SEO KPIs: what to check monthly, and what to ignore

Why the list is decided in advance

The difference from a report is in the purpose. A report narrates, a review decides. If nothing about next month's work has changed by the end of the half hour, the review still worked: it told you to carry on.

A dashboard showing forty numbers does not produce decisions, it produces interpretations. Every month an explanation is found for any variation, and none of those explanations changes the following week's work.

A closed list, decided in advance, does the opposite: it forces you to say up front which variation would change something. It is the only useful definition of an important indicator.

What a good indicator needs

1. It has a threshold past which you do something specific. If you cannot say in advance what you would do on seeing it fall, it is not an indicator: it is an interesting fact.
2. It moves because of the work done, not only on its own. A number that swings with seasonality or with algorithm updates measures the market, not you.
3. It reads without you having to construct an explanation every time. If it needs a story to make sense each month, the metric is the problem.

The third condition eliminates the most candidates, and is also the easiest to skip.

The four indicators worth a monthly look

Four, because a review containing ten ends up done badly. The rest belongs to a diagnosis, not to a calendar.

1. Impressions and average position on the target phrases. Not the whole site: the group of phrases you decided to work on. The two metrics have to be read together, because apart they lie. The official definitions matter: an impression means a user saw, or may have seen, a link to the site, while average position is a relative ranking in which 1 is the highest. Google itself warns that the latter is a complex metric that can be misleading if you do not understand its subtleties, and further down you see why.
2. Citations recorded in AI answers. Search Console has a generative AI performance report, which Google made available to all websites globally from 31 August 2026 and which includes impressions for AI Overview and AI Mode. The available dimensions are pages, countries, devices and dates; the report exposes neither clicks nor queries. It is little, but it is your own data and it is free: as long as the only available measure was a paid tool's count, this indicator was not verifiable against your own numbers.
3. Crawl errors on the pages that bring traffic. The threshold here is not a number but a condition: a crawl error counts if it sits on a page that produced clicks in the last month. A 404 on a URL nobody visits and nobody links to is not a monthly-review problem.
4. Conversions assisted by the organic channel. It is the only one of the four that measures the result rather than the means. Read it knowing that time passes between the visit and the conversion, so the current month is always understated.

Thresholds: how far a number has to move

This is the part most often missing, and also the reason monthly reviews turn into meetings about nothing. With no stated threshold, any variation looks like a signal.

• On the site aggregate, under one position of movement nothing gets opened. The reason is how the number is built: for each query it takes the highest position the property occupies — or the individual page, if the data is grouped by page — and averages those across every query you appeared on. Change the set of queries and the average changes, without a single page having moved.
• On the target phrase group the filter does not remove the problem, it reduces it: there too, queries entering and leaving the set move the average. So it has to be read alongside the number of queries that produced impressions, because that is what explains most of the movement.
• On impressions, compare comparable periods. Same length, same number of weekends, mind the movable holidays. A month with five Mondays against one with four produces a gap by itself that looks like a trend.
• After a change or an update, wait before judging. Google writes that some changes may take effect within hours and others take several months, and generally advises waiting a few weeks to assess whether the work had a positive effect. An immediate reading measures neither the update nor your work.
• On errors, the threshold is binary. Did the page produce clicks? Then you act, whatever the site's total error count.

What moves on its own

A good share of a monthly review's alarms concern things that do not depend on the work done. They are worth recognising before chasing them.

Google lists the causes of a traffic drop in its debugging guide and distinguishes six: algorithm updates, technical issues, security issues, spam issues, seasonality and changing interests, and site moves and migrations.

On seasonality Google writes that user behaviour changes sometimes shift demand for particular queries, following new trends or the seasonality of those queries across the year. The examples it gives are diets in January and turkey in November.

And on updates there is a sentence worth rereading before every meeting where somebody proposes rewriting everything: if you suspect a traffic drop is due to an algorithmic update, it is important to understand that there may be nothing wrong with your content. Google also adds that it cannot guarantee changes made to a website will have a notable impact on search results.

