How can I track my keyword rankings?

Rank tracking is one of the easiest things in SEO to set up and one of the easiest to misread. Most of the damage comes from three habits: averaging away the signal, trusting a sampled location as if it were universal, and monitoring a surface that no longer decides the outcome on its own.

Two sources, two purposes

Google Search Console is your source of truth. It reports average position derived from real impressions — every device, every location, every personalised variation, weighted by how often each actually occurred. Nothing else can do that, and it is free.

Its limitation is that it is an average, which makes it truthful and coarse. It also only knows queries you already appear for, and it cannot see competitors at all.

A third-party rank tracker checks a fixed list of terms from a fixed location on a schedule. That gives you daily granularity, competitor positions, and the ability to ask “what does someone in Manchester on mobile see?” — none of which Search Console offers.

Its limitation is that it is a sample. One location, usually desktop, one moment a day, with no personalisation. It is precise about a narrow context and unrepresentative of the whole.

Search Console tells you what happened. A rank tracker tells you what a specific hypothetical searcher would see. Comparing their numbers directly is meaningless.

What to actually track

Five things, in descending order of usefulness.

1. Query count by position band. How many queries sit in positions 1–3, 4–10, 11–20, and 21+. This is the clearest picture of visibility available and it is hard to game or misread.

2. Clicks from non-branded queries. Branded traffic reflects marketing generally; non-branded reflects SEO specifically. Reporting them together hides which one is working.

3. Impressions by query group. Segmented by intent — informational, commercial, branded. A site-wide impression number moves for reasons nobody can attribute.

4. CTR against typical rates for your positions. Materially below par at a good position is the cheapest opportunity in SEO, and it is invisible unless you compare against the expected rate rather than an absolute threshold.

5. Positions for a small set of genuinely strategic terms. Ten to twenty, not five hundred. This is the one place where a third-party tracker earns its cost, and it should be a deliberate list rather than everything you rank for.

Why position bands beat averages

Average position is the most commonly reported and least useful number in the discipline.

Consider a site whose average position improves from 18.4 to 16.1. That looks like progress. It could be produced by fifty irrelevant queries moving from 90 to 40 while your five commercial terms slipped from 4 to 7 — an unambiguously worse quarter reported as an improvement.

Bands do not have this problem, because they only move when something meaningful happens. A query entering the top ten is a real event. A query moving from 34 to 29 is not, and a band view correctly ignores it.

Track the counts, chart them over twelve months, and the trend is both honest and legible to people who do not do SEO.

Reading the data honestly

Four rules prevent most misreadings.

Compare like periods. Year over year, not month over month, unless you genuinely have no seasonality. Most sites do, and comparing November to August produces a crisis that does not exist.

Expect fluctuation. Positions move daily for reasons unrelated to anything you did — personalisation, testing, ordinary volatility. A two-position change over a week is noise. A five-position change sustained over a month is signal.

Check impressions before concluding a ranking change caused a traffic change. Position can hold while impressions fall because demand fell, or because an AI Overview absorbed the visibility. Those are different problems with different responses.

Segment before averaging. Every useful finding in ranking data is inside a segment. The blended number is where findings go to hide.

How often to look

Weekly for monitoring, monthly for reporting, daily only when actively diagnosing something.

Daily checking is actively harmful in normal operation. Normal fluctuation looks like a trend at short intervals, and teams that check daily make changes in response to noise — which then makes it impossible to attribute anything, because three variables changed in the fortnight before the number moved.

The exception is incident response. If traffic has dropped, daily granularity and a stored history matter enormously, because the shape and timing of the decline is most of the diagnosis. Our guide to diagnosing a traffic drop covers reading that shape.

What no rank tracker can see

Conventional trackers report a fixed, largely US-centric, daily-sampled, blue-link keyword set. That was a complete picture of search once. It is now a partial one.

Two things are invisible to it. AI Overviews appear on roughly a third of monitored queries — tracking through 2026 puts presence at around 32.5%, in a narrow 31–34% band — and can absorb the click while your position is unchanged. A tracker showing a flat line during a quarter when your clicks halved is not reassurance; it is a measurement gap.

Assistant answers are invisible entirely. When a buyer asks ChatGPT which tool to use and it names three companies, no impression was served, no position assigned, no visit occurred. Your tracker records nothing, because nothing happened on a results page — but a shortlist was formed, and if you were not on it you never entered the funnel.

Tracking that surface requires a different unit. Not position, but: were you mentioned, were you cited, who appeared instead, and what share of the answer space do you hold across a prompt set. Our guide to tracking AI brand visibility covers the method.

Setting up tracking properly

Four decisions at setup determine whether the data is useful for the next two years.

Choose the keyword set deliberately, and freeze it. Ten to twenty genuinely strategic terms, chosen because they matter commercially rather than because you rank well for them. A list that changes when the numbers get awkward is worthless for trend analysis and destroys credibility the first time someone notices.

Set the location and device to match your customers. A default US desktop configuration is wrong for most businesses and produces numbers nobody in the company would recognise from their own searching.

Add competitors at setup. Your position is far more meaningful relative to the two or three sites you actually compete with than as an absolute number. Adding them retrospectively means no historical comparison.

Export Search Console monthly to your own storage. Search Console retains 16 months of data. Anything older is gone permanently, which means a year-over-year comparison in month eighteen is impossible unless you started archiving in month one. This costs nothing and is invariably regretted only in retrospect.

What to do with the data weekly

Fifteen minutes, three questions. Has anything in the top ten dropped out? Has anything entered the 8–20 band that was not there before, which is a new opportunity rather than a problem? And has CTR fallen anywhere with stable position, which usually means the results page changed around you rather than anything on your side.

That is the entire routine. Anything more elaborate becomes a ritual nobody sustains, and anything less means finding out about a problem from a quarterly report.

The annotation habit

Keep a dated log of everything that could plausibly affect rankings: deploys, migrations, content updates, redesigns, tracking changes, known algorithm updates. Five lines a month.

When something moves in six weeks, the log turns an unanswerable question into a lookup. Without it, diagnosis becomes reconstruction from memory and commit history, which is slow and usually wrong. Nearly every team that has been through one bad incident starts doing this, and nearly none do it beforehand.

Reporting it to other people

Three numbers, monthly, each with one sentence of context:

• Queries in the top ten, as a count and a trend.
• Non-branded organic clicks.
• Conversions or pipeline from organic, which is the only one anyone outside the team genuinely cares about.

Resist reporting positions for a handpicked keyword set. It is the metric most vulnerable to selection bias, and once a stakeholder notices that the list changes when the numbers get awkward, every other figure you report loses credibility alongside it.

Add one more line if you can: whether assistants name you when someone asks about your category. It sits in the same funnel, it is invisible in every conventional report, and by the time its absence shows up as a pipeline problem it has usually been true for months.

Sources

• Google Search Console documentation — on how average position is calculated from impressions across devices and locations.
• SERP feature tracking, 2026 — AI Overview presence averaging ~32.5% of monitored queries within a 31.02%–34.40% band.
• Ahrefs — CTR impact of AI Overviews; approximately 58% decline for top-ranking pages (February 2026).

Frequently asked questions