Writing for the email AI summary

By Bill Evans, Deputy Director, MarTech

Illustration of a robot reading an email on a computer screen

Your email subject line used to be the pitch. Now it has competition from something you didn’t write: a one- or two-sentence summary, generated by AI, sitting right next to — or instead of — the preview text you designed.

Gmail, Apple Mail, Outlook, and Yahoo Mail have all shipped some version of this AI summary in the last two years, and open rates are declining, partly because of it. A machine now often answers, in one sentence, the question a subject line used to have to earn an open to answer. 

But treat that decline as a symptom, not the real story. Open rate stopped being a reliable engagement metric long ago. 

And now, between AI compressing the reason to open and regulators narrowing how opens can even be measured, open rate has drifted even further from being a trustworthy signal.

That changes what a marketing email actually has to do. Here’s what the four major mailbox providers are doing, why it matters more than the hero image debate, and how to test your copy against it before you hit send.

What are Gmail, Apple, Outlook, and Yahoo doing?

Gmail. Google’s Gemini-era update introduced AI Overviews that summarize email threads and answer questions about your inbox in natural language, alongside writing tools like Help Me Write and Suggested Replies. Google describes the goal as turning Gmail into a “personal, proactive inbox assistant” rather than a passive list of messages. Some of these summary features are on by default, meaning recipients don’t have to opt in to be affected.

Apple Mail. With Apple Intelligence turned on, a short AI-written summary automatically appears under each unread email in the inbox, replacing the traditional first-lines preview unless the user turns summaries off. Apple Mail also runs a separate feature, Priority Messages, which pulls time-sensitive emails (a same-day invite, a boarding pass) into their own section at the top of the inbox. Summarization and prioritization are two different mechanisms, but together they mean your email is competing for attention before it’s opened, twice.

Outlook. Microsoft allows the reader to request a summary, using Copilot, versus automatically delivering it for every message. A “Summary by Copilot” (or “Summarize”) button appears at the top of an email thread; clicking it generates a summary of the key points, sometimes with numbered citations back to the original messages. Microsoft has also extended this to users without a full Copilot license, provided Copilot chat is enabled, with rollout completing in phases throughout 2026. It’s a meaningfully different interaction model than Gmail’s or Apple’s passive, always-on preview — worth knowing so you’re not writing for a false equivalence across all four clients.

Yahoo Mail. Yahoo’s announcement of its inbox redesign describes a Priority tab that surfaces AI-generated, one-line summaries next to each message automatically. Once a message is opened, the top of the email shows a fuller summary as a bulleted list, along with suggested actions, tasks, or replies. This has since expanded to mobile: Yahoo’s iOS app pulls out specifics like event dates, travel confirmations, billing details, and verification codes into the inbox view before you open anything. Yahoo’s version is the closest to Apple’s in that it’s automatic and on by default rather than something the reader has to request.

The common thread isn’t identical mechanics. It’s that all four companies are building infrastructure specifically to let a reader act on an email without fully reading it.

A note on the wider landscape. Beyond the big four consumer inboxes, a second tier of AI-native email tools — Superhuman, Shortwave, Spark, Notion, and others — layer their own summarization and triage on top of Gmail or Outlook accounts. These matter less for reaching a general consumer list, but if your recipient list skews toward professional, B2B, or executive inboxes, some of your readers may already be seeing a third AI-generated summary of your email before Gmail’s or Outlook’s own summary ever gets a chance to run. Worth knowing, not worth over-optimizing for, since none of these tools publish how their summarization works.

The “live text” problem

Civic Shout has argued that Gmail, Apple, and Outlook’s summary tools scan roughly the first 150-200 characters of live text — the actual readable HTML text in your email, as opposed to text baked into an image or hidden in alt text — to build their summary or preview. That specific character count comes from that publication’s own testing, not from a number that Google, Apple, or Microsoft has published, so treat it as a working estimate rather than a confirmed spec. Yahoo doesn’t appear in that particular testing, but its own documentation describes pulling the same kind of specifics — dates, confirmations, amounts — directly out of message content, which points to the same underlying mechanic: it works off live text, and content locked inside a hero image isn’t content that an AI (or a screen reader, for that matter) can read.

