Google Ads Report Segments: 12 Breakdowns That Change Your Conclusions
The most common cause of a bad decision in Google Ads isn’t bad data — it’s averaged data. A $42 campaign CPA is not a fact about the campaign; it’s the midpoint between mobile traffic at $110 and desktop at $19. Until the number is split, you’re optimising for a user who doesn’t exist.
The tool that splits it is Google Ads report segments — a built-in mechanism that expands any table row into several rows by a chosen attribute. Below: the full documented list, the 12 breakdowns that most often reverse a conclusion, the limitations nobody mentions until you hit a wall of dashes, and separately what changed in Google Ads reporting in September 2026.
Applying a segment: three steps
- Use the page menu to navigate to the table you want to segment.
- Select the segment icon and pick the data you want to isolate.
- To remove it, select the segment icon again and choose None.
Google’s own caveat matters here: your segment choices vary depending on which table you’re viewing. Some segments appear on almost every tab, others only on particular tabs, and in some cases you need to download a report to apply a segment at all. A missing option in the dropdown is a property of that table, not a bug.
The complete list of Google Ads report segments
| Category | Contains |
|---|---|
| Ad destination | Where the ad sends the user (requires setup) |
| Brand lift type | Type of Brand Lift study |
| Click type | What was clicked |
| Conversions | Conversion action, Conversion category, Conversion source, Ad Event Type, Days to Conversion, Conversion Adjustment, Days to Conversion or Adjustment, New vs Returning Customers, Value rule adjustment |
| Device | Device type |
| Format | Ad format |
| Network (with search partners) | Network, split out by search partners |
| Time | Day of the week, Day, Week, Month, Quarter, Year, Hour of day |
| Top vs. Other | Top of page versus other positions |
12 breakdowns that most often change the answer
1. Device
The first cut in any diagnosis. A two-to-three-times CPA gap between mobile and desktop is normal, not anomalous, and almost always means you have two different landing-page realities. Bid adjustment decisions come after this split, never before — the mechanics are in the guide to device bid adjustments.
2. Network (with search partners)
Separates Google Search proper from search partners. Behaviour differs enough that identical CPCs can produce very different junk-lead rates. If partners deliver volume without revenue, you turn them off — but only after seeing it in the split rather than assuming it. More in the piece on the Google search partners network.
3. Top vs. Other
Shows what share of impressions landed at the top of the page versus everywhere else. The practical question it settles: are we overpaying for the top slot, or failing to reach it at all? One limitation to know — search partner data has not distinguished Top from Other since October 2020, so partner traffic tells you nothing in this split. Read it alongside impression share and auction insights before touching bids.
4. Click type
Breaks clicks out by what was actually clicked: the headline, a phone number, a sitelink, an asset element. It exposes the case where half the budget goes to clicks on an element that never leads to the goal action. One quirk: a single impression can relate to several click types, so in a downloaded report Google replaces summary totals with a dash to prevent you from double-counting.
5. Hour of day
Not for a pretty chart — for one specific question: do your peak spend hours line up with your peak conversion hours? A mismatch here is the cheapest saving available in most accounts. What to do with the answer is in the guide to budget pacing and the 30/4 rule.
6. Day of the week
The same question at weekly scale. In B2B and service businesses, the gap between Tuesday and Saturday is routinely 2×, and fixing the schedule moves numbers faster than any ad copy rewrite.
7. Day / Week / Month / Quarter / Year
Primarily a diagnostic set: when exactly did performance drop? The limitation everyone trips on — the Day segment is only available when the date range is 16 days or less. For longer periods you get Week and Month.
8. Conversion action
Essential in any account tracking more than one conversion. It answers what the headline number is made of: form fills, calls, or the micro-conversions someone added “to help the algorithm learn.” Until you apply this split, a conversion report proves nothing.
9. Conversion category and Conversion source
Source is especially useful when data is imported from outside the platform: it shows which conversions came from the website, which from offline import, and which from other connected sources.
10. Days to Conversion
Days elapsed between click and conversion. This is the split that explains why yesterday’s numbers always look worse than reality, and why decisions made on yesterday’s data are systematically wrong. Window mechanics are covered in the piece on conversion windows and conversion lag.
11. New vs Returning Customers
The split that most often reverses a ROAS conclusion. A campaign with excellent ROAS can be running almost entirely on existing customers — recording growth rather than creating it. For a business paying for acquisition, those are different outcomes.
