Discover Performance: complete guide

Interpret Discover clicks, impressions, pages, countries, devices, and volatile traffic patterns.

Automatable now Intermediate

Search Console's Discover Performance report shows how often content from a property was scrolled into view in Google Discover, how often users clicked it, and the resulting CTR. It can group this evidence by canonical page, country, date, and supported Discover appearance.

Discover is an interest-based, proactively served feed—not a list of answers to explicit search queries. Its traffic is therefore more volatile than keyword-driven Search traffic, and the report does not contain a query or average-position metric.

This guide reflects Google's documentation on July 18, 2026.

What qualifies for Discover

Content is automatically eligible for Discover when:

Google has indexed it.
It meets Discover content policies.

No special tag or structured data is required. Eligibility does not guarantee selection, impressions, or sustained traffic.

Discover can surface many content formats and topics, including older content when it remains helpful and relevant to an individual's interests. Do not define the eligible inventory as “articles published today” or assume that an older page appearing again is a reporting error.

Why the report may be missing

The Discover report appears only after a property reaches Google's minimum Discover-impression threshold. Google does not publish the numeric threshold.

A missing report can mean:

The property recorded no Discover impressions.
It recorded impressions but not enough to make the report visible.
The selected URL-prefix property does not contain the canonical URLs receiving credit.
The relevant period falls outside the available data window.
The site's pages are not indexed or are ineligible under Discover policies.
The property or content is new and reporting has not populated yet.

Google's introductory help describes report eligibility in terms of a minimum number of impressions in the last three months. The detailed report's default date range is also three months, while available Discover history can extend to 16 months.

Do not add markup, resubmit pages repeatedly, or change titles simply to make a navigation item appear. First verify the property, canonical ownership, page indexing, date range, and whether any Discover traffic exists through the Search Analytics API or bulk export.

Metrics and counting rules

The report exposes three metrics:

Metric	Meaning in Discover
Impressions	A Discover result from the property was scrolled into view, as a standard item or inside a carousel.
Clicks	A user clicked the Discover result. Shares and other actions are not clicks.
CTR	Clicks divided by impressions.

There is no average position. A personalized, continuously scrolling feed does not provide the same stable ranking context as a conventional results page.

Impression counting

Only one impression is counted for the same result in a session even if the user scrolls past it, leaves it, and scrolls back. If two distinct results from the same property appear in the same Discover feed, each can receive its own impression because Discover data is aggregated by page.

An item inside a carousel must be scrolled into view to count.

Click counting

A click records navigation through the Discover item. Sharing, saving, or another non-navigation action does not count as a click.

Web Story behavior

Web Stories have an additional counting rule. Search Console can count the story clicked from Discover and each additional story opened from that initial story. It does not count repeated views of the same story twice within the session.

Keep this behavior in mind when comparing a Web Story with a standard article or reconciling Search Console clicks with a one-session analytics model.

All data is page-aggregated

Discover reports data by page rather than using the property aggregation common in parts of the Search Performance report.

This means:

Two visible results from the same property can each count an impression.
Page-level data is assigned to the canonical URL.
A duplicate or alternate URL can show zero while its canonical receives all clicks and impressions.
The data appears only in the Search Console property that contains the canonical URL.

If a user lands on an alternate URL after a click but Google credits a canonical in another property, the alternate property's Discover table will not inherit that data.

Diagnose property gaps through canonical ownership

When a directory or subdomain property shows fewer pages than expected:

Inspect affected URLs.
Record Google's selected canonical.
Check which property contains that canonical.
Compare the canonical property rather than adding duplicate page totals.

Migrations, syndication, cross-domain canonicals, AMP arrangements, and protocol changes can all shift where Discover evidence appears.

Dimensions available in the interface

The current Discover report can group its table by:

Page.
Country.
Date.
Discover appearance.

It can filter by date, page, country, and appearance type. It does not offer a query dimension because Discover is not triggered by a user-entered query in the way Search results are.

Country

Country is where the content was viewed or clicked, not necessarily the language, market, or publication edition of the page.

Use country changes to investigate:

Topic interest by market.
Localization and headline fit.
Availability of Discover or a specific feature.
Changes in audience mix that alter total CTR.

Do not infer language directly from country. Combine Search Console with page locale and first-party analytics where that distinction matters.