That’s the practical shift. For years, the top of an email was real estate for a beautiful hero image and a punchy headline baked into the design. Now, if your value proposition lives only inside that image, the AI has nothing to summarize but whatever’s underneath or around it — which is often nothing more than a wall of alt text or blank space. The summary a reader sees may say almost nothing about the primary purpose of your message.

Five copywriting principles for the summary layer

  1. Front-load one sentence that can stand completely alone. Assume the summary will strip out your design, your image, and your context. Whatever your first real line of HTML text says needs to work as a complete, standalone pitch — not a fragment that depends on the layout around it to make sense.
  2. Pick one offer per send. A summary compresses. If your email is pushing three things at once, the AI will pick one (or blend them badly), and you lose control of which one. One clear primary objective per send gives the summary something coherent to compress.
  3. Pair every image-based CTA with live-text copy. Hyperlinked text counts as live text that a summarizer can read; a button baked into an image generally doesn’t. If your call to action lives only inside a graphic, add a plain-text link or sentence nearby that carries the same message.
  4. Keep the HTML clean and simple. Short paragraphs, real semantic structure, no deeply nested tables standing in for basic formatting. The easier your markup is to parse, the more accurately an AI model extracts what you actually meant, rather than guessing from fragments.
  5. Write the summary in your head before you write the email. Before you finalize the copy, ask: if an AI reduced this to one sentence, what would it say? If the honest answer doesn’t match your actual message, move your key point earlier and simplify.

Don’t over-correct

Don’t treat 150–200 characters as a hard spec — no platform has published one, and optimizing to an exact number you can’t verify is a good way to write copy that reads as robotic rather than compelling. Don’t abandon design, either; a summary is a preview, not a replacement for the full email experience once someone opens it. And don’t mistake a lower open rate for a failing campaign. If an AI summary satisfies a reader’s question without a click, that’s not automatically a loss — it’s only a problem if your click-through and click-to-open rates are falling, too. Watch those together, not open rate in isolation, to tell the difference between “the summary is doing its job” and “the summary is doing your job for you.”

The real skill shift

Copywriters are now writing two things with every send: the email, and the one-sentence version a machine is going to write about, whether you plan for that or not. The winning move isn’t fighting the summary or trying to trick it. It’s making sure the summary and the full email tell the same true, compelling story — so that whichever one your reader acts on, it’s the story you actually meant to tell.

How to test this before you send

After making your best effort to write copy that will summarize well with AI, the next step is to test it to see if those efforts have paid off. This step will likely be a part of the production process and will require enlisting the help of your friendly neighborhood Martech specialist.

We’ve established that the rules are fuzzy and changing constantly, and there is no official testing methodology, but here are a few things you (or your favorite Martech person) can actually do:

  • Run a live-text audit on your last 5–10 sends. Pull the first 150-200 characters of real HTML text from each (not the hero image, not the subject line) and read it as if it were the whole message. Does it stand alone as a pitch? If you have to squint, so will the AI.
  • Seed-test across real accounts. Send to test addresses on Gmail, Apple Mail, Outlook, and Yahoo Mail, and manually check what preview or summary text actually renders on each. Behavior isn’t identical across clients, and none of the four offers a public preview tool, so this has to be done by hand. Keep in mind that AI summaries aren’t always turned on, or even available by default.
  • A/B the hook placement, not just the wording. Test the same offer two ways — value prop in live text near the top vs. value prop only inside the hero image — and compare click-through rate and click-to-open rate. 
  • Retest on a schedule, not once. Summarization models change without a changelog. What rendered well in Gmail’s summary card last quarter may render differently after the next model update. Build this into a recurring QA pass, not a one-time fix.

Looking ahead

Some tools are already beginning to add features to make testing these AI summaries as easy as we currently test rendering for the various email clients. Inbox Monster already includes this feature in their standard Creative Suite package. While this tool is currently priced for the enterprise, and won’t work with every budget, it’s a sign that this type of testing will (hopefully) become more commonplace and accessible for everyone in the future.

