Labelling Search Console Queries by Intent

A query you can count is a query you can act on. Take the table in Search Console's Performance report, give each row one intent label, and add up impressions and clicks per label. The totals show which kinds of content your pages already earn attention for and which kinds they do not answer. The method suits a site owner, an editor or a freelance SEO working with free data. Every example comes from one niche, vegan protein bars.

What the Performance report gives you to label

The Performance report in Search Console (Opens in a new tab) shows clicks, impressions, click-through rate and average position, grouped by tabs for queries, pages, countries, devices, search appearance and dates. According to the help page, grouping by queries groups your data by the search query users typed. Data grouped by queries, countries, devices or dates is aggregated by property. Data grouped by pages or search appearance is aggregated by page. The difference matters when you add figures up in a spreadsheet, because a page-level total and a property-level total are not the same number.

The four measures

Clicks count the times a user clicked through to your site from Google Search results. Impressions count the number of times your site appeared in search results. Click-through rate is the click count divided by the impression count. Average position is the average position of the topmost result from your site, and in the table it is the average for the specific URL or grouping shown in that row. Keep these definitions in mind, because labelled totals need recomputed ratios.

Date range and search type

The default view covers the past three months, and a date filter changes the period. A search type filter shows data for web (text-based or multimodal), image, video or news searches. Practical suggestion: build the first label table from web searches only. Image and video queries answer different needs and deserve their own table later.

A label vocabulary: question, comparison, price, brand

A label is a guess about what a searcher wanted, made from the words typed. It is not data Google gives you. Keep the vocabulary short and write the rule for each label before you touch the first row, so two people labelling the same export land on the same result.

Table: A label vocabulary with vegan protein bar examples

Why best-of gets its own label

A best-of query is not a comparison and it is not a question. The searcher is ready to choose and wants a shortlist. A comparison query already has the options and wants the difference. Lumping the two together hides the fact that they need different pages: a ranked list in one case, a side-by-side table in the other.

Why the ingredient label is worth keeping

Ingredient and allergen queries are narrow, and a product page or a short explainer can answer them. If the totals show this label growing, the signal is about a specific need, not about general interest in protein bars.

Handling short, misspelled and ambiguous queries

Real queries are short, misspelled and ambiguous. Word rules catch the clear cases and leave a remainder for hand review. That remainder is where judgment goes.

Two labels on one query

"Best cheap vegan protein bar" holds a best-of intent and a price intent. Pick one primary label, the one the searcher is likelier to want answered first, and write the other in the second column. This avoids counting the row twice while keeping the second need visible.

Reading the page behind the query

When a rule fails, open the Pages tab, filter to the query and look at which URL appears. If a product page appears for a query you labelled as a question, the mismatch itself is the finding: the searcher may want an explanation and the site offers a purchase page.

From screen to spreadsheet

Reading hundreds of rows on screen is slow. The help page notes that reports provide an export button to download both the chart and table data, and it adds that values shown as "~" or "-" in the report will be zeros in the downloaded data. Treat those zeros as unknown, not as true zeros.

Row limits and the API

The Search Console API data export page (Opens in a new tab) states that Performance report data is limited to 50K rows of data per day per type (web, news, image and so on) per property. The API offers filtering, sorting and aggregation type, not freeform SQL, and it is free with a usage quota. A small food site rarely needs the API. Google's limits can change, so check the current page before you plan around them.

The labelling routine

1. Export three months of web search data, with the Queries tab as the base.
2. Sort by impressions, highest first, and keep the top rows for hand labelling.
3. Add a label column and, next to it, a column for a second label.
4. Label the first fifty rows by hand and note any rule you invent on the way.
5. Write simple word rules for the rest: contains "vs" or "versus" gives comparison, contains "best" gives best-of, contains "cheap" or "price" gives price.
6. Check the rows the rules touched on a small sample before trusting them.
7. Pivot the table: label in rows, sum of clicks and sum of impressions in values.
8. Compute click-through rate per label as total clicks divided by total impressions.

