Google brings multimodal BigQuery analytics into GA

Google has made ObjectRef and multimodal data analytics in BigQuery generally available, extending BigQuery's handling of unstructured data stored in Cloud Storage.

The update lets users represent files such as images, video, audio and documents as native ObjectRef columns within BigQuery tables alongside structured data. Teams can therefore query structured and unstructured data together, rather than moving outputs between separate systems for analysis.

ObjectRef is a built-in BigQuery STRUCT data type that points to objects outside BigQuery. It allows organisations to keep large files in Cloud Storage while using BigQuery to analyse them alongside metadata, timestamps and other relational fields.

The move addresses a long-running challenge for data science, data engineering and machine learning teams. Many have relied on separate pipelines for structured and unstructured data, then had to feed inference outputs back into data warehouses for reporting and further analysis.

Unified workflow

With the general release, users can pass multimodal inputs directly to Gemini models in a single SQL query or through conversational analytics. Extracted features and annotations can also be stored in the same row as the ObjectRef, keeping the original file linked to the structured output.

This is designed to reduce the loss of context that can occur when audio, images or documents are converted into purely structured formats. Customers are already using the system for multimodal vector search, large-scale classification and feature extraction from document and media archives through SQL.

The release also includes changes aimed at widening the range of use cases, including support for End User Credentials for ad hoc analysis, broader support for Cloud Storage metadata sources, and native array support for ObjectRefs so multiple files can be analysed in a single call.

Users can create ObjectRefs from BigQuery object tables, Cloud Storage insights datasets or raw string URIs in standard BigQuery tables. Production pipelines can still use explicit BigQuery connections, while personal identity-based access is now available for faster exploration.

SQL focus

A central part of the product's pitch is that full ETL pipelines for unstructured data can be built with SQL. Teams are using BigQuery to prepare data, run AI-driven classification and annotation, and keep original files attached to the resulting records for later review or deeper analysis.

Examples include call centre analysis, retail compliance checks and phishing detection. In one case, audio files from customer service calls can be referenced directly in BigQuery, classified into issue categories and then used as a source for conversational analytics.

Another example involves filtering product uploads by querying whether an image contains a restricted item or policy violation. A third uses arrays of ObjectRefs to compare a trusted login page screenshot with a suspicious page to identify possible phishing attempts.

Customer use

During the preview period, the technology moved into production use across moderation, security, retail and customer service workloads. These use cases combine multimedia assets with other operational data in a single pipeline.

One customer cited by Google was Snap, which uses the technology in content review workflows.

"At Snap, the safety of our community is a top priority, and we aim to limit users' exposure to content that violates our Community Guidelines. We regularly review samples of public posts, and by leveraging ObjectRef with BigQuery AI, we transitioned from simple random sampling to post-stratification sampling. This transition improved the efficiency of our human review processes, enabling us to reinvest reclaimed capacity into new and expanded workflows that further enhance our ability to detect and address potentially harmful content," said Carlos Ortiz, Data Scientist, Snap.

ObjectRef and multimodal data analytics are now available across all BigQuery editions and on-demand pricing tiers.