Borrow a page from web dev: check your email’s accessibility tree

Something adjacent to all of this has been playing out on the web development side, and it’s worth stealing the underlying idea even though the tooling doesn’t transfer directly. In May 2026, Google added a new “Agentic Browsing” category to Lighthouse — the audit engine behind Chrome DevTools and PageSpeed Insights — sitting alongside the existing Performance, Accessibility, Best Practices, and SEO scores. Rather than a 0-100 score, it reports a simple pass/fail ratio across checks like accessibility tree quality, layout stability (Cumulative Layout Shift), and the presence of a machine-readable llms.txt file. The point of the category is to answer a new question: Can an AI agent actually read and act on this page, not just a human visitor or a search crawler? Google’s own documentation is explicit that the category is experimental and, at least for now, informational rather than a ranking factor.

No email client publishes an equivalent audit for email HTML, and Lighthouse itself is built to score hosted web pages you can point a URL at — not the HTML rendered inside an inbox — so this isn’t a tool you can point directly at a campaign. But the underlying logic maps cleanly onto the “live text” argument we’re making: whether it’s Lighthouse grading a webpage for agent-readability or Gmail’s AI generating an inbox summary, both are working from the same kind of structure — a DOM-derived accessibility tree, not a rendered image.

A couple of ways to actually use this:

  • Inspect your email’s accessibility tree the way a web developer would inspect a landing page. Chrome DevTools’ Accessibility panel (or Lighthouse’s own accessibility audit) will show whether every interactive element has a programmatic name, whether headings follow a logical order, and whether anything visually important is hidden from the tree. Those are close to the same signals an inbox summarizer is likely relying on.
  • If your ESP gives you a hosted preview URL for a campaign, run it through PageSpeed Insights or a local Lighthouse scan. You’ll get a real accessibility score and, currently, an informational Agentic Browsing readout — with the caveat that this reflects the hosted preview page, not necessarily how a given email client actually renders and parses your message.
  • Treat a clean accessibility tree as work that pays twice. The fixes — semantic HTML, real alt text, a logical heading structure — serve recipients using screen readers and, very plausibly, the AI models generating your inbox summary. It’s one audit doing double duty for two audiences that used to be considered unrelated.

Sources and further reading

Official platform documentation

  • Google, “Gmail is entering the Gemini era” — blog.google
  • Google Mail Help, “Catch up on email threads with AI Overview conversation summaries” — support.google.com
  • Google Mail Help, “Collaborate with Gemini in Gmail” — support.google.com
  • Google Mail Help, “Learn about Gemini features in Gmail” (availability and rollout) — support.google.com
  • Apple Support, “Use Apple Intelligence in Mail on Mac” — support.apple.com
  • Apple Support, “Use Apple Intelligence in Mail on iPhone” — support.apple.com
  • Apple Support, “Use Apple Intelligence in Mail on iPad” — support.apple.com
  • Apple Newsroom, “Introducing Apple Intelligence for iPhone, iPad, and Mac” — apple.com/newsroom
  • Microsoft Support, “Summarize an email thread with Copilot in Outlook” — support.microsoft.com
  • Microsoft Support, “Frequently asked questions about Copilot in Outlook” — support.microsoft.com
  • Yahoo Inc., “Yahoo Mail Redesigns The Inbox For A Cleaner, Smarter And Simpler Experience” — yahooinc.com
  • Yahoo Help, “Turn AI-powered message summaries on or off in New Yahoo Mail” — help.yahoo.com
  • Yahoo Help, “Use AI features in the Yahoo Mail app for iOS” — help.yahoo.com

News and rollout tracking

  • CNBC, “Google is unleashing Gemini AI features on Gmail. Users will have to opt out” (Jan. 8, 2026) — cnbc.com
  • Microsoft 365 Message Center Archive, MC1124564 — Copilot chat summarization rollout timeline — mc.merill.net
  • Litmus, “AI-Generated Email Summaries: What Marketers Need To Know” (context on Gmail, Apple, and Yahoo rollouts from an email-marketing lens) — litmus.com

Adjacent web development context 

Industry commentary

  • Acoustic, “Is email marketing dead, or just evolving beyond batch-and-blast?” — acoustic.com
  • Civic Shout Newsletter, “What Gmail, Apple, and Outlook are doing to your emails” — civicshoutnewsletter.com