Position tracking: what to measure and when to sound the alarm

You've written the pages, they got indexed — now you need to understand whether they're growing or falling. For that, positions for keywords are measured regularly and the trend is watched. Without tracking you work blind: it feels like "everything's fine, I guess," while in fact half your queries have slid to the second page.

Let's break down what exactly to record, how to tell normal swings from a real disaster, and when it's time to step in.

What to record in a measurement

A position on its own is meaningless without context. One measurement is five fields:

Query — which phrase we're checking.
Position — what rank in the results.
URL — which page exactly is ranking. More important than it seems: if the wrong page suddenly ranks for a query, that's a signal of trouble, even when the position is decent.
Region — Moscow, St. Petersburg, all of Russia. Positions depend heavily on geo.
Engine — Yandex and Google rank differently, they must be measured separately.
Date — without it there's no trend.

Watching the trend for a single keyword is hard — there are hundreds of them. So queries are grouped into clusters and you watch how the whole group moves: it's steadier and more honest than the jumps of a single phrase.

Normal swings versus a real problem

Positions always fluctuate. This is called SERP volatility, and there's no need to fear it.

What's normal and what isn't

Normal volatility

One phrase jumps 2–3 positions day to day
A dip for a couple of days, then a rebound
Different phrases move out of sync

A real problem

The whole cluster slid down in sync
The drop holds and doesn't rebound
A different page ranks for the query, or it dropped out entirely

One phrase jumped three positions and came back — noise. The whole cluster slid down and isn't returning — a signal.

The key difference is scale and persistence. One phrase twitched and came back — noise. The whole group dropped and holds for several measurements in a row — something happened: an algorithm update, a technical failure on the page, a competitor stepping up.

The alarm rule

To avoid reacting to every rustle and to avoid missing a real disaster, set yourself a threshold in advance. A working rule:

Sound the alarm if five or more keywords from one cluster dropped by five or more positions in a single measurement — and it didn't rebound by the next one.

This double threshold (many keywords + a noticeable shift + persistence) filters out volatility and catches real drops. It ignores lone jumps, but a synchronized cluster collapse it highlights right away.

How not to sit in reports by hand

Watching tables every day is impossible. So tracking is set up on alerts: the system measures positions on a schedule itself and writes to you only when the alarm threshold triggers. In our service it's arranged like this — the measurement runs on a schedule, and the notification about a cluster drop arrives by email and on Telegram. You open the report not "just in case," but when you actually need to.

That's the point of tracking in passive mode: not to watch by hand, but to get a signal when something breaks.

Common questions

How often should I measure positions?

For most projects once a week is enough. Daily measurement is needed in hot commercial niches or when you've just changed something and are waiting for a reaction.

Positions in Yandex and Google differ — is that normal?

Yes, they're different algorithms. They must be measured and analyzed separately; you can't merge them into one number.

Why can't I just look at traffic instead of positions?

Traffic is the outcome, positions are the cause. Traffic can hold on by inertia while positions have already slid. Tracking catches the problem before it hits your visitor count.

The last step of the guide — how to go beyond classic search and get into the answers of neural networks.