From Raw Data to Decisions: Reading Search Console and SERP Analytics

A search report is not a decision. It turns into one only when somebody has written down in advance which number, at which level, sets off which piece of work. In a city that sells capability rather than price, that written rule looks nothing like the textbook version.

Every verified domain hands you daily figures on impressions, clicks, click-through rate and position. The shortage was never data. What goes missing is the translation step: the moment a number becomes a task with an owner and a date on it.

Two habits block it: trusting one averaged figure that has flattened a dozen unrelated situations, and reading without a threshold, so observations end as remarks rather than jobs. Both are dealt with below, across the fourteen analytical views in the rebuilt Semalt panel.

Two data sources that answer different questions

The two get treated as views of one thing, and they are not. Search Console is an official connection to a verified property. It knows every query that produced an impression for you, and nothing about queries you have never appeared for. Rank tracking runs a standardised, depersonalised query and records what comes back, including the domains above and below you. It knows the field; it knows nothing about clicks.

The panel splits this into eight Search Console views and six ranking views — a menu that works as a division of labour, each answering one question well and every other badly. A fabricator chasing a drop in enquiries opens the Pages view; a machine shop asking who bids against it online opens the competitor list, where Search Console has nothing to say.

Impressions, clicks, CTR and position: four ways to be misled

The four values at the top of every report are the most quoted numbers in the trade and the most misread. Each alone describes a situation that may not be yours.

Impressions rise for reasons unrelated to you. A results page showing ten organic entries instead of eight lifts your count with no change on your side: read them as surface area, not demand. Clicks are honest but coarse. For a supplier of guarding and conveyor components, forty clicks from students copying a standard number look exactly like forty from maintenance planners, and only one group requests a quote.

Click-through rate hides its denominator. Nine per cent on eleven impressions is noise; one and a half per cent on three thousand is a title problem worth a morning. Sort by impressions first, then read the rate.

Average position weights by impressions, not by value, and that is where it fails hardest here. A category name with thousands of impressions at twenty-six drags the average down harder than a specific query with two hundred at position two lifts it — and the second is the one attached to a purchase order. The scale is uneven too: twelve to seven changes traffic materially, thirty-one to twenty-six changes nothing, and both are five places. It also punishes growth, because every new term enters low.

Position bands answer all three objections by replacing one number with a distribution: eleven terms in the Top 3, forty-one in the Top 10, a hundred and six in the Top 30. Those counts survive comparison month on month.

The value sits in the clustering, not the row. One term slipping off page one is weather; four leaving at once from one URL share a cause.

When the query carries a part number

Search here is technical. The strings arriving at a fabrication shop, a machining supplier or an equipment dealer carry alloy grades, standard numbers, capacities and model codes: small volumes, no ambiguity. Somebody typing a material spec and a plate thickness is not browsing.

That rewrites what a keyword report is worth. Where consumers search, sorting by volume is a defensible start. Here it is backwards — the terms with the biggest numbers beside them are the vaguest, and the ones at the bottom are what a specifier types with the drawing on the desk.

The consequence is a reporting rule. Split the list into a specification set and a general set. On the specification set judge clicks and enquiries, and ignore click-through rate: the denominators are too small to mean anything. On the general set judge impressions and rate, and accept that a Top 10 place there is a branding outcome.

The query view and the page view are not two versions of one report

The Keywords view answers what people searched for; the Pages view answers which URL the engine chose to show. They overlap enough to feel redundant and differ exactly where the money is.

One page earns impressions on dozens of related strings, which is why a single strong specification page carries a product family. The reverse is more damaging: a query where the engine keeps switching between a catalogue listing and a service page that each half-answer it. Keywords shows a mediocre position and no cause; Pages shows the term split across two URLs.

