Google Cloud AutoML

Google Cloud AutoML is Google’s family of tools for training a custom machine learning model on your own labelled data without designing the model yourself. You upload examples, AutoML handles model selection and training, and you get an endpoint that makes predictions. In MLforSEO’s courses it is the route for teaching Google’s language models your own categories instead of the fixed ones the Natural Language API provides.

How it works

The workflow is supervised learning with the hard parts automated:

• Prepare a dataset of examples with the labels you want (for example, page types or intent classes).
• AutoML splits it into training, validation and test sets, tries model configurations and tunes them.
• You review evaluation metrics (precision, recall, a confusion matrix) and deploy the model for predictions.

AutoML launched as separate products (such as AutoML Natural Language) and was folded into Vertex AI, Google Cloud’s machine learning platform, in 2021.

Current status for text. Since 15 September 2024 you can no longer train or update Vertex AI AutoML models for text classification, entity extraction or sentiment analysis. Existing AutoML text models could be used until 15 June 2025. Google now directs these tasks to Gemini, using prompts for quick results or supervised fine-tuning for deeper customisation. AutoML remains available in Vertex AI for other data types, such as tabular data.

How it’s used in SEO and marketing

The use cases MLforSEO teaches still apply; the tool you build them with has changed:

• Custom content classification for specialist niches (medical, legal, B2B software) where generic categories are too broad.
• Page type identification (blog, product, category, support) as a multi-class classification task.
• Custom search intent labels that match your funnel.
• Detecting elements on a page, such as Hamlet Batista’s AutoML approach to finding structured data opportunities.

Example

A legal publisher wants to tag 20,000 articles with its own 12 practice areas. Previously that meant labelling a few hundred articles and training an AutoML text model. Today the equivalent is a labelled sample, a Gemini prompt with clear category definitions, an accuracy check against the sample, and fine-tuning only if prompting falls short.

Limitations and things to watch

• Older tutorials, including some course material, reference AutoML text features that no longer exist.
• Custom models are only as good as your labels; inconsistent labelling produces inconsistent predictions.
• Training and hosting endpoints cost money even when idle; check pricing before deploying.

Learn it on MLforSEO

• What Machine Learning Classification Is, and Where It Fits in SEO
• How to do Text Classification with Google’s Natural Language API in Google Sheets
• Structured Data Automation with Python
• Introduction to Machine Learning for SEO course