Signatures

A signature

A paid click on this screen is the first appearance of a Google Ads gclid or a Microsoft Ads msclkid on your site (tag your ad links). A later appearance of the same id is the ad link opened again — a trait of the original click, not a click of its own. Paid sessions that carry no such click id, including every session tagged by UTM alone, are not counted here, so this screen's Paid clicks can be lower than the Overview's.

A signature is the set of paid clicks of one campaign, in one 7-day block, that share one value of one keyed trait, when that value is at least 10 times more frequent than usual for the same audience and there are at least 5 such clicks. Three traits are keys:

The usual rate is how often the trait appears among the 28 days of clicks before the block, for the click's audience: the same ad network, operating-system family and surface the ad ran on (Search, Display, YouTube, Feed, the Google app, or no recognized referrer). When that audience is small, the screen names the wider population the rate was read on, beside the rate. Expected is the count the trait would reach at that rate over the campaign's clicks in the block.

The campaign is the signature's scope, not its name. Clicks that arrived without a campaign tag are grouped as one campaign, "Untagged" — although the Campaigns directory gives untagged clicks no row. When two groups of the same block and campaign overlap, one signature carries the name and the other appears under it as "Also …". The same campaign, trait and value in two adjacent blocks is one signature spanning both.

A configuration

A configuration is a set of observed browser settings, shown as an opaque 8-character label such as 50127444. It is never a key of a signature. It is listed on its own as a configuration group when over the period it carries at least 5 paid clicks from at least 2 identities — visitor ids (Visitors) plus sessions whose browser blocked storage — and is both:

• rare — seen on at most one non-paid visitor of your site over the period and the 28 days before it;
• coherent — its clicks come from at most 2 cities, one browser clock and one language. A click with no city, no clock or no language counts as a value of its own.

A group is named by its label — "Configuration 50127444". When all of the above holds, its What we infer panel ends on Probably the same device: the one place the product says it, beside the rule that earns it.

The pages

Signatures sits in the rail's Traffic group, after UTM tags and before Visitors, with three pages: Signatures, Configuration groups and Rules. The first two take the period picker (see Time periods), which selects the 7-day blocks shown. Each of them is a Reading card with figures, beside a Breakdown card whose one tab states the rule, then a table.

When this month's event allowance stopped the read inside the window, a line under the cards says where the numbers stop; the clicks after that point were collected and are kept (plans and billing). When it stopped the read before the window opened, nothing was read, and the figures print a dash rather than a zero.

Reading

On Signatures the card is titled "Signatures", with the block's first and last day, or the number of blocks and the span they cover ("4 blocks · Sep 1 → Sep 28"). Three figures:

Two doors at the foot open the other pages: "Configuration groups" and "Rules".

Breakdown

The card is titled "What decides the reading" and has one tab, Signature rule, one fact a line. Nothing in it is editable; it states what the page applies. A link opens A signature. When the page reads one of your own rules, the tab carries the rule's name instead, and its lines are the rule's own — see A rule's results.

The Signatures table

The bar is titled "Signatures", and its first line, in every state, is the rule: Groups of paid clicks in which one trait is at least 10 times more frequent than usual for the same audience, over the 28 days before. Sorted by number of clicks. There is no sort control, and no ratio column: observed and expected are two plain numbers, and the reader divides.

On a phone each signature is a card with the same facts. The footnote names the three traits compared and says what Expected is.

When the period has paid clicks but no signature, the result reads "No unusual signature in this period.", then one sentence with the highest group that stayed under the line — "No trait reached 10 times its usual level on 5 clicks or more; the highest was 6 clicks with … and the language set to English, on campaign …, for 2.1 expected." When nothing could be compared, the sentence says why instead: either your site's data starts too recently for a usual rate ("No usual rate to compare with yet: this site's data starts …"), or no campaign carried 5 clicks or more with one value of a trait in a block that had a usual rate. When the allowance stopped the read, the card says "Nothing to read before …" or ends the statement "before" the cutoff. With no paid click at all: "No paid click observed in this period."

Configuration groups

The second page has its own Reading card, titled "Configuration groups" ("rare and coherent"), with four figures: Paid clicks, Configuration groups, Inside a signature and Unlinkable shown. Its Breakdown card has one tab, Configuration rule:

The table is titled "Configuration groups" and has two tabs, Groups and Unlinkable.

