Dividing a site's traffic by the number of keywords it ranks for tells you almost nothing about how well its SEO is working. The ratio mixes searches of wildly different frequency, positions that move from day to day, and estimated figures set against measured visits. A result above or below one separates nothing.
To judge keywords properly, keep three questions apart: which searches the site appears for, which of those bring clicks, and which pages contribute to something useful such as an enquiry, an order or a booked call. Each question needs different data and a different level of analysis, and collapsing them into one number is how reporting quietly stops being useful.
Most of the work is scope and discipline: fixed filters, assumptions written down, and a clear record of what changed and when. The sections below set out what each tool actually measures, how to build a tracking table that survives comparison, and how to size an opportunity without inventing a figure.
Why the traffic-per-keyword ratio misleads
A site can pick up dozens of new rankings on queries that are barely visible yet, without losing a single visitor. The keyword count rises, the ratio falls, and coverage has in fact widened. The reverse happens too: losing a batch of near-dormant queries pushes the ratio up while commercial results stay exactly where they were.
An average also hides the distribution. A handful of brand searches can account for most of the clicks while hundreds of other terms contribute almost nothing. Counting terms will not tell you whether the site depends on one page, one product line or one need.
Finally, the number of keywords detected depends on the tool's database. When that database is expanded or pruned, the count moves without any matching change in traffic. Treat the number as a coverage indicator inside a defined scope, not as a measure of return.
What each tool is actually measuring
Ranking tools: an estimate drawn from a database
The Semrush Organic Rankings report shows positions detected within the first hundred results of its selected database, together with an estimated monthly traffic figure. That is genuinely useful for comparing trends or spotting competing pages, provided the settings stay identical between checks.
Semrush's own documentation distinguishes between its estimation methods. Ranking-based figures combine positions, search volumes and click assumptions. They are not a window into a competitor's analytics account, and no universal multiplier converts the estimate into real visits. Read them as an order of magnitude, not a measurement.
Search Console: interactions with Google results
Google defines impressions, clicks, CTR and position in Search Console. CTR is clicks divided by impressions. Average position is an aggregate of appearances, not a fixed rank that every user sees at every moment, which is why comparing it across periods with different query mixes goes wrong so easily.
Look at query, page, country, device and period together. The data is generally attached to the canonical URL Google has selected, so a change of canonical version can break a comparison built on a single address. If pages keep appearing and disappearing from the report, that is an indexing question before it is a content one.
Analytics: what the site itself records
Your analytics tool tracks visits and events that were actually collected. The results depend on the configuration, on consent choices and on whether the tracking fires as intended. Check forms, orders and duplicate events before you let a conversion rate drive a decision.
The Search Console integration in GA4 keeps a queries report and a landing pages report separate. You cannot freely join every organic query to every analytics dimension, or to an individual order. For commercial outcomes, the page and the intent family are usually the level you can work with.
Build a tracking table that survives comparison
Set the scope first: web search, target country, period and the pages in scope. Keep the same filters every time you compare. Separate brand searches from generic ones where you can, and write down which name variants you counted as brand, because that single decision moves a lot of totals.
| Element | What it is for | Question to ask |
|---|---|---|
| Intent family | Grouping close formulations | Does the need match the offer? |
| Landing page | Identifying the content to work on | Is the right page appearing? |
| Impressions and clicks | Separating discovery from acquisition | Has demand, visibility or click behaviour moved? |
| Enquiries or orders | Observing the commercial contribution | Are the events and their qualification reliable? |
| Planned action | Connecting analysis to a decision | Which specific problem are we solving? |
Do not force the totals to agree. The dimension limits in Search Console include anonymised queries and a selection of rows. A filter on queries can exclude data that is still counted in the chart total, so an exported list does not necessarily represent every search observed. Note the gap, state it in your reporting, and move on.
Reading the long tail without a word-count rule
In demand analysis, the long tail describes the large set of queries that are each searched rarely. It is not defined by a boundary between three and four words. A short product reference can be searched almost never, while a long sentence can be a common question asked in a common way.