Date and time granularity

The report supports hourly, daily, weekly, and monthly granularity in applicable views:

Hourly is available in the 24-hour view and contains preliminary data.
Daily helps investigate individual spikes and drops.
Weekly and monthly views smooth daily volatility for trend analysis.

Weeks run Sunday through Saturday; months are calendar months. Partial first or last weeks can make a comparison misleading when the selected range does not align with those boundaries.

Discover appearance

Current appearance types are:

UI name	Meaning	API value	Bulk-export field
AMP articles	An AMP page, including AMP story results	`AMP_TOP_STORIES`	`is_amp_top_stories`
News Showcase	An article in a publisher-curated News Showcase panel	`NEWS_SHOWCASE`	Not supported as a bulk-export appearance field
Web Story	The visual Web Story subset of AMP content	`AMP_STORY`	`is_amp_story`
Videos	A video from the site that appears in Discover	`VIDEO`	`is_video`

The appearance tab and filter show only types for which the property has recorded data. A missing appearance option is not proof that the markup or content type is invalid; it means no reportable evidence is available for that appearance in the selected context.

One result can qualify for more than one appearance classification. Do not add appearance-row totals and assume they form mutually exclusive shares of all Discover traffic.

Thresholds hide rows, not their contribution to totals

Search Console applies a minimum-impression threshold to table rows. A row below the threshold can be omitted while its metrics remain included in site totals.

The chart can likewise omit low-volume daily points while the selected-period totals remain accurate and include them.

Consequences:

Visible page rows may not sum to the chart total.
A page can receive a small amount of traffic without appearing in the table.
Country or appearance subtotals can be incomplete.
“Other” cannot be reconstructed reliably as proof of specific hidden URLs.

Treat the total as the best property-wide measure and the table as a thresholded diagnostic breakdown.

The interface table is limited to 1,000 rows

The report truncates the table before display at 1,000 rows. Reversing the sort order does not reveal row 1,001 because that row never reaches the displayed table.

This is separate from threshold omission. A row can be absent because it is below the privacy/reporting threshold, outside the top 1,000 rows, or both.

Use the API or BigQuery bulk export for deeper analysis, but retain their own limits and anonymization rules. A spreadsheet export of the interface does not bypass the 1,000-row table limit.

Time zones and preliminary data

The 24-hour view displays hourly data in the browser's local time zone. Other report views use Pacific Time.

The newest data can be preliminary:

It appears as a dotted line.
It can change during collection.
Today and sometimes yesterday may be preliminary.
The 24-hour view includes preliminary hourly points by design.

Before reconciling with analytics or publishing logs:

Convert time zones consistently.
Compare finalized dates.
Align the definition of a click with the downstream session model.
Account for consent, blockers, redirects, and page-load failures in analytics.

Do not alert on small differences in an unfinished day.

Discover is not keyword Search

Discover selects content based on individual interests and activity. It is not a query-results report.

Therefore the report cannot show:

Queries that “triggered” an item.
A stable ranking position.
Search-volume demand for a keyword.
Whether a title ranks for a target phrase.

Analyze Discover through pages, topics, formats, countries, and time—not keyword positions. Search Performance remains the correct evidence source for query-driven web, image, video, and News-tab analysis.

Why Discover traffic is volatile

Google documents three broad sources of Discover fluctuation:

Changing user interests: people move toward or away from topics.
Changes in content types: Discover adjusts the kinds of content shown in the feed.
Google Search updates: Discover is an extension of Search and can change with Search systems.

Google also warns that traffic can change for reasons unrelated to a site's quality or publishing frequency. Treat Discover as supplemental to keyword-driven traffic, not a predictable guaranteed acquisition channel.

A spike is not a new baseline

One article can be recommended broadly for a short time and dominate the property total. Compare:

Median daily traffic, not only the peak.
Traffic with and without the top page.
Number of pages receiving impressions.
Top-page concentration.
Weekly or monthly cohorts by publication date.

A surge can represent one temporary audience-content match rather than durable site-wide growth.

A drop is not automatically a penalty

A decline can follow topic saturation, shifting interests, replacement by other content types, seasonality, a previous outlier ending, canonical changes, Search updates, or reporting anomalies.

Check Security & Manual Actions for an explicit Discover policy action, but do not label every decline a manual action or quality penalty.

