Analytics
Search performance
Google Search Console data for your site: what you rank for, which positions moved, and where a small change would pay.

The screenshot shows the Search screen for the Quillstone demo project on development data. The demo's Search Console rows are generated and its topic map was built from them, so its numbers illustrate the layout rather than what a real project will show.
What you see
Search performance, the
Search tab under Traffic in the sidebar, reads your Google Search
Console property: the searches that actually
surfaced one of your pages in a Google result. "Query" on this screen is
Search Console's own word for a literal search string typed into Google, not
this product's tracked prompts against AI engines; the import that fills this
screen asks Search Console for exactly the date, query, page and
country dimensions it defines
(api/services/integrations/google/reconciliation.py#import_gsc_daily_for_project).
The hero number is clicks for the selected window, with impressions,
click-through rate, and average position stepped down beneath it
(api/routers/integrations.py#gsc_search_performance). Click-through rate is
the window's total clicks divided by its total impressions, not an average of
the underlying rows' own rates, and average position is weighted by
impressions the way Google itself aggregates it (#_ctr, #_round2). The
backend keeps that figure to two decimal places; the screen itself shows one,
everywhere a position appears on it
(frontend/src/lib/format.js#formatPos). Because a lower position number is
the better ranking, the opposite of clicks or impressions, the position card
spells the direction out in words rather than leaving a signed number to
misread: "2.0 positions better" means the average position fell by two. The
click-through rate card gives its change in percentage points ("0.4 pt
better"), and either card reads "Unchanged" when the change rounds to 0.0
(frontend/src/routes/(app)/performance/search/+page.svelte#posPoints,
#ctrPoints).
A "Closest to page one" card beside the hero surfaces the top row of Quick
wins below, in whichever of its two lists is showing: among the searches
within striking distance of page one, the
one with the most impressions, not necessarily the one with the best
position. topWin reads quickWins[0], and quick_wins sorts its
candidates by impressions then clicks, using position only as the 4-to-20
admission filter described below
(frontend/src/routes/(app)/performance/search/+page.svelte#topWin,
api/services/integrations/google/gsc_search_metrics.py#quick_wins).
Leading with impressions points at the near-miss with the most search
demand behind it, even when a better-positioned search sits further down
the list. When one of your prompts already tracks that search, the card
links to that prompt's Google section ("See the prompt that tracks it");
otherwise company members and editors get "Track this as a prompt", which opens the
add dialog on Prompts with the search filled in
(frontend/src/lib/search-performance.js#trackHref).
When the window has missing days
Search Console data reaches this screen through a daily import, and a
newly mapped property, a paused import or Google's own two- to three-day
lag can leave days in the window with nothing imported. The screen counts
the days that have any imported row, in any country
(api/services/integrations/google/gsc_coverage.py#imported_days). When that count is more
than 3 days short of the window, a note under the title says so: "Search
Console has data for 5 of the last 28 days (Oct 6 to Oct 10). Totals cover
only those days."
(frontend/src/lib/search-performance.js#coverageNote,
#COVERAGE_SLACK_DAYS).
Every change on the screen is a comparison with the window before, and a
comparison where either window was barely imported would read an import gap
as a rise or a drop. So changes are shown only when each window has imported
days for at least 80% as many days as the other
(api/services/integrations/google/gsc_search_metrics.py#is_comparable,
#COMPARABLE_SHARE). A short previous window fails that test, and so does a
current window that is still catching up: on the 7-day window, Search
Console's two- to three-day lag often leaves the current week short, so
changes there are often hidden until the import catches up. When the test
fails, the hero's change chip and the cards' changes are left out, the
explorer's Change column is removed, and Movers lists nothing, showing
instead a note that names the window that is short, as in "No change is
shown: this period has Search Console data for 5 days so far, too few to
compare with the previous 7 days (7 days)."
(frontend/src/lib/comparison.js#comparisonGapReason). A change that is
withheld is never shown as zero.