Groups

Lists the period's rare and coherent configuration groups that no signature already holds; a group with at least half of its clicks inside a signature is listed on that signature instead.

With no group, one line says which is true: "No rare and coherent configuration carried 5 clicks or more from 2 identities or more in this period." — or "No group outside the signatures." when every group sits inside a signature. The footnote says what Identities counts, and that mobile addresses locate approximately and hosting and VPN addresses at their exit — a city count on a mobile network is weak evidence.

Unlinkable configurations

The Unlinkable tab names the period's most-clicked configurations that fail the rule, among those carrying 5 clicks or more.

An empty tab says so in one line: "No unlinkable configuration to list …". Opening a row shows the configuration's clicks in the shape of a configuration group (One signature); its What we infer panel ends on the row's reason, so the screen never calls those clicks linked. A row opens nothing when a single identity carries the configuration. The footnote says "visitors" here means all of your site's traffic, not only paid clicks. On Free, configuration labels are hidden by the plan.

Rules

The third page lists the rules this site's Signatures are read with: the built-in rule first, then the site's own rules, oldest first. The bar is titled "Rules", with the count beside it ("4 · 1 built in"). The page takes the period picker, which only changes the Groups column.

Owners and admins write the site's rules and flip the switch; members read the page as it is. The footnote states when detection reads and whom it emails.

A rule

A rule of your own is a name in your words (up to 60 characters), a scope, one to three conditions that must all hold, and how strict:

The rule's name is yours and is shown as you wrote it; nothing else in a rule's results changes its words. While you write, the editor shows what the rule would find in the period — groups, paid clicks and expected — before it is saved.

Conditions

A click whose country or operator is unknown meets neither is nor is not: an unknown is never counted toward a group. Surface, referrer, configuration and hour are not conditions — the evidence for leaving them out is in What this screen does not do.

A rule's results

A rule's name or its Groups count opens the Signatures page read with that rule, for the same period. The page keeps its shape; three things change:

• The Breakdown tab carries the rule's name, and its lines are the rule's own: Scope, Conditions, At least ("3 times usual · 5 clicks"), Blocks, Usual rates. An "All rules" link returns to Rules.
• The table's first line is the rule in a sentence: Groups of paid clicks in which the network class is hosting, VPN or Tor at least 3 times more often than usual for the same audience, over the 28 days before, on 5 paid clicks or more. Sorted by number of clicks.
• Each row is named by the rule's conditions rather than by one trait.

Read a row as you read any signature: Clicks beside Expected, the count the same conditions would reach at their usual rate in the same audience over the campaign's clicks in the block. A rule never changes how usual is measured — the audience, the 28 days before each 7-day block and the expected count are the built-in rule's — so a group under your rule is held to the same comparison. The scope only narrows which paid clicks are grouped; the usual rate is still read on the whole audience. Opening a row gives One signature, with its three panels and its members.

With no group in the period the result reads "No group under this rule …", then the highest group that stayed under the line, as for the built-in rule.

Automatic detection

A rule switched to Automatic detection is read every morning (from 06:00 UTC) on the 7 days that ended the day before, in UTC. When it finds a group that was not already found, the workspace's owners and admins get one email:

One email per new group, not one per day: a group that grows on the following days, or runs on into the next block, is the same group and is not mailed again. The email states what was observed; what the clicks have in common, what can be inferred and what is not known are on the group's page. Each recipient can turn these emails off under Automatic detection in Notifications; the switch on the rule turns them off for everyone.

Rules and plans

On Free the Rules page and the editor open — try the fields and read the preview — and the editor's Save button becomes the way to the plan that writes rules (plans and billing). When a site holds more rules than its plan includes, after a change of plan, nothing is deleted: every rule stays listed and its results stay readable, all of them read-only and with detection stopped, until enough are deleted to fit. Which ones to keep is your choice; each kept rule's detection switch then comes back as you left it.

One signature

Opening a row shows a signature, a configuration group or an unlinkable configuration — the last two in the same shape. Every count on this screen is whole on every plan; what a plan caps is the list of members at the bottom (plans and billing).