A reliable drop investigation

1. Verify data integrity

Confirm the property and date range.
Exclude preliminary hours and days.
Check Google's Search Console data-anomalies page.
Confirm the report is still visible and history has not moved to another canonical property.

Google's current anomalies log records several Discover logging errors in 2026, including June 24, May 21, and May 7–8. Those incidents affected reporting rather than actual visibility. Always check the live log instead of relying on a remembered list.

2. Segment the loss by page

Compare a stable period with the decline period and classify:

Pages that disappeared.
Pages that lost impressions but remain present.
Pages with stable impressions and lower CTR.
New pages that partly offset the loss.

Then measure whether one former winner explains most of the site-level decline.

3. Segment by country and appearance

A regional or format-specific loss points to a different hypothesis from a property-wide decline. Check whether Web Stories, video, AMP, or News Showcase changed separately.

4. Audit canonical and indexing evidence

Inspect top lost pages for:

Index status.
Google-selected canonical.
Redirects and migrations.
Large-image crawl access.
max-image-preview changes.
Accidental noindex or resource blocking.

5. Review content and policy changes

Check release history for headline systems, images, templates, structured metadata, page experience, intrusive interfaces, paywalls, or content-policy risks. Review manual actions directly.

6. Add external context

Use Google Trends, news cycles, editorial calendars, and first-party engagement data to distinguish falling interest from a site-specific loss. This is an inference layer; Search Console itself does not provide the causal explanation.

Improve eligibility and presentation responsibly

Google recommends:

Titles and headlines that accurately capture the content.
No clickbait, misleading omissions, or exaggerated previews.
No sensational tactics based on outrage, titillation, morbid curiosity, or shock.
Timely content, strong storytelling, or unique insight.
Relevant, compelling, high-quality large images.
A strong overall page experience.

Large-image guidance

Google's current Discover guidance recommends images that are:

At least 1,200 pixels wide.
More than 300,000 total pixels.
Suitable for a 16:9 presentation and meaningfully cropped.
Enabled for large preview through max-image-preview:large or AMP.

Specify a relevant preferred image with schema.org markup or og:image. Avoid generic logos and text-heavy images.

These practices improve the likelihood and quality of presentation; they do not guarantee inclusion or a specific CTR.

Evaluate content without rewarding clickbait

High CTR is not a sufficient content-quality metric. A misleading preview can raise clicks while damaging trust and violating Discover guidance.

Evaluate a Discover winner across:

Accurate promise-to-page alignment.
Helpful or unique information.
Image relevance and quality.
Engagement and satisfaction after the click.
Returning readership or subscription outcomes.
Policy compliance.
Durability beyond the initial spike.

Use Search Console for acquisition evidence and first-party analytics for post-click behavior. Do not infer engagement quality from CTR alone.

Search Analytics API

Use type: "discover" in a Search Analytics request. The older searchType request field is deprecated in favor of type.

Example request body:

{
  "startDate": "2026-06-01",
  "endDate": "2026-06-30",
  "type": "discover",
  "dimensions": ["date", "page", "country"],
  "rowLimit": 25000,
  "startRow": 0
}

The API can return up to 25,000 rows per request and supports pagination with startRow, but Google states that Search Analytics returns top rows rather than guaranteeing all rows.

Important Discover boundaries:

byProperty aggregation is not supported for type: "discover".
Page grouping uses the canonical URL.
No query analysis is available for Discover.
Search appearance values should be discovered by grouping on searchAppearance before hard-coding filters.
dataState: "all" can include fresh data; final or omission uses finalized data.
dataState: "hourly_all" is required for the hourly breakdown and includes partial data.

Build monitoring from totals and stable segment aggregates before collecting individual low-volume rows.

BigQuery bulk export

Bulk export supports Discover performance with search_type = 'DISCOVER'.

Use the URL-impression table for canonical-page analysis and always aggregate rows because Search Console does not guarantee that rows are pre-consolidated.

SELECT
  data_date,
  url,
  SUM(impressions) AS impressions,
  SUM(clicks) AS clicks,
  SAFE_DIVIDE(SUM(clicks), SUM(impressions)) AS ctr
FROM `project.searchconsole.searchdata_url_impression`
WHERE search_type = 'DISCOVER'
  AND data_date BETWEEN DATE '2026-06-01' AND DATE '2026-06-30'
GROUP BY data_date, url
ORDER BY impressions DESC;

Use is_anonymized_discover in the export schema to identify rows affected by Discover anonymization. Do not expect bulk export to reveal user identities or the missing “query” behind Discover selection.