Comparable windows can still be too small to compare. A change is shown only
when both windows hold at least 100 impressions
(api/services/integrations/google/gsc_search_metrics.py#CHANGE_MIN_IMPRESSIONS,
#change_has_volume). When the whole site falls short in either window, the
hero's click change and the cards' click-through rate and position changes
are hidden, with a note under the hero saying one of the two periods has
fewer than 100 impressions. The same test runs on every explorer row: a row
short in either window shows a dash in its Change column, and a line
under the table explains the dash. Below that floor, two clicks becoming five
would read as a 150% rise.
The daily trend draws clicks and impressions on their own separate scales,
so it shows shape rather than one shared magnitude. A day with nothing
imported is a gap in the lines, not a drop to zero; with a country chosen,
a day that was imported but had no searches from that country is a real
zero (api/services/integrations/google/gsc_search_metrics.py#build_daily_series).
The trend is drawn once at least 5 days in the window have data, and
otherwise replaced by a line saying there are not enough days yet
(frontend/src/routes/(app)/performance/search/+page.svelte#hasTrend,
frontend/src/lib/search-performance.js#daysWithData).
Where your impressions rank
Every query with a measured position falls into exactly one of four bands, in
a fixed order, labelled on screen as "1-3", "4-10", "11-20" and "21+", based
on that query's own impressions-weighted average position for the window.
Because that average is a continuous, often fractional number, the labels'
low edges are not exact: the cutoffs are position 3 or better for the first
band, then better than 3 up to and including 10, then better than 10 up to
and including 20, then worse than 20, so a query averaging 3.4 lands in
"4-10", not "1-3"
(api/services/integrations/google/gsc_search_metrics.py#_BUCKETS,
#position_bucket). A query with no position on record is left out of every
band and out of the total the bands divide by, so the bar and the list beside
it show a share of the impressions that had a ranked position, not a share of
every impression in the window (#position_mix). The fourth band carries its
own label on screen: "Position 21+, effectively unseen."
Where your impressions come from
Beside the position bands, a second bar splits the window's impressions by
kind of search, with the clicks each kind got
(api/services/integrations/google/gsc_search_metrics.py#segment_totals).
Each search is sorted in this order (#query_kind,
api/services/search_topics/plan.py#segment_key):
- URLs and site: checks: a search using an operator such as
site:,inurl:orintitle:(api/services/prompt_demand.py#SEARCH_OPERATOR), or a URL typed into the box, meaning it contains://or starts withwww.(api/services/integrations/google/gsc_search_metrics.py#URL_QUERY). Asite:check of your own domain counts here, not as your brand. - Your brand name: a search containing one of your own names: the
project's name, the bare label of its domain, and any alias or
sub-brand registered for your own brand
(
api/services/prompt_demand.py#_own_names). - Every other search. Without a topic map from the
Search plan, these are Other
searches. With one, they split by the relevance of the topic that
holds them: What you sell, Related reading and Not about your
business, and Not in a topic yet for the searches no topic holds.
Brand searches, URLs and
site:checks are never counted there: they keep their own segments above (frontend/src/lib/search-performance.js#SEGMENT_LABELS).
Searches for your name, a URL or a site: check are people already looking
for you, not new demand, which is why the explorer starts without them.
Quick wins and movers
Quick wins lists searches with an average position from 4.0 to 20.0, both
ends inclusive, and at least 10 impressions in the window, ranked by
impressions then clicks, capped at 10 rows
(api/services/integrations/google/gsc_search_metrics.py#quick_wins, called
with that floor and cap in api/routers/integrations.py#gsc_search_performance).
Searches for your brand, URLs and site: checks are left out, and so are
searches in a topic the Search plan judged not about your business. Each
row carries the same Tracked or Track link as the explorer (below).
Once the Search plan has a topic map, the card offers two lists: What you
sell, which it opens on, and All searches. What you sell applies the
same rules to the searches in a topic the plan's What you sell view shows: a
core topic or a Target, and not dismissed
(api/routers/integrations.py#gsc_search_performance, sell_quick_wins). It
is empty, and the card shows only All searches, until a map exists
(frontend/src/lib/search-performance.js#quickWinScope). The weekly report
uses the All searches list: when a report is
generated it asks for this screen's 7-day window ending that day and keeps
the first 5 rows (api/services/weekly_report.py#_assemble_quick_wins).
