Currently Available: Need a skilled Software Developer for your next project?
Categories
SEO, GEO and AEO

Reporting Image and Video Discovery Cases in BigQuery

When a Google Search Console BigQuery export shows traffic from image or video searches, decide what you want the report to show. Search Console’s search_type field identifies a search surface, such as the Google Images tab or video search. A landing-page report asks which URLs received the traffic; a query-group report asks what people searched for. Those questions need different groupings, and some visual or video appearances require more than a search-type filter.

Build each report from the table and fields that answer its question. Before relying on the totals, check which values and columns the export actually contains. That matters especially for image clicks, which Search Console credits to the page hosting the image, and for newer multimodal reporting, whose BigQuery value is not fully specified in the published table reference.

Choose the search surface first

Search Console’s search type describes where a result appeared. image refers to results in Google Images, video to video-search results, and web to the combined All tab. The web category can include results featuring images or videos, so filtering for image or video alone answers a tab-specific question, not every question about visual content in search. Google documents these categories in its BigQuery table reference.

Google announced web multimodal reporting for image-based searches such as Lens and Circle to Search, then announced that this data would be available in BigQuery for properties with an active bulk export. The Search Analytics API reference uses the value webMultimodal, but the published BigQuery table reference does not list a corresponding export value. Inspect the search_type values in your own table rather than assuming the API value also appears in BigQuery. Google’s multimodal reporting announcement and API and BigQuery update describe the rollout.

A simple check reveals which values are present:

SELECT
  search_type,
  COUNT(*) AS row_count
FROM <code>project_id.searchconsole.searchdata_url_impression
GROUP BY search_type
ORDER BY row_count DESC;

Use the results to define the report’s scope. If a question concerns video-rich appearances across search surfaces, a VIDEO filter alone does not establish which results qualify. The URL-level export has true-or-false fields that mark search appearances, but the table reference does not list every current field. Check the live table schema for the appearance field you need; if it is absent, the export does not support that particular split.

Group landing pages by their role

Use searchdata_url_impression when the report needs landing URLs. This table aggregates performance by URL, while searchdata_site_impression aggregates at the property level. The URL table can show which pages or page families received image- or video-search activity; the property table suits property-wide query and search-type trends.

Search Console’s image metrics identify the page that hosts an image, not the individual image file. When several images appear on one page, the export does not separate their performance. Treat URL-level image data as a landing-page measure, not an asset-level measure, as Google explains in its image metric definitions.

The following hypothetical query groups URLs into page families based on path patterns. Replace the project, dataset, date range, and path rules with those used by your site.

SELECT
  data_date,
  UPPER(search_type) AS search_surface,
  CASE
    WHEN REGEXP_CONTAINS(url, r'/products/') THEN 'product pages'
    WHEN REGEXP_CONTAINS(url, r'/videos/') THEN 'video pages'
    WHEN REGEXP_CONTAINS(url, r'/guides/') THEN 'guide pages'
    ELSE 'other pages'
  END AS landing_page_group,
  SUM(clicks) AS clicks,
  SUM(impressions) AS impressions,
  SAFE_DIVIDE(SUM(clicks), SUM(impressions)) AS ctr
FROM <code>project_id.searchconsole.searchdata_url_impression
WHERE data_date BETWEEN DATE '2026-10-01' AND DATE '2026-10-31'
  AND UPPER(search_type) IN ('IMAGE', 'VIDEO')
GROUP BY 1, 2, 3
ORDER BY clicks DESC;

These page-family labels come from your URL rules, not from Search Console. Check that the rules classify the intended pages before using the totals to compare sections. For more on choosing and querying the exported tables, see querying image and video data in BigQuery.

Group queries without hiding missing text

Create query groups in SQL when individual query strings are too detailed for a report. For example, regular expressions can assign visible queries to categories such as how-to searches or product searches. These categories are analyst-defined; Search Console does not provide a standard intent-group field.

Some rows contain performance metrics without query text. Google excludes anonymized queries from the query text, and the export represents that missing text as an empty string. Keep those rows in a separate group if the report needs to account for the full performance totals. Removing them gives a list of visible queries, but that list does not represent all query-related activity. Google describes this behavior in its bulk export documentation.

This hypothetical example keeps rows with unavailable text visible:

SELECT
  UPPER(search_type) AS search_surface,
  CASE
    WHEN query IS NULL OR TRIM(query) = ''
      THEN 'query text unavailable'
    WHEN REGEXP_CONTAINS(LOWER(query), r'\b(how to|tutorial|guide)\b')
      THEN 'how-to or guide'
    WHEN REGEXP_CONTAINS(LOWER(query), r'\b(price|buy|review)\b')
      THEN 'commercial or review'
    ELSE 'other visible queries'
  END AS query_group,
  SUM(clicks) AS clicks,
  SUM(impressions) AS impressions
FROM <code>project_id.searchconsole.searchdata_url_impression
WHERE data_date BETWEEN DATE '2026-10-01' AND DATE '2026-10-31'
GROUP BY 1, 2
ORDER BY clicks DESC;

The example rules are deliberately simple and will misclassify some searches. Review the queries assigned to each group, and adjust the rules to suit the analysis. Keep the raw query field available for diagnosis rather than treating custom groups as a replacement for it.

Keep each report at one data grain

A report’s grain is what each row represents, such as one row per date and search type or one row per URL and query group. Use the URL-level table for landing-page questions and the property-level table for property totals. Do not join both tables and add their metrics as though they represented separate activity: they summarize Search Console data at different levels. The table guidance describes those distinctions.

The export can contain multiple rows with the same values for the fields you group by, so use SUM to add their metrics for each group in the report. Calculate click-through rate from the summed clicks and impressions, as in the examples, rather than averaging row-level rates. If you need country or device breakdowns, add those dimensions only when they answer the report question, and include them in the grouping.

Check the export before interpreting totals

Before sharing a report, check that its date range and property match the question. Confirm that the search surface and table fit it, and that the grouping answers it too. Then inspect the actual search_type values and appearance fields in the export. This catches capitalization differences and prevents a report from relying on a multimodal or video-appearance value that the table does not contain.

Also confirm that repeated rows have been summed and that query rows with blank text have not disappeared unnoticed. Search Console bulk exports run daily, but separate tables can arrive at different times. The ExportLog table records successful exports and revisions to previously exported dates; failed exports do not appear there. BigQuery groups the exported data into date partitions, and those dates use Pacific Time. Check the export log and date boundaries when a recent period looks incomplete. Google’s query guidance covers aggregation behavior, while its Search results report documentation explains why interface totals can vary with filters and aggregation level.

When comparing BigQuery with Search Console, match the property and dates in both sources. Then use the same search type and filters, and compare results at the same aggregation grain. A page-level result and a property-level total answer different questions, so they should not be expected to match row for row.

What I'm building

Delegate tasks. Get software.

Give Vroni a GitHub issue, bug report, spec, or rough idea. It reads the repo, plans the change, writes code, runs checks, and works toward a review-ready pull request.

Take a look at vroni.com

Email updates

Usually a new article and a few links I found interesting.

No spam. Unsubscribe with one click.

Leave a Reply

Your email address will not be published. Required fields are marked *