Bulk export is the best long-term source for high-volume cohort and concentration analysis because it accumulates daily data after setup. It does not backfill dates before the export was configured.

Separate the standard and generative-AI Discover reports

Search Console now documents a separate Generative AI performance report for Discover, currently rolling out to a subset of property owners.

That report:

Measures impressions from supported generative-AI features in Discover.
Appears only for eligible rollout properties with enough impressions.
Groups by page, country, and date.
Does not expose clicks, CTR, or queries in the documented report.
Uses canonical-page attribution and Discover-style impression counting.

Do not infer generative-AI visibility from the standard Discover report alone, and do not add the two reports together without checking whether Google documents overlap for the comparison being attempted.

The dedicated Generative AI visibility reports guide covers that evolving feature in detail.

A practical monitoring framework

Daily integrity checks

Finalized site clicks and impressions.
Preliminary-data boundary.
Missing or zero days.
Search Console anomaly annotations.
Abrupt canonical or indexing changes on top pages.

Weekly performance checks

Top gaining and losing pages.
Page concentration.
Active pages with impressions.
Country mix.
Appearance mix.
New versus evergreen page cohorts.
Median and 75th-percentile page performance.

Monthly strategic checks

Topic and format portfolios.
Discover versus keyword-driven Search contribution.
Large-image and template quality.
Post-click engagement from first-party analytics.
Content-policy and manual-action status.
Dependence on one page, topic, country, or appearance.

Annotate editorial launches, migrations, template changes, image-policy changes, and known Google reporting incidents before interpreting causality.

Common mistakes

Searching for Discover keywords

The report has no query dimension. Page topics and Google Trends can support hypotheses, but they are not a recovered Discover query log.

Treating a viral peak as normal traffic

Compare robust baselines and exclude the dominant page before calling a property-wide trend.

Summing visible rows to rebuild the total

Thresholds and the 1,000-row limit omit rows while their data can remain in totals. The arithmetic will often differ legitimately.

Ignoring canonical attribution

A duplicate page can receive the user visit while the canonical URL and its property receive reporting credit.

Comparing unfinished local-time data with finalized PT data

The 24-hour view is local and preliminary; most other views are PT. Normalize both time zone and finalization state.

Optimizing CTR with sensational previews

Google explicitly discourages clickbait and manipulated appeal. Accurate preview-to-page alignment matters more than a short-lived CTR increase.

Assuming a traffic loss means a technical or policy failure

Interests, content types, system updates, and reporting incidents can change traffic without a site defect. Test the alternatives before making large editorial changes.

Confusing standard Discover with generative-AI Discover

They are separate Search Console reports with different metrics and rollout rules.

What Search Console can and cannot conclude

It can show:

Property totals for Discover impressions, clicks, and CTR.
Canonical pages, countries, dates, and supported appearances above reporting thresholds.
Up to 16 months of available history.
Preliminary hourly data in the 24-hour view.
API and bulk-export evidence for automation.

It cannot show:

A query that triggered an item.
Average ranking position in the feed.
The interests or identity of an individual user.
Every low-volume page row in the interface.
Why Google selected one item for one person.
Whether a traffic change was caused by quality, demand, competition, layout, or an algorithm update without further evidence.
Post-click engagement or conversion quality.
A guarantee of future visibility.

Review checklist

The correct property contains the credited canonical URLs.
Missing-report diagnosis accounts for Google's unpublished threshold.
Impressions, clicks, CTR, and Web Story rules are interpreted correctly.
No keyword or position claims are inferred from Discover.
Site totals are not reconstructed from thresholded table rows.
The 1,000-row interface limit is separated from API and export limits.
PT, local hourly time, and preliminary data are normalized.
Page, country, appearance, and time are segmented before explaining a change.
One-page spikes are separated from portfolio-wide growth.
Data anomalies and manual actions are checked.
Headlines and images follow Google's accuracy and quality guidance.
Large images meet current Discover recommendations.
Search Console acquisition data is paired with first-party engagement data.
Standard Discover is kept separate from generative-AI Discover reporting.