12. Conversion Adjustment and Value rule adjustment
Show how conversion adjustments and value rules changed the raw numbers. If you import cancelled orders or apply value rules by geography and customer type, without this split you’re looking at post-processed output with no view of what caused it.
Three segments carry access conditions worth noting: Format is available for Demand Gen and most video campaigns (excluding Video Action Campaigns) and can be applied via the segment icon or in Report Editor; Ad destination requires setup — Firebase or a third-party SDK, connector pings, a GA4F or third-party account link; Brand lift type is only available to accounts with a Google account representative.
Why the table fills with dashes
The most common complaint is “I applied a segment and half my columns went blank.” That’s intended behaviour. Conversion-based segments (Conversion action, Conversion lag) work only with the Conversions, All conv. and Conversion value columns. Every other column shows a dash, because the values in them can’t be meaningfully attributed to a single conversion action.
The second dash scenario is Click type in a downloaded report, where Google replaces summary totals to avoid double counting. Neither case is worth a support ticket.
What changed in reporting in September 2026
This needs an exact frame. A reporting-connector vendor documented the changes below. They describe data returned by the Google Ads API and date to a rollout on September 21, 2026. That is the API and export layer, not necessarily the interface. If you build reporting in BigQuery, Looker Studio or through a third-party connector, verify on your own stack; if you live entirely in the UI, you likely won’t notice.
| Field | Before | After |
|---|---|---|
| Network (Performance Max) | A single Cross-network value | The specific channel: Search, Search Partners, YouTube, Display, Gmail, Discover, Maps |
| Asset performance (Search and Display) | Labels Pending, Learning, Good, Best, Low | Field returns empty; Performance Max and Demand Gen unaffected |
| Click type on asset reports | Available | Unavailable on Customer asset, Campaign asset and Ad group asset reports; unchanged elsewhere |
Consequences of the Network field change
The consequences of the first row are not cosmetic:
- Filters on Cross-network now return an empty result — with no error. The report simply shows zero, which is easy to mistake for a traffic collapse.
- One Performance Max row can expand into as many as seven rows, changing export volumes and totals in any dashboard that sums rows naively.
- Channel-level data exists only for dates from June 1, 2025 onward. Earlier rows return Cross-network, later rows return the channel name. Any report spanning that boundary blends two schemas.
- Channel-level Performance Max data is available only through the API and cannot be reconciled against the same breakdown in the Google Ads interface. Trying to match it row by row wastes effort.
We covered the underlying PMax channel reporting separately in the piece on Performance Max channel-level reporting, and the warehouse side in the guide to Google Ads BigQuery Data Transfer reporting.
The second row is quieter but more painful for anyone who automated creative monitoring: if you had a rule or script watching asset performance labels on Search and Display campaigns, it now receives an empty value, and grouping or filtering on that field stops returning those rows entirely. How to read asset quality where the labels still exist is in the piece on Ad Strength and RSA asset performance.
Four ways people misread a segmented table
Drawing a conclusion from a row with no volume
Segmentation splits rows, and each one holds less data than before. A row reading “tablet, Tuesday, top of page” with four clicks and one conversion says nothing at all — but it looks authoritative, complete with its own CPA and CTR. The rule: before concluding anything from a row, look at the absolute numbers inside it, not the percentages.
Splitting on one attribute when two causes are at work
You split by device, see expensive mobile, cut mobile bids — and the real story was that mobile traffic arrived mostly at night, when conversion rate drops across every device. A second split by hour of day would have shown it immediately. If the first split gives you a suspiciously clean picture, test it with a second one.
Comparing segments with unequal bases
A segment carrying 5% of impressions and one carrying 70% can’t be compared on relative metrics as equals. The small segment will almost always look either anomalously good or anomalously bad, purely because it holds fewer observations.
Treating a finding as permanent
You run the split once and keep the conclusion forever. Segment behaviour drifts, though — seasonality, auction shifts, platform changes. A diagnostic split has a shelf life, and for consequential decisions it’s worth re-running at least quarterly.
When segments aren’t enough
Segments have a natural ceiling: they work inside one table and one account hierarchy. The moment the question leaves those bounds, you need a different instrument.
| Task | Right tool |
|---|---|
| Expand a row by an attribute | Segments |
| Build a derived metric or formula | Custom columns |
| Combine data across several accounts | Manager-account reporting or export |
| Join with CRM data or stock levels | Warehouse export |
| Retain history beyond standard retention | Warehouse export |
| Establish cause and effect | An experiment, not a report |
That last row is the most important and the most ignored. A segment shows correlation — “performance is worse in this slice.” It does not prove the slice attribute caused it. Only an experiment does, and that’s a separate tool with a separate method, covered in the piece on Google Ads experiments and Experiment Center.
