AI Knows If Your Email Will Be Opened

Marketers are in an endless pursuit of the truth. What words should the headline contain? What to include in the CTA? What to exclude from the header? And of course, the burning question remains — will the email campaign perform? Now, imagine knowing the result of your campaign before you hit send. No guesswork, no gut feeling involved.

How to Predict Open Rate Before Sending Your Email

In 2023, a research team released a preprint called NLORP — N-gram LSTM Open-Rate Prediction. They developed a powerful AI model that analyzes email copy, subject line length, word order, and even emotional tone to predict the likelihood of an email being opened.

It might sound like a snippet from a machine learning textbook, but essentially it’s an answer to a marketer’s ultimate question: Can you forecast open rates in advance? The answer is yes.

How the Model Works

The training dataset comprised over 300 diverse subject lines featuring special offers, collected via Google Search. The dataset captured a broad spectrum of promotions: car discounts, exclusive apparel deals, loyalty point accumulation, movie-related offers, email marketing software promos, and more. Subject lines ranged from 15 to 60 characters and were all in English.

Using millions of emails from various brands labeled as “opened” or “not opened,” the system was trained with 5-fold cross-validation. NLORP treats email content as a sequence of language patterns, identifying subtle insights invisible to humans.

For example, the word “reminder” decreases open rate when placed at the beginning but increases it when in the middle of the subject line. Such nuanced correlations form the backbone of the model’s predictions.

Beyond just providing an open rate forecast, the algorithm highlights key three-word phrases (trigrams) along with their individual open rates to explain the overall subject line score. After analysis, it outputs the top five impactful trigrams for every subject, giving marketers actionable insights to optimize their campaigns.

Key Findings

• Best subject length: 7–12 words.

• Action verbs increase opens by 12–15%.

• Negatives (“don’t miss”) reduce performance.

• Emojis lost their edge after 2022 (–9%).

Example
A brand integrated NLORP scoring before each send. Low-score emails were rewritten — more verbs, fewer negations. Within 3 months, open rates rose from 27% to 38%.

Why This Matters
Marketers used to rely on intuition. Now, they rely on the model. NLORP doesn’t replace a sense of style but helps eliminate blind spots. It’s not “the machine writes for us”; it’s a tool that reveals where we make mistakes.

What Marketers Need to Know

• Use past emails as a training dataset.

• Gather your top 50 highest-performing emails by open rate and look for patterns.

• Test action words and avoid negative constructions (e.g., words starting with “un-“, “no-“, “not-“).

• Integrate open rate prediction into your email approval workflow.

This approach empowers marketers to make data-driven decisions, reduce guesswork, and optimize campaigns more effectively.