You rebuilt the template. Your spam rate did not move. That is the usual end of a content-first deliverability project, and the studies already on this site are why.
This page does not add a dataset. It points you to InboxEagle’s Q1 2026 content signals study (774,828 emails) and the link density study (192,651 emails), and to the posts that hold their numbers. None of those findings is a Gmail, Yahoo, or Microsoft requirement.
In all three studies, the content most people are told to fear (trigger words, heavy images, many links) did not raise the spam rate. That is a correlation in one ecommerce corpus, not permission to ignore content: the authors tie it to healthier audiences sending that content, not to the content itself.
What the content signals studies found
| Study | Sample and period | Observation | Limit |
|---|---|---|---|
| Spam trigger words | Content signals study, 774,828 emails, Q1 2026 | Discount language did not land in spam more often than mail without it | One ecommerce corpus. Small gaps are not a wording ban. |
| Image-to-text ratio | Content signals study, 774,828 emails, Q1 2026 | Image-heavy mail had a lower spam rate than balanced mail | The image-heavy group was the small slice. The post says so. |
| Link density | Link density study, 192,651 emails, Q1 2026 | Higher link counts lined up with lower spam rates | Very large HTML did not keep improving. The post separates link count from file size. |
Go to each post for its numbers. The pattern the authors draw is the useful part for you. The “risky” content showed up more often on mail that already had a healthier audience. The filter outcome tracked the audience and the sender, not the myth. Subject line personalization and sale subject lines belong in the same family. Read them for their own numbers. Do not collapse them into “content does not matter” as a slogan.
Where content still matters
Corporate gateways and old scoring tools still score keywords. The spam word checker tells you what those engines flag. It does not tell you where Gmail will file your Klaviyo campaign. The trigger-word study says that directly.
HTML that is broken, a tracking link that points at a bad domain, or a public URL shortener are not “image ratio.” They are infrastructure in the body of the message. Tracking domains and shorteners is that problem. Order confirmations should not be rewritten into newsletters to chase the image study.
Authentication still comes first. A DMARC failure is a different class of event from a word in the subject. The DMARC failure study is the one to read before you A/B test adjectives.
Reputation over content: what to change when placement drops
Google’s guideline is to keep your user-reported spam rate below 0.10% and never reach 0.30%. Above 0.30%, you are ineligible for mitigation support until the rate stays under 0.30% for 7 consecutive days. Those reports happen after the message is delivered, so they measure the audience, not the template. Yahoo’s bulk complaint rate is under 0.30% on inbox-delivered mail. Neither document lists a maximum number of images or links.
If your mail is opted in and still complained about, the lever is who is on the send and how often. Spam complaint rate, frequency, and sunset are the operational pages. A new template on the same cold segment reproduces the complaint rate in a different color.
HTML that is so broken a client cannot read it is a rendering problem. Gmail’s formatting section says to follow HTML standards and not to hide content with HTML or CSS. Hidden content is called out as something that might cause spam classification. That is a guideline against concealment, not a score for how polished the file is. Attachments are a different risk: you rarely need them on marketing mail, and a file the recipient did not ask for is a reason to complain. Neither one replaces the complaint rate. Do not add an attachment to “look more legitimate.”
Methodology
No new collection was run for this article. Findings are summarized from the three posts in the table, each of which has its own methodology: Q1 2026, InboxEagle placement data, the 774,828-email content signals study and the 192,651-email link density study. They are not a census of all ecommerce email. Image-heavy volume is a minority of the image study. Link-count results are correlations with sender type. Do not quote them as proof that adding links or images will move your spam rate.
InboxEagle
Test your email for the content issues that do matter
The free spam tester scores a real send on authentication, SpamAssassin rules, blocklists, and content. Use it to rule content out quickly, then look at your complaint rate and list.
Sources
- InboxEagle Q1 2026 studies linked above. Observations, with the sample sizes in the table.
- Email sender guidelines, Google. The published bulk-sender signals are authentication and spam rate, not a template ratio.
- Sender best practices, Yahoo. Complaint rate and authentication for bulk senders.