Which split answers which question
| Question | Segment |
|---|---|
| Why doesn’t the average CPA resemble anything real | Device, then Network |
| Where does the money go during the day | Hour of day + Day of the week |
| What is the conversion number made of | Conversion action, then Conversion source |
| Why does recent data always look worse | Days to Conversion |
| Are we growing or serving existing customers | New vs Returning Customers |
| Are we overpaying for the top slot | Top vs. Other |
| What are people actually clicking | Click type |
| When did the drop start | Day (range ≤16 days), otherwise Week |
A segment isn’t a way to make a report prettier. It’s a test of whether the “average campaign” you think you’re managing exists at all. In half of accounts, the first split shows that it doesn’t.
What segments can’t do
Segments expand existing rows; they don’t create new metrics. If you need a derived figure — share of conversions by type, margin-adjusted ROAS, a custom formula — that’s the job of custom columns, covered in the guide to custom columns and Report Editor. Google documents the segment list and column compatibility but does not state a limit on how many segments can be combined, nor how segments behave in scheduled reports — verify both against your own account.
And the thing most teams forget: a segmented table is a diagnostic instrument, not a reporting format. A table broken out by seven attributes is unreadable and adds nothing but noise to a weekly call. Apply a split to answer a question, then remove it once you have the answer.
None of this pays off on an account with no history — segments need data volume, and restarts reset it. If a campaign lives for weeks rather than quarters, there is nothing to split, which is why stable infrastructure such as Google Ads agency account rental gets settled before the reporting question, alongside a working reference like the Google Ads guide for the people building the reports.
Two companion reads that use these splits in anger: the breakdown of the audience retention curve in video campaigns, where device segmentation decides whether the curve means anything, and the guide to running Google Ads on a small budget, where the hour-of-day split is usually the cheapest saving available.
FAQ
How do I apply a segment in Google Ads?
Navigate through the page menu to the table you want, select the segment icon and choose the data to isolate. To remove it, select the segment icon again and choose None.
Why isn’t the segment I need in the list?
The available set varies by table. Some segments appear almost everywhere, others only on specific tabs, and some require downloading a report to apply. That’s documented behaviour, not an interface fault.
Why do columns show dashes after I apply a segment?
Conversion-based segments such as Conversion action and Conversion lag only work with the Conversions, All conv. and Conversion value columns. Other columns have no valid value under that split, so Google shows a dash.
Why are summary totals missing with the Click type segment in a download?
A single impression can relate to multiple click types. To prevent double counting, Google replaces summary totals with a dash in downloaded reports.
Why is the Day segment unavailable?
The Day segment only appears when the selected date range is 16 days or less. Use Week, Month or Quarter for longer periods.
What changed in Google Ads reporting on September 21, 2026?
According to a reporting-connector vendor, three API-level fields changed: Network for Performance Max now returns the specific channel instead of Cross-network, asset performance labels are no longer returned for Search and Display campaigns, and Click type became unavailable on customer, campaign and ad group asset reports.
What happens to existing Cross-network filters?
They return an empty result with no error. The report shows zero, which reads like a traffic collapse — audit filters in custom dashboards before escalating.
From what date is Performance Max channel data available?
From June 1, 2025. Earlier rows return Cross-network, so any report crossing that date mixes two different data schemas.
Can PMax channel data from the API be reconciled with the interface?
No. The channel-level Performance Max breakdown is available only through the API and does not reconcile row by row against the same breakdown in the Google Ads interface.
Why is Top vs. Other useless for search partners?
Search partner data has not distinguished Top from Other since October 2020. The split remains informative only for Google Search itself.
How are segments different from custom columns?
Segments expand existing rows by an attribute but create no new metrics. Derived figures and custom formulas belong to custom columns and Report Editor.
How many segments can I combine at once?
Google does not document a limit on combined segments, nor segment behaviour in scheduled reports. The practical limit arrives first: a table with three or more splits stops being readable.
Should a segment stay applied permanently?
No. A segmented table is a one-off diagnostic. For recurring reporting it’s too noisy — apply the split to answer a question, then take it off.