Movers diffs each query's clicks in the selected window against its clicks in
the window immediately before it (the two are adjacent, though not quite the
same width; see below for the exact day counts), keeping a query only
where one of the two windows reached 5 clicks, then splits the result into
gainers and losers, each capped at 8 rows
(api/services/integrations/google/gsc_search_metrics.py#compute_movers,
called with that floor and cap in api/routers/integrations.py#gsc_search_performance).
URLs and site: checks are left out; your brand's searches stay, since a
drop there is real news. A query that fell to zero clicks still shows up as
a loser rather than disappearing, because the comparison is a diff across
both windows, not a filter on the current one. Movers lists nothing when the
two windows' imported days are too far apart to compare (see above).
The explorer
Beneath the trend, four tabs: Topics, Queries, Pages and
Query → Page. Topics comes first and is the one that opens when the
project has a topic map; otherwise Queries opens. The last three each list
up to 100 rows ranked by clicks, with that row's own impressions,
click-through rate and average position alongside it; a full table says
"Showing the top 100 queries by clicks. Smaller ones are not listed." (or
pages, or query and page pairs). The one Change
column compares clicks against the prior window, and only when the two
windows are comparable and the row has the impressions described above; the
other figures carry no change
(frontend/src/routes/(app)/performance/search/+page.svelte#gscRows,
api/routers/integrations.py#gsc_search_performance).
On Queries and Query → Page, a switch reads Without brand and noise
(the default) or All queries. The first hides searches for your brand,
URLs and site: checks from the rows the server sent, so it can show fewer
than 100 rows, and the row count says how many were hidden
(frontend/src/lib/search-performance.js#visibleQueryRows). Each search
row links out in up to two ways:
- Tracked, when an active prompt of yours covers the search: every
content word of the search, at least two of them, appears in the prompt,
plurals folded together, the same rule a prompt's Google demand uses
(
api/routers/integrations.py#_tracked_prompts,api/services/prompt_demand.py#query_tokens_match). It links to the oldest such prompt's Google section and names up to 3 on hover. With no covering prompt, company members and editors get Track, which opens the add dialog on Prompts with the search filled in. - Topic, the Search plan topic that holds the search, linking to its card on the plan. A topic not about your business has no card, so its name is not a link.
Each page row (Pages and Query → Page) opens the page in a new tab and,
when an active prompt lists the page as its own, shows "Your page for:"
with the oldest such prompt (api/routers/integrations.py#_owned_prompts_by_page).
Its Cited by AI column counts the answers that cite that exact page as
your own, after both URLs are normalized the same way (trailing slash and
tracking parameters removed), whichever prompt the answer was for
(#_ai_citations_by_page, api/services/urls.py#normalize_url). It
follows the Explorer's rules for which answers count: completed answers
dated in the window's own number of days (28 by default) ending yesterday,
one day fewer than the Search Console figures, which run through today (see
below), with one retained answer per prompt target, engine and day
(api/services/execution_reads.py#project_citations). On Query → Page the
count is the page's, so every search landing on one page shows the same
number.
Topics lists the Search plan's topics: their impressions, clicks, best
position and most shown page, and their citation rate in AI answers (a
share of answers, unlike the page count above), all over the window's own
number of days (28 by default) ending yesterday, one day fewer than the
Search Console figures above, which run through today, in its country; then
their state and their monthly searches. They are ranked by impressions,
with the topics not about your business grouped last and a Not in a
topic yet row for the other searches no topic holds, brand searches,
URLs and site: checks aside. Each topic links to its card on the plan,
except the topics not about your business, which have none
(api/services/search_topics/plan.py#read_topics,
frontend/src/lib/search-plan.js#searchTopicRows). A line above the table
says so: "Numbers for the selected window and country. The search plan
uses the last 28 days for every country." A topic's state here can
therefore differ from the one on the Search plan. Monthly searches is
the topic's total over the countries looked up, or, with the country chip
set, that country's own figure, and "–" when it was not looked up there
or had no data (frontend/src/lib/search-plan.js#monthlySearches). See
Search plan for how
each number is computed. Without a topic map, the tab offers a link to
build one on the Search plan instead.