Why not average the click-through column

Practical suggestion: do not average the per-row click-through rates. A row with a very large number of impressions and a row with a handful count equally in an average, and the result misleads. The help page defines the rate as clicks divided by impressions, so apply the same definition to the label totals. For position, an impressions-weighted average is a steadier summary. The documentation has no statement on weighting; it is a working choice.

A word-root table for writing rules

Rules need a shared list of trigger words. Kept on one sheet, the list makes two people give the same label to the same query. It is a starting point, and it grows with the queries of each site.

Table: Trigger words and their false matches

The third column explains why rules leave a remainder. A row holding "how much" gets better treatment from a person than from an automatic price label. Test every new rule on a small sample before it touches the full sheet.

Synonym handling

Searchers name one product several ways: protein bar, protein snack bar, plant protein bar. The three phrases carry one need. Labelling folds them together, so the label total is less fragmented than a keyword list. Grouping by intent instead of tracking each spelling also keeps the content plan short.

Turning labels into page and content decisions

A label table is worth the effort only if each label leads to a decision. The decision depends on two things: how much attention the label receives and whether a page on your site answers it.

Table: Label, expected page and decision when it is missing

When to act on a label

Practical suggestion: when a label earns high impressions but few clicks and a matching page exists, compare that page's title and opening lines with the label. If a label has high impressions and no matching page, plan a page. If impressions are low, re-check next month before spending effort. Keep the brand label out of this rule.

Cross-checking with the Pages tab

A label drawn only from the words of a query is half finished. The Pages tab aggregates the same data by page, so filtering a query and switching tabs shows which URL appeared for it. The check answers three questions. Does the query meet the right page? Do several queries of one label gather on one address? Does a label have no home on the site at all?

What the cross-check returns

Each label ends in one of three states: matched correctly, matched to the wrong kind of page, or unmatched. A price query that lands on an explainer is the middle case. Practical suggestion: add a page of the right type and link it from the old one, then read the result in the same label table a few weeks later.

Two intents on one page

If one page tries to serve an information need and a price need together, look at what the first screen answers. A flavor page whose opening lines skip price and pack size serves the price label weakly. Put two or three sentences that answer the label's first question at the top and leave the detail after them.

From labels to an editorial calendar

Three columns are enough for a calendar: label, page type and finish date. Two measures set the order. One is the impression size of the label. The other is how weak the current page is. A label with big impressions and a weak page goes first.

Table: From label to calendar row

The success column ties production back to the label table, so a month later the same table reports what the calendar achieved. Without that tie, a calendar turns into a to-do list cut off from the analysis that produced it.

Where a listing page ends and a guide begins

Price and ingredient labels fit listing and product pages; question and best-of labels fit guides. For a small shop one page can carry two intents. As a page grows, readers find it harder to reach the part they came for, and clear subheadings are the light way to separate intents.

Reading a pivot: a vegan protein bar shop

Example scenario: a small shop sells vegan protein bars. It has a product category page, a few flavor pages and one blog post about protein sources. The owner exports three months of web queries, labels the top rows and builds the pivot. The numbers here stay out of the example; only the reading is shown.

Table: Example scenario, reading the pivot

Four jobs come out of five labels, and each job points at one page. Next month the owner repeats the export and checks whether the label that received work changed.

A one-page monthly report

A monthly report keeps the work alive. It fits on one page: a dated header, the pivot table, one example query per label and a single action for the month.

Table: Monthly report layout

Comparing months fairly

The report offers date comparison, but the help page has no statement on the details of its filter syntax, so a spreadsheet comparison is the simplest route: two exports, two pivots, side by side. Check both ranges cover the same number of days before comparing totals.

The first thirty days after labelling

Week one: take the label with the highest impressions and check whether the matching page answers it in its title and opening paragraph. Week two: draft the page for the label with no match. Week three: publish and inspect the new URL in Search Console. Week four: repeat the export and compare pivots. Tying the work to a calendar keeps the table from becoming a slide in a deck.

Small sites and large sites

A site with a few dozen pages can assign each label to one page. A site with thousands of listings adds a page group column and decides at group level. The rule stays the same: every label maps to something you can measure.