• Open the query view to ask whether the search result is doing its job. High impressions with a low rate is a title problem, not a ranking problem.
• Open the page view to ask which asset carries the result, whether two URLs split one term, and which pages deserve the next edit.
• Cross the two when a number moved and nobody can say why. Filter the page, read its queries: the cause is usually one term, not a general decline.
• Trust the page view for value because enquiry forms, spec sheets and quote requests live on URLs. Queries do not convert; pages do.

Set the best-pages list from the ranking module beside the Pages view monthly. It rarely matches the pages the team believes are important.

The country split and the device split, read properly

Two of the eight views exist to stop you averaging across situations that have nothing to do with one another. Both are quick to read and routinely skipped.

The country split matters more for technical catalogues than for most sites, because specification pages travel. A page on a fitting standard attracts readers anywhere the standard is used, and few will buy from a supplier in Alberta. If a third of your impressions arrive from outside Canada, a third of your position, rate and volume describe a market you do not serve.

The device split is different. A specifier reads on a desktop with a drawing open beside the browser; the same firm's field staff search on a phone for a parts counter or an opening time. Two jobs on one domain, failing in different ways.

Date comparison, and the two days you should not read

Comparing two periods is the most useful thing you can do with either source and the easiest to get wrong. The presets, twenty-eight and ninety days, already allow for the two-day lag in Google's own processing — the biggest single source of false alarms.

Somebody opens the report on Monday, sees a sharp fall across the last days of the window and calls a meeting. Two days later the graph is flat, because those days were never fully computed. Three agreements remove it: the last forty-eight hours are not interpreted, reports keep three days of clearance from the end date, and the weekly check sits late in the week.

Beyond the lag, compare like with like. Twenty-eight days against the previous twenty-eight keeps the same count of working days in each half, which month-against-month does not. And date your own work: a rebuilt catalogue template or a migrated URL structure shows up in these graphs.

Where the numbers stop and judgement starts

Here analytics in this market parts company with the standard advice. A specifier finds you, reads for an afternoon, leaves, comes back three weeks later from a bookmark, checks one table and sends one enquiry. The decision may then take two more months and involve people who never visited the site.

No click-based measurement survives that. The last click is usually your own brand name, which says nothing about which page did the work. Traffic is a poor proxy in both directions: a month with fewer visits and four serious quote requests beats a month with double the traffic and none. The answerable question is which pages precede a quote request, across the run-up rather than at the end of it.

• Ask on the form. One optional field asking what brought the enquiry. A third of people answer, and that third names pages no report would have flagged.
• Ask on the phone. Whoever takes the call notes where the caller read the specification. Two months of notes reshape the priority list.
• Match queries to enquiries. Find the technical terms last quarter's enquiries used, and check whether you appear for them. The gaps are the work list.
• Count enquiries, not sessions. Report one number upward: qualified enquiries per month, with the search figures underneath as explanation.

Thresholds turn this into a routine that fits in half an hour a week. Each line below carries a number, and the number decides whether anything happens.

That routine raises the question of who does the resulting work. For firms without in-house search capacity, the panel connects analysis to execution at two levels, and these views stay available in both.

Those thresholds are starting points. A domain with seventy ranking terms needs different boundaries from a portfolio of four thousand. Calibrate by running last quarter through the analytical views Semalt provides and counting how often each would have fired: anything firing weekly is too loose, anything that never fires is decoration.

Frequently asked questions

What actually tips the balance

Whether search data is worth anything to an industrial supplier rarely depends on access to it. It depends on whether somebody wrote down which observation leads to which action, and whether that rule contains a number.

The Edmonton version has one extra clause. Before reading any figure, ask which market it describes: the broad category traffic that will never buy, or the exact technical queries that precede a quote request. Both live in one report and behave like opposites, and averaging them produces a number that is true, stable and useless. A dated record of the technical and editorial changes you make turns the rest into evidence; further analyses sit on our blog.

To see your own figures arranged this way, connect your property to the Semalt dashboard with your own domain and let a full quarter run through it. What helps first is almost never the average position. It is the list of specification queries that quietly left the first page — and on a technical catalogue that list is short, specific, and worth more than everything above it.