Group these expressions by the problem they describe, the product concerned or the stage of the decision. You do not need a page for every variant. One well-built page can answer several close formulations, and that is usually a better use of effort than a scatter of thin pages, which is worth settling before you commission any SEO content.
How much of your traffic the tail represents depends on the site and on the data available to you. Do not assign it a fixed percentage, and do not assume a precise phrase always converts better. Stock, price, trust and whether the offer genuinely fits still decide what happens next.
Prioritising work beyond position
A page that already appears can be a good candidate, but its ranking alone is not a reason to work on it. Cross the value of the need, the volume of useful enquiries, the problem you have identified and the effort required. A distant average position can belong to a strategic service that deserves sustained work, while an easy climb on an unrelated query may be worth very little.
Read the situation rather than applying a recipe
- Impressions up, few clicks: check the new queries, the result being displayed and whether the page matches the need behind them.
- Clicks down at a comparable position: look at demand, seasonality and how the results page itself has changed.
- Several pages for one intent: compare their function and their content before calling it internal competition.
- Clicks up, enquiries flat: check who is arriving, what the journey asks of them and whether the events are still being collected.
- Plenty of enquiries, few of them relevant: be more explicit about services, conditions and the area you cover.
Google's method for debugging search traffic drops points to seasonality, technical problems and groups of pages. A broad decline often has several causes at once, which is why a technical audit and a content review usually need to run alongside each other rather than in sequence.
Estimating an opportunity without inventing certainty
You can build a scenario from observed impressions and an assumed CTR. State the assumption, the period and the conditions you are treating as stable. The result is an estimate of clicks, not a sales forecast, and it reads better as a range than as a single number.
Avoid a universal table that assigns the same CTR to every position. Brand recognition, device, intent and the features present in the results page all change the context. Use your own comparable cases when they are genuinely close, and present two or three scenarios rather than one figure with false precision.
Measuring the effect of a change
Record the change, its date and the pages affected. Confirm it is actually live, then wait for enough data to compare comparable periods. If another campaign, a price change or a stock-out lands in the same window, keep that in the note.
A plain text log is enough. What matters is that the next person reading the chart can see what else was happening:
2026-03-04 /services/gutter-repair/
change: rewrote intro, added pricing section
scope : UK, web search, 28 days before / after
note : paid campaign paused from 2026-03-10A before-and-after shows movement alongside a change; it does not prove the change caused it. The useful decision is whether the content now answers the need better, and whether what you observe justifies continuing, adjusting or testing a different assumption. Deciding that question in advance is largely what it means to scope an audit properly.
Common questions
Can traffic divided by keyword count work as an SEO score?
It can describe an average inside a stable scope, but no threshold tells you whether a strategy is good or bad. It hides how clicks are distributed and it depends on the coverage of whichever tool produced the keyword count.
Why do Semrush and analytics give different numbers?
Ranking reports estimate traffic from positions, volumes and click assumptions, while analytics measures interactions collected on the site itself. The scopes and the methods differ, and no fixed conversion factor reconciles them.
Can I see which order each search query generated?
Not in general, using only the Search Console and GA4 link. Queries and analytics data cannot be freely combined. Work instead at the level of landing pages, intent families and the commercial outcomes you can actually observe.
Should I always start with positions four to ten?
That group can reveal opportunities, but priority also depends on commercial relevance, on the problem you have identified and on the effort required. A rank on its own guarantees neither a quick gain nor a profitable one.
Why do the Search Console query rows not add up to the total?
Some queries are anonymised or absent from the displayed rows, and filters and aggregation level change the scope as well. Check that you are comparing the same settings before you treat the difference as an error.
Is the long tail simply queries of four words or more?
No. It describes a distribution of individually infrequent searches. Phrase length is not a reliable boundary, and the share of traffic the tail represents varies from one site to another.