Mitigating the Risks of Inbox AI Summaries

Whether provided by inbox providers or via agents, AI summaries of promotional emails are inherently reductive, introduce errors, overwrite brand voice, and deprive marketers of a significant amount of their “first impression” real estate. That has created a host of new risks for marketers, including lower campaign performance and sender reputation damage.

To help you mitigate those risks, Zeta’s Creative and Strategy Teams have developed three potential postures you can take in response to inbox AI summaries:

• Lean Forward - a posture that seeks to actively optimize for inbox AI summaries
• Neutral Stance - a posture that seeks to reduce the likelihood of AI summaries being wrong
• Lean Away - a posture that seeks to train subscribers to ignore AI summaries

For a full discussion of each of those approaches, watch the on-demand webinar below. But for a quick rundown of each approach, read on.

Lean Forward

This posture seeks to actively optimize for inbox AI summaries. Brands can accomplish this through message simplification and additional coding.

• Focus your messages on a single CTA that AI summaries can understand. Most AI summaries reduce promotional emails down to a single CTA, so lean into that by only having one CTA. Moreover, this CTA should be limited to offers that AI can easily understand, such as a straightforward percent-off deal.
• Code Deal Annotations and schema. While not always honored, Gmail gives marketers the ability to code Deal Annotations into their campaigns to better control the pre-open AI summary experience. Similarly, Yahoo Mail enables coding schema, which is similar but not identical to Annotation coding.
• Code Deal Cards. To control the post-open AI summary experience, Gmail gives brands the ability to code Deal Cards. Similar to Deal Annotations, Gmail sometimes ignores this coding and will even sometimes overwrite it with their own Deal Card code through their process of Automatic Extraction.

Alarmed by AI summary inaccuracies, some of our clients have begun coding Annotations, schema, and Deal Cards. While the impact is difficult to measure, both in terms of reach and impact, we believe it improves AI summary accuracy.

Neutral Stance

This posture seeks to reduce the likelihood of AI summaries being wrong. Brands can accomplish this through a variety of copywriting, design, and coding tactics.

• Write clear and descriptive subject lines that make sense without preview text. Just like SEO killed off creative news headlines, marketers need to write straightforward subject lines. With preview text now often replaced by AI summaries, brands shouldn’t rely on that to give their subject lines meaning.
• Use extra From field characters. With inbox-provided AI summaries routinely replacing preview text and sometimes even subject lines, brands should use any extra characters they have remaining for their sender name to help clarify the topic of their message. Examples include “YourBrand Holiday,” “YourBrand Newsletter,” and “YourBrand Cart.”
• Use plenty of HTML text. If you want the key messages in your email to be accurately summarized more often, using HTML text is essential. Inbox AI doesn’t read graphical text and rarely reads alt text.
• Have a clear hierarchy. HTML text that’s earlier in an email is given priority over text that’s positioned closer to the bottom.

The good news is that many of these recommendations align with accessibility best practices we’ve been recommending for many years. It turns out that accessibility is great for people and machines.

Lean Away

This posture seeks to train subscribers to ignore AI summaries by routinely demonstrating that summaries can’t remotely capture what your campaigns entail. Brands can accomplish this by fully utilizing email’s capacity to create rich subscriber experiences.

• Use multiple content blocks and calls-to-action. Don’t shy away from including many content blocks and CTAs, just like most B2C brands have done more than two decades.
• Have a unique brand voice. If you have a unique brand voice, don’t snuff it out because of AI. Also don’t avoid highlighting the unique voices of your customers, employees, and others connected to your brand.
• Regularly include animation and video content. This is powerful content that AI can’t even begin to summarize.
• Regularly include personalized content. Customers expect this, so deliver.
• Regularly include interactive content. CSS-based interactivity offers lots of capabilities that are supported for most subscribers across most B2C brands.
• Regularly include real-time content. Countdown timers, real-time inventory, live loyalty point balances, and weather forecasts for travel destinations are just a few of the possibilities.
• Regularly include polls, quizzes, and surveys. Not only can these be fun and community-building for subscribers, they can collect valuable zero-party data brands can then use to follow up with segmented and personalized campaigns.

Just like brands create rich and interactive experiences on the web and in their apps, they do the same in email. So, it’s not so much about leaning away from AI summaries as leaning into a strong brand experience in the inbox.

Which Approach Should You Use?

Given the content of your emails and the goals of your email marketing program, one of these three postures probably makes more sense for your brand than the others. However, don’t get locked into one approach.

First, you may send occasional campaigns that might benefit significantly by adopting a different posture. So, stay flexible.

And second, the state of inbox AI summaries is still evolving. Recent years have seen big ups and downs with AI summaries, but in general they’ve expanded. And as consumers become more sophisticated users of LLM plug-ins and agents, we expect more consumer-initiated summaries of inboxes. So, expect more changes.