There is no universal safe ratio
An image-heavy message can be difficult to understand when images are blocked and may leave filters little readable context. That does not create a universal text-to-image threshold. Providers combine the message with sender, reputation, audience, and recipient history.
Make the email useful without images, compress assets, provide meaningful alt text, and avoid building the whole offer as one graphic. Judge the rendered production message, not only the designer canvas.
Links carry their own evidence
Tracking domains, destination domains, redirects, and public shorteners can all affect trust and make diagnosis harder. Open the final post-tracking URLs and check every redirect. Remove unused links and replace public shorteners with direct or branded paths.
A branded tracking domain cannot protect a compromised destination. Monitor landing pages and domain ownership as part of the sending system.
Clarity beats spam-word superstition
A word can be risky in a deceptive or exaggerated message, but filters do not operate from one shared forbidden dictionary. Fake reply prefixes, hidden identity, misleading urgency, and an unclear unsubscribe path are whole-message trust problems.
Write so a skeptical recipient can identify the sender, purpose, offer, and next step. That is better advice than swapping every commercial word for a euphemism.
Test the final template
Send through the same ESP or CRM that applies the tracking and HTML transformations. Compare provider placement, then simplify one meaningful element if the report and content evidence support it.
If placement is broad and poor, investigate sender and audience evidence before rebuilding the design. Content is one branch of the system, not the automatic explanation.