Contextual Image Analysis

Summary

Contextual Image Analysis is an AI-powered feature that evaluates whether alt-text accurately describes an image by comparing the text to the visual content and classifying the result as descriptive or not descriptive.

Overview

Contextual Image Analysis is an AI-powered feature that helps you assess how accurately your images are described. It analyzes existing image–alt-text pairs to determine whether the text meaningfully represents the visual content.

This enables you to validate the quality of your image descriptions, ensuring they contribute to a more accessible and inclusive digital experience.

How it works

When you trigger an Image Analysis, Siteimprove collects the images and their associated alt-text from the selected occurrences you choose to review. 

Each image–alt-text pair is then sent to an AI model trained to evaluate how closely the description matches the image content.

After processing, each image is categorized as one of the following:

• Descriptive – The alt-text accurately describes the image content.
• Not descriptive – The alt-text does not appear to accurately describe the image.

The results are displayed directly within the Accessibility module, under AI-powered reviews, allowing you to review the findings and take appropriate action.

When to use it

Use Contextual Image Analysis when you want to:

• Verify the accuracy of image descriptions provided by content authors.
• Identify potential mismatches between visuals and alt-text that may affect accessibility.
• Gain AI-driven validation of existing accessibility efforts without manually reviewing each image.

Frequently Asked Questions

Do I need to enable or configure this feature?

No setup is required. You’ll see the option to trigger Image Analysis once it’s available in your account.

Is the analysis automatic?

No. Image Analysis is performed on demand — you decide when and where to run it.

Which images can I analyze?

Any image with an existing alt-text that appears in the Images or Pages with images views.

How are results categorized?

Each analyzed image is marked as Descriptive or Not descriptive, depending on how accurately the alt-text represents the image.