Keyword Metrics Explained
A guide about numbers that contains none. Every figure here is explained by what it is and how it was built rather than by what counts as a good one, because almost all of them were produced by the tool showing them.
Where These Numbers Come From
Google does not publish how often people search for things. So no keyword tool is reading a real total. Each one builds its own figures from sampled behaviour, patterns across similar searches, data bought from elsewhere and assumptions filling the gaps. That is reasonable work and it is not measurement.
The one exception. The advertising figures.
Those derive from a system where money changes hands, which makes them the only numbers in the whole set grounded in something real. Block four covers why that matters more than it sounds.
The clearest proof of the rest. They disagree.
Run one search through two tools and the figures differ. If either were reading a total they would match, so their disagreement demonstrates the point without any argument.
Why they look authoritative. Precision implies certainty.
A specific number in a clean column reads as a fact. Shown as a range with a note about method, nobody would treat it the same way.
Why this page publishes no figures. They get repeated.
Any number printed here would be quoted back as a rule, including a figure given as an example of a bad one. So none appears, including in the refusals.
What the metrics are still for. Comparison.
Judging one search against another, roughly. That survives the estimation and it is genuinely useful.
What they are not for. Forecasting.
Multiplying figures together to predict visitors produces a number with no basis. It appears in proposals constantly and it should be treated as a warning sign.
Search Volume
The number everybody looks at first and the one that decides most keyword lists. It estimates how often a search is made, usually as a monthly average. Its reliability varies enormously depending on how large the search is.
What it is. A modelled average.
Not a count and not last month's total. An average across a period, smoothed, which hides seasonal peaks completely.
Why the averaging matters. Seasons disappear.
A search that is enormous for three months and absent for nine shows a modest average all year. For a trade with a seasonal shape, that is actively misleading.
Where it fails hardest. Small numbers.
Estimating a large national search is achievable. Estimating a specific service in one town has almost nothing to model from, which our local guide covers.
The rounding problem. Bands hide detail.
Figures grouped into ranges conceal how small or uncertain a number really is. Two searches shown identically can be very different in reality.
What zero does not mean. Nobody searched.
An empty column frequently means too small to estimate rather than absent. Never delete a commercially sensible search because of it.
How to use it. Relative scale, then stop.
Whether one search is much larger than another is usually right. Beyond that it should not be deciding anything.
Keyword Difficulty
A single figure estimating how hard a search would be to win. Every tool has one and they all calculate it differently. Every one of them shares the same fundamental limitation: the number describes the search without knowing anything about the site asking.
What it is measuring. The incumbents.
How strong the sites currently appearing are, expressed as one figure. Reasonable information and half of what you actually need.
What it cannot know. You.
The same search is achievable for an established local firm and impossible for a site launched last month. One number cannot express both. It does not try to.
Why the tools differ. Different recipes.
Each calculates it from its own inputs with its own weighting. Two tools disagreeing about difficulty is expected rather than a sign one is wrong.
What we will not publish. A threshold.
No figure below which a search is safe to pursue. Any such rule ignores your site entirely, which is the one thing that matters most.
What to use instead. The results themselves.
Two minutes looking at who ranks tells you more than any score. If it is national brands and directories, that answers it. If it is businesses like yours, it is available.
The signal worth more than the score. Weak answers.
Poor pages ranking means nobody bothered rather than everybody tried, which is the strongest opportunity signal there is.
Where the score still helps. Breaking ties.
Choosing between two otherwise equal options. That is its entire legitimate use.
The Advertising Figures
The most useful number in a keyword tool is the one about advertising. Almost nobody reading it wants to advertise. Cost per click is the only figure in the whole set derived from money somebody actually spent, which makes it the closest thing to hard evidence available.
What it represents. Real bidding.
What advertisers are prepared to pay to appear for that search. Nobody bids repeatedly on something producing nothing, so the figure carries information no estimate can.
What it tells you as an unpaid search. Commercial value.
A search with meaningful advertising behind it has been established as profitable by businesses risking their own money. That is the strongest signal of buying intent in any keyword report.
Why that inverts the usual reading. Cheap is not good.
People scan for low competition and low cost. A search nobody will pay for is frequently a search nobody makes money from, which is the opposite of an opportunity.
How to actually use it. As a sorting question.
Ask which searches on your list have advertisers behind them. Those are the ones the market has already tested. They belong near the top of your list.
The caution. It measures a different game.
Advertising placement and unpaid placement are separate contests. A high figure says the search is valuable rather than that appearing without paying is easy.
Why no amounts appear here. They vary by everything.
Trade, region, season and competition all move it. Any figure quoted would be wrong for most readers, so we describe the signal rather than the number.
The related figure. Volume of advertisers.
How many are bidding, which is block five and gets confused with difficulty constantly.
Competition, Which Is Not Difficulty
Reports show a competition figure alongside difficulty. Businesses read them as the same thing. They are not. Competition describes the advertising contest and difficulty describes the unpaid one, which means a search can be crowded in one and open in the other.
What competition measures. Advertisers bidding.
How many are competing for paid placement on that search. Nothing about the unpaid results at all.
Why the confusion persists. The label.
Two columns sit beside each other with similar names and no explanation. Nobody reading a report is told they describe different contests.
The combination worth noticing. High competition, low difficulty.
Advertisers fighting over a search while the unpaid results are weak. That is an unusually good opportunity and the two columns together are what reveal it.