Splitting a label by device and country

The Performance report also offers device and country tabs. After the label table exists, you can use one of them to look into a single label. The aim is to see whether the same intent behaves differently across devices or countries. If the bulk of impressions for price-labelled queries comes from phones, price information needs to sit in the first view on a phone screen. That is a guess until you open the page on a phone yourself, which is the shortest check.

Which labels gain from a split

For a local label a country split adds little. For question and best-of labels, a device split hints at where readers consume the page. For a comparison label, desktop weight tolerates wide tables, while phone weight needs tables that stay readable on narrow screens.

Knowing where to stop

Each extra dimension cuts the table into smaller parts, and fewer impressions fall into each part. Splitting a low-impression label by device can turn random wobble into a false decision. Practical suggestion: split only high-impression labels, and confirm the pattern a month later before acting on it.

Keeping labels consistent across a team

With several people labelling, consistency becomes the issue. Have two people label the same hundred rows without talking, then compare. Rows where they disagree point to gaps in the vocabulary. Write those rows up as new rules and the second round agrees more.

When a rule changes

If a rule changes, comparisons with earlier months can break. Practical suggestion: date the vocabulary and relabel the previous month with the new rule before comparing. Otherwise a jump in the monthly report can reflect a changed vocabulary and not changed content.

A time budget

The first pass can take a few hours, while later months shrink because rules exist. Spend the first pass on building rules and later passes on new rows the rules do not cover. That split keeps the routine light enough to survive a busy month. Keep the vocabulary sheet, the rule list and the monthly pivot in one shared folder, with the date of the last vocabulary change in the file name. A new teammate can then reproduce last month's table from the export alone, and a handover does not restart the work from the first hundred rows. If the sheet only lives in one person's head or one person's laptop, the monthly routine stops the first time that person is away.

Ways labelling goes off course

Letting brand queries flatter the content results

Brand searches reflect awareness. Report them apart from the rest so a rise in clicks is not credited to content work. For a closer look at the report itself, see the Performance report guide.

Reading a label as a fact

A label is an estimate drawn from words. Use the table to order your work, not to settle arguments about what searchers want.

Letting the label list sprawl

Fifteen labels drift apart in use, while six or seven keep the table readable. Before adding one, ask whether it would lead to a different page decision.

Overlooking queries just off the first page

Queries that sit just past the first page deserve their own pass. The page two queries work list shows how to build one without repeating this labelling work.

Judging from one window of dates

A three-month window can hide seasonal swings; a bar brand sees different interest around the new year than in summer. Export two periods and compare the pivots before concluding anything.

Preflight list for the next export

Checklist: tick these off before you export and label.

• Is each label's rule written in one line, with brand as its own label?
• Is there a column for a second label?
• Is the search type set to web, with image and video kept apart?
• Does the date range have a comparison range of equal length?
• Are "~" and "-" values recorded as unknown?
• Is click-through rate computed from totals rather than averaged?
• Does every label have a page and a decision?

Frequently Asked Questions

Which Performance report dimensions help before I label anything?

The report offers queries, pages, countries, devices, search appearance and dates. Intent labels are your own classification layered on top.

How many labels is too many?

Practical suggestion: more than six or seven. Add a new label only when it would lead to a different page decision.

What limits an export of every query?

The report has an export button for chart and table data, and the API limits Performance report data to 50K rows per day per type per property.

Why keep brand queries apart from the rest?

Brand searches reflect awareness. Mixed into totals, they make content work look more effective than it is.

How is click-through rate worked out per label?

Divide total clicks by total impressions for the label, following the report's own definition, instead of averaging per-row rates.

How should "~" and "-" cells be treated?

The help page says values shown as "~" or "-" in the report will be zeros in the downloaded data, so treat them as unknown.

Conclusion

Labelling queries by intent turns a long table into a short list of decisions. Export the web data, apply six or seven labels with written rules, total impressions and clicks per label, and recompute click-through rate from the totals. Then give each label a matching page and a next step, and track the brand label on its own.

If the pages your labels call for are not yet written, our content writing service can help you fill them.