The header names the group as its row did: a signature by its trait ("Browser clock on UTC · English"), a configuration group by its label ("Configuration 50127444"). A back link returns to the list for the same period, and a signature carries the Export control. The figures follow: Clicks; Expected, on a signature only, beside Clicks and never divided by it; Addresses; Visitor ids, with the sessions whose browser blocked storage; Networks — the distinct connection providers, not ad networks; and Conversions. A group whose block starts before the earliest day your plan reads is drawn from that day and says so. A group that ends before that day cannot be opened, and the screen says so. A link to a group this site does not hold, or a read that did not come back, says that instead of showing an empty group.

Details is a column beside the reading holding one card of the group's facts:

Timeline ("Clicks by hour", hours in UTC) places every click of the group on its block. Taller marks are hours with more clicks. A click whose session reached the primary conversion is ringed (Conversions).

Facts and inferences are three panels, always all three, each a list of short lines: a label on the left, its value on the right. A panel with nothing to add says so.

• What we know — counts only: the clicks and their span in Paris time (the one place besides the Members column where this screen leaves UTC; see Time periods), the campaign, the shared trait, the addresses, visitor ids and connection providers, the surfaces the ads ran on, where the clicks landed, how many sessions ended after one event, the conversions, how many links were opened again from another address, and how the clicks split across configurations.
• What we infer — comparisons, with the base rate always printed: the reference window, then one line per population read ("Usually · Google on Android — 0 of 1 275"), then "Expected at that rate — 0.5 this week" beside "Observed — 47". When a population was too small and a wider one was read instead, a Fallback line names it. Every other trait shows its count with its usual rate.
• What we don't know — stated, never empty: who clicked and how many devices or operators are behind the clicks; whether the ad network already set these clicks aside, which your ad account knows and Intelligence/Ads does not; what they cost you; where the clicks really came from, since hosting and VPN addresses are located at their exit and mobile addresses approximately; and that clicks with different settings are not in the signature. The last line counts how many clicks came from an address or a visitor id that ever converted on your site.

A configuration group's three panels read its own rule. What we know: how many days, connection providers and cities its clicks spread over, its campaigns and its visitor ids — or that storage was blocked on every session — then the surfaces, the landings and the links opened again. What we infer: how many non-paid visitors of your site showed the configuration and over how many cities, clocks and languages its clicks spread, ending on Probably the same device or on Cannot be linked with the reason. What we don't know: who clicked and whether one device or several share the configuration, what they cost you, and Converted before.

Traits is a table of every trait as it stands in this group, beside its usual rate:

Configuration groups appears on a signature that holds nested groups. Its Nested groups panel lists each with its label, clicks, identities, cities and the days it was seen; each opens on its own.

Members lists every click of the group in the order it arrived:

On a plan that caps identity lists, the first rows are shown and the boundary states how many more clicks the period holds; the counts above already include them. On an uncapped plan the list stops at 500 rows and says so.

Export downloads a signature's members as CSV or JSON on plans that carry export (export): the same columns as a connection's export, plus the browser clock beside the time zone the address resolves to, the language, the referrer host, the surface, and when and from where the click was reopened. A configuration group has no export.

What this screen does not do

• It does not grade. There is no score, no severity and no tier. No color carries meaning: a hosting connection wears the same neutral mark it wears everywhere. The ratio is never a column and the list is never sorted by it.
• It does not say who clicked. A signature says that clicks share a setting, not that they share a hand. The one sentence that reaches further — Probably the same device — appears only in the What we infer panel of a rare and coherent configuration group.
• It never keys on a configuration, a surface, a referrer or an hour. The built-in rule keys on the three traits above, and a rule of your own on the five traits listed under Rules; everything else is described on every group and compared on none. Campaigns and the ad network are a rule's scope, never a condition.
• It does not give an all-clear. The empty state says what was read and compared, and names the highest group under the line. Nothing here says a period is clean.
• It does not alert on its own. The one email this screen sends is automatic detection, on a rule you wrote and switched on.
• It offers no action. There is no button to exclude addresses and no claim to send to the ad network. The addresses are in the member list and in the export; whether to act on them is yours.