The country chip
The country chip in the filter bar applies to this screen
(frontend/src/lib/nav.js#HUBS). Each half of the screen filters by what
it knows. Search Console numbers count searches from people in that
country, the country Search Console records for each search
(api/services/integrations/google/gsc_search_metrics.py#gsc_country_filter).
AI numbers, the Cited by AI counts and the Topics tab's citation rates,
count answers to prompts aimed at that country, and a topic counts only
its prompts with a target in that country. The Tracked, Track and "Your
page for:" links are not narrowed by country. The note about missing days
still counts imported days in every country. When the chosen country has
no searches in the window, the screen says so and offers to show every
country.
The window, and how it catches up with Google
Left alone, this screen opens on the last 28 days. The date range filter at
the top of the page switches it to 7, 28 or 90 days, carried in the URL as
?days=; any other value in the URL reads as 28, so 90 is the widest window
this screen shows, and the API behind it accepts only those three windows
(frontend/src/routes/(app)/performance/search/+page.server.js#windowDays,
frontend/src/lib/filters.js#readDays,
api/routers/integrations.py#gsc_search_performance).
Every change figure, the hero's clicks, each explorer row, and the queries
feeding Movers, compares that window against the window immediately before
it: the two are adjacent, with no gap and no overlap, but not quite the same
width. The current window runs from window_days days ago through today,
both ends inclusive, which is window_days + 1 days; the prior window covers
exactly window_days days, ending the moment the current one starts. At the
default 28-day setting that is 29 days of current data against 28 days of
prior data, one day wider; the same window_days + 1 is why the trend chart
above draws 29 daily buckets for that same default, not 28
(api/routers/integrations.py#gsc_search_performance).
The import behind this screen does not treat Google's reporting lag as a
special case: on every scheduled run it re-pulls the last 7 days plus today,
8 calendar days inclusive, and overwrites what is already stored, so a day
that first lands incomplete gets corrected on a later run instead of needing
separate lag-handling logic. The first time a property is mapped, the import
instead pulls 181 days, from 180 days ago through today, both ends
inclusive, in one run split into 26 chunks of up to 7 days each rather than
a single request
(api/services/integrations/google/reconciliation.py#import_gsc_daily_for_project,
#_IMPORT_WINDOW_DAYS, #_BACKFILL_WINDOW_DAYS, #_BACKFILL_CHUNK_DAYS,
#_backfill_windows). See
Troubleshooting: Search or Site traffic is empty
for exactly when the recurring import runs and what a window that lands
before your most recent import means for this screen.
Before you see data
Search performance needs a Google Search Console property mapped to this
project. The mapping belongs to the project rather than to you: a property
mapped through any member's Google connection counts, so a teammate who never
connected Google still sees the project's search data, while a mapping whose
connection was disconnected, or whose property Google no longer lists, does
not count
(frontend/src/lib/server/google-status.js#loadGoogleState,
api/routers/integrations.py#project_google_status). With nothing mapped,
the screen prompts you to connect Search Console when your own account is not
connected, and to map a property when it is; mapped with no impressions
recorded yet for the selected window, it says so and points at the mapping to
check, rather than showing an empty chart
(frontend/src/routes/(app)/performance/search/+page.svelte#gscHasData).
When the report request fails, or the Google connection check itself fails,
the screen says it could not load the data and offers Retry, instead of
reading the failure as no search data or as not connected.
Related
- Search plan: these searches and your tracked prompts grouped into topics, each with a state and, where there is one, the next step
- Google rankings versus AI citations: why ranking well here does not by itself mean an AI engine cites you
- Troubleshooting: what an empty or stale-looking screen usually means
Last verified 2026-10-14