The reverse combination. Nobody advertising, hard to win.
Strong unpaid competition and no commercial interest usually means an informational subject that large publishers cover thoroughly.
What to do with it. Read it as intent.
Treat competition as a second confirmation of commercial value rather than as a measure of how hard anything is.
Where the whole argument sits. The tools guide.
Our guide to keyword tools covers what to distrust across all of them.
Click Estimates
Some tools project how many visits a given position would bring. That figure is the most derived thing in any keyword report, because it multiplies an estimated volume by an assumed share of clicks. Both inputs are uncertain before they are combined.
How it is built. Two guesses multiplied.
An estimated volume, times an assumed proportion of people who click at a given position. Errors in each compound rather than cancel.
Why the second input is weak. It varies by search.
How many people click at all depends entirely on what the results page looks like. A page answering the question directly changes it completely.
What we will not publish. Any click curve.
The figures circulating for clicks by position are third party estimates that differ enormously by search type. None appears here.
Where these projections appear. Proposals.
Usually as a promise of visitors if a position is achieved. Treat that as a sales device rather than a forecast.
What to ask about one. The assumptions.
Which click rate was assumed and where it came from. There is rarely a good answer and the question ends the discussion.
The related figure. Clicks against searches.
Some tools estimate what proportion of searches produce any click at all. Directionally interesting, though built on the same foundations.
What to use instead. Your own reporting.
It records what actually happened on your site rather than projecting what might happen on any site.
Trend And Seasonality
Trend data is the most trustworthy thing in a keyword report, for a reason that sounds like a criticism. It is usually relative, comparing a search against itself over time rather than stating a total, which means it makes a smaller claim and keeps it.
Why relative is better here. Less to get wrong.
Comparing something to its own past requires fewer assumptions than estimating an absolute figure. The uncertainty largely cancels.
What it shows well. The shape of a year.
When demand rises and falls for a trade, plus whether the pattern repeats. That decides when to publish rather than what.
The second use. Direction over years.
Whether interest in a subject is growing or fading. Useful for deciding whether something is worth building at all.
The third use. Settling vocabulary.
Which of two phrasings for the same thing is used more. That resolves arguments quickly and without any subscription.
Why it corrects the volume figure. Averages lie about seasons.
A monthly average flattens a subject that is enormous for one quarter. The trend view restores what the average removed.
Its limit. Small subjects vanish.
Anything specific or local frequently shows nothing, for the same reason volume estimates fail there.
The better version for your own site. Your reporting.
A year of your own search data shows the shape of your year rather than the shape of the market's.
The Scores That Are Not Keyword Metrics
Keyword reports frequently display a score describing a website's overall standing, sitting beside the search figures as though it belonged there. It does not. Those scores were invented by tool makers and they are not used by search engines. They describe sites rather than searches.
What they are. Commercial estimates.
One company's attempt to summarise how strong a website is, on its own scale, using its own inputs. A reasonable thing to build and not something anybody outside that company uses.
Why they appear here at all. Difficulty needs them.
Difficulty scores are frequently calculated from the standing of the sites currently ranking, so the two figures come from the same machinery.
What they are useful for. Rough comparison.
Whether the sites above you are much stronger than yours. That is a legitimate reading and it is what the number was built for.
What they are not. A target.
Working to raise a score produced by a third party is optimising for somebody else's product rather than for your business.
The reporting problem. They look official.
A score presented alongside search data reads as a measurement from the search engine. Nothing about the presentation says otherwise.
Where they belong. Another discussion.
Our measuring performance material covers site level scores properly, including why they should never appear as a headline in a report.
What To Actually Do With Them
Two questions decide whether a search belongs on your list. Could you realistically appear well for it. Would appearing produce work you want. Every figure above serves one of those questions or neither. Knowing which is what turns a report into a decision.
Serving the first question. Difficulty, weakly.
Difficulty and the standing of incumbents give a rough sense of contestability. Reading the actual results serves it far better.
Serving the second question. The advertising figures.
Cost per click and competition are the best available evidence of commercial value, because they reflect money somebody spent rather than a model.
Serving neither well. Volume.
It tells you how many people might arrive and nothing about whether they matter. Useful for breaking ties and dangerous as a sorting column.
Serving the calendar. Trend.
When to build rather than what. That is a scheduling input and it is the most reliable figure on offer.
Serving nothing. Click projections.
Built from other estimates and presented as forecasts. Nothing on your list should be decided by one.
The order to work in. Value first.
Filter for commercial worth before assessing difficulty, so you never spend time judging the winnability of things you would not want. Our method guide covers the whole exercise.
What no figure can supply. The judgement.
Which searches lead to work you want, at margins you accept, in places you serve. That stays with you. Everything sits on the keyword research guide.
The most useful
number in a
keyword tool is
the one about
advertising.
Cost per click is the only figure in the set derived from money somebody actually spent. Nobody bids repeatedly on a search that produces nothing, so it carries evidence no estimate can. Meanwhile people scan for low competition and low cost, when a search nobody will pay for is frequently a search nobody makes money from.
How we read a keyword report:
If somebody shows you a forecast of visitors, ask which click rate was assumed and where it came from. There is rarely a good answer.
Every guide.
One subject.
What the words mean, what the numbers in a keyword tool actually are, how to find the searches worth targeting, how to turn a list into a publishing plan and what to do when a page you relied on slips.
