by an event, and events have timestamps. The candidates are narrow:

Something broke on the page or the site — a noindex shipped, a redirect went in, a template change, a robots rule.
A ranking system update rolled out.
The result set itself changed shape: a new feature block appeared, or the number of organic slots on the page changed.
Your tracker’s measurement changed — different location, different device, different provider.

The diagnostic move is to fix the date and ask what happened on it. If several unrelated keywords cliffed on the same check, the cause is not page-specific, which points at a site-wide change or something outside your control. If one keyword cliffed alone, look at that URL first.

A cliff on a single keyword with the URL intact and indexed is often not your page at all — a competitor arriving, or a feature block pushing everything down. What Position One Means on a Crowded SERP covers why the slot you’re counting can move without you moving.

Slide

Small losses accumulating over weeks. No single check is remarkable; the trend line is.

A slide is not an event, so looking for one wastes time. The mechanisms that produce slides are gradual by nature:

Content ageing against fresher competitors.
Competitors improving incrementally.
Slow loss of relevance as the query’s meaning drifts.
A gradual change in what the result set rewards.

The measurement trap here is that a slide is the shape most easily faked by noise. Daily positions move on their own, and a run of unlucky checks looks like a trend when it isn’t. Why Google Rankings Change Daily and Baselines Before You Read a Position are the two things to read before believing a slide exists. The test: compare a weekly average now against a weekly average two months ago, rather than comparing single checks.

If the slide is real, it’s a content and competitive problem, and it will not be fixed by anything you can deploy on a Tuesday.

Step

Flat, then a drop, then flat again at the new level. Sometimes several steps.

Steps are the most informative shape and the least discussed. A step means something changed state and stayed changed. Common producers:

A page was consolidated, redirected, or replaced, and the new state settled.
Two of your own pages started competing for the query, and which one gets shown changed — Keyword Cannibalisation Looks Like Flapping covers the version where it doesn’t settle.
The result set gained a persistent new feature or a persistent new competitor.
A rollout finished. Some updates arrive over days, which reads as a step rather than a cliff.

Multiple steps in the same direction usually mean a sequence of related changes — which is a good reason to keep a record of what you shipped and when. That’s the subject of Rank Movement and Your Deploy Log.

The shape you should hope for

A fourth shape exists and it’s the one people misread most: flapping. A keyword oscillating between two positions on alternating checks has not dropped at all. It’s sitting near a boundary — between two results with similar relevance, or between two of your own pages, or across a personalisation or location boundary in the measurement itself.

Flapping is not a drop and shouldn’t be alerted on as one. If your alerts are firing on it, the threshold is wrong: Rank Alerts That Do Not Cry Wolf.

Shape at the set level

The same reading applies across a group of keywords, and it’s more reliable because the sample is bigger.

Many keywords, same check, same direction → something site-wide or external. Look at your own release history and at whether the whole result set moved.
Many keywords, same page → that page. See Sitewide Drops Versus Page-Level Drops.
One keyword, everything else flat → that query. Competitor arrival, feature block, or intent shift.
Everything sliding slowly → competitive erosion, or your tracked set is ageing into queries that matter less.

What to actually do

Plot the history before reading the number. Endpoints hide the shape, and the shape is the diagnosis.
Classify it: cliff, slide, step, or flap. Do this before forming a hypothesis, so the hypothesis has to fit the shape.
For a cliff, fix the date and look for an event. Deploys first, then ranking-system news, then whether the result page itself changed.
For a slide, confirm it’s real with period averages rather than single checks. Then treat it as a content and competition problem.
For a step, find what changed state and check whether the new level is stable before responding to it.
For a flap, do nothing except widen your alert threshold.
Check the set, not just the keyword. A shape shared across keywords is a much stronger signal than the same shape on one.

The general diagnostic sequence, once you know what shape you’re dealing with, is in How to Diagnose a Ranking Drop. The shape is what tells you which branch of it to take.