Why article translations can fail when images are pasted from Google Docs

If a news article translates only partially, such as translating the title but leaving the body in the original language, a common cause is that one or more images were inserted as Base64-encoded data instead of being uploaded as normal image files.

Symptoms

• The article title is translated, but the article body remains in the original language.
• After updating and republishing the source article, some localized versions do not refresh correctly.
• Localized versions may revert to English in the body while keeping translated titles.
• The translated body may update, but the localized title may stay outdated until the problematic images are removed and translations are refreshed again.
• The article HTML contains image sources that start with data:image/... and include base64.

Cause

Images pasted directly from tools such as Google Docs may be embedded as Base64 data URLs instead of regular hosted image links. This can interfere with the localization workflow for news articles.

This issue can also continue to affect later edits to the same article until all inline Base64 images are replaced with normally uploaded image files.

How to fix it

1. Open the source article.
2. Remove any images that were pasted in from Google Docs or another editor.
3. Re-upload those images as standard .png, .jpg, or similar files using the editor upload flow.
4. Save or republish the article.
5. Regenerate or update the translations if needed.

Best practice

For articles that will be localized, do not rely on pasted inline images from Google Docs, Word, or Pages. Instead, download the images first and upload them directly into the article editor so they are stored as normal hosted assets.

How to verify

After re-uploading the images, check that image URLs are standard hosted links rather than Base64 data URLs, then confirm that the translated versions update correctly.

If localization still does not update after the images are replaced, collect the article URL and the affected locales and escalate for further investigation.