This is not an invitation to inertia. It is the reason the list of indicators is decided in advance: so a movement in the market is not mistaken for a result of your own work, in either direction.

What can be ignored

Ignoring does not mean they do not exist, nor that they should never be opened. It means not giving them a monthly slot, because they do not produce a decision every month.

Day-to-day position swings

Watching them daily anticipates nothing: it only adds occasions to react to noise. Position makes sense read by query group over windows of weeks.

Composite scores calculated by tools

Domain authority, visibility indices, audit platforms' scores out of a hundred: these are proprietary estimates, each vendor calculating its own with its own model, and not comparable with one another. The point that matters is a different one: they are not Google metrics. Google does not produce them, does not expose them, and documents no correspondence between these scores and its own signals. You often read that they measure a site's authority in the eyes of search engines: any relationship with ranking is in fact an estimate by whoever sells the tool, not a verifiable figure. They are useful at most as a relative indicator of your own trend over time, within the same tool. They are not a target.

The PageSpeed Insights score out of a hundred belongs in the same group: it is a simulated lab test. The Core Web Vitals assessment is based instead on data collected from real users, at the 75th percentile of loads, segmented between mobile and desktop. The score is for diagnosis, not for establishing whether a page is inside the threshold.

Bounce rate and time on page

They appear in almost every KPI list in circulation, rarely accompanied by an explanation of how they are built. In Google Analytics 4 an engaged session is a session that lasts longer than ten seconds, or contains a key event, or contains two or more page or screen views — and the ten-second threshold is a default, adjustable in the settings. Bounce rate is the opposite of engagement rate: the percentage of sessions without engagement.

Once you know that, the number says a great deal less than it seems to: someone who opens the page, finds the answer in eight seconds and leaves satisfied is counted exactly like someone who leaves disappointed. Improving that number does not necessarily mean anyone was served better. Engagement time has a different but equally awkward flaw: it measures the seconds the page is in the foreground, so it does not distinguish close reading from a tab left open while doing something else.

The number of indexed pages, read without context

Google is explicit in the page indexing report: do not expect every URL on your site to be indexed, because some URLs may be duplicates or contain no meaningful information. A falling count can mean the site is being cleaned up. What counts is which pages are missing, not how many.

Google's own list, which almost nobody cites

There is a section of the official documentation titled “Things we think you shouldn't focus on”, and it is the closest thing to an authoritative list of pointless worries. Among the entries:

• The keywords meta tag, because Google Search does not use it.
• Content length: length alone is not important for ranking, and there is no magic minimum or maximum word count.
• Keywords in the domain or the URL: from a ranking point of view, using keywords in the domain name or URL path has almost no effect.
• Duplicate content reachable from several URLs: not a problem, so do not worry about it. Google adds that it is not an efficient arrangement, though — so the entry leaves the monthly review, not the list of things to tidy up in due course.
• The idea that E-E-A-T is a ranking factor, which Google denies directly.

None of these deserves a place in a monthly review, and yet several still appear in the report templates going around.

The half-hour review

An order that works, and that finishes in the time it takes to drink a coffee.

1. Open the performance report filtered to the target phrases, compare against the same period last year, and look at impressions and average position together. Five minutes.
2. Open the generative AI report and check which pages the site appears on inside answers. Five minutes.
3. Cross the crawl errors against the pages that produced clicks in the last month. If the intersection is empty, close it. Ten minutes.
4. Look at conversions assisted by organic, remembering the attribution lag. Five minutes.
5. Write two lines: what changed, and what you will do differently. If the answer is “nothing”, write that too. Five minutes.

The fifth step is the one usually skipped, and the only one that makes the review something other than a habit.

What to look at once a year

Some things matter but do not change at the pace of a month, and putting them in a monthly review produces only noise: the category structure and the navigation, how content is distributed across the themes you hold, the pages that have never produced a visit since they existed, the consistency of external references to the brand. Once a year, unhurried, they are useful work. Every thirty days they are not.

Common questions