As at late September 2026. This is a moving picture and I’ll update this page as it moves.
The useful thing about designing email for an AI reader is that it asks for the same email a good CRM team was already trying to write. One clear purpose, stated early, in a structure a human, and now a machine, can follow. None of the technology shifts in the last year has rewarded the six-topic multi-cta newsletter.
So this isn’t a new discipline to learn. It’s a much better reason to do the thing you already knew was right, and a few places where the detail genuinely matters now.
Where AI email summaries actually stand, and it isn’t one picture
Gmail: the direction of travel, nine months on
I wrote about Gmail’s Gemini changes in January, when most of it was still announcement rather than experience. The direction has held. The inbox isn’t a list sorted by arrival any more, it’s a layer that interprets, prioritises and summarises before anyone opens anything.
What’s changed since is that the consequences have arrived in the reporting rather than in the roadmap. – Emails compete with calendar prompts and personal messages, not just with other brands. – Summaries let a subscriber take the value out of a message without opening it. – The Promotions tab orders by relevance rather than time, so a lower-engagement sender sits further down.
Apple: the biggest change for consumer brands, and the most recent
iOS 27 landed on 14 September 2026, and Mail got the most substantial rework in years.
Search now ranks by relevance and intent rather than by keyword and recency. Apple’s own example is that searching a retailer’s name surfaces your recent orders above that retailer’s marketing. Read that twice if you send promotional email to consumers, because it means your campaigns are now competing for retrieval against your own transactional messages, and losing on purpose.
Ask Siri summarises a message on request, finds items in it, tracks packages and pulls out flight numbers. Writing tools and Smart Reply are tuned to the reader’s own style.
And then the part the coverage tends to skip: the AI features need an Apple Intelligence-capable iPhone, and the Siri features aren’t available in the EU or China. Contextual Suggestions launched in English only.
Yahoo: first to do it, least talked about
Yahoo Mail has had one-line AI summaries in the message preview, an inbox that groups by sender and content, and Quick Actions that pull bills, verification codes and package tracking straight out of a message, since November 2024. It launched on iOS in the US first.
One honest caveat: that November 2024 announcement is the most recent dated source I could verify. Yahoo did this before Gmail and Apple made it a conversation, and then the conversation went elsewhere. If a meaningful share of your list sits on Yahoo, it has been reading summaries of your email for nearly two years while everyone wrote about Gemini.
Yahoo will be speaking directly at The London Forum in November, and brands can access my exclusive discounted ticket rate here.Outlook: the one where somebody else decides
Microsoft’s own message centre notice, MC1124564, published in July 2025, took Copilot email summarisation to targeted release that August and to general availability at the end of January 2026. It’s on by default for eligible users, and administrators control it through policy settings and group membership.
That last detail makes Outlook a different shape from the other three. In Gmail, Apple Mail and Yahoo, the reader’s device and their own settings decide what happens to your email. In Outlook, an IT department decides, for a whole organisation at once. If any part of your list is a work address, whether your email gets summarised before a human sees it is now somebody’s procurement setting, and you’ll never see which way it went.
So there isn’t one inbox to design for
Four platforms, four sets of rules, and they don’t overlap neatly. Apple’s need a capable iPhone and exclude the EU and China. Yahoo’s started on iOS in one market. Outlook’s are an administrator’s switch. Gmail’s have moved fastest and furthest.
Depending on the device someone holds, the country they’re in and who runs their email, the same campaign lands into a different set of rules. If you sell across Europe, or to a mix of consumer and work addresses, a meaningful share of your list sees none of this while another share sees all of it, and no report you currently run splits them.
What almost everyone writing about this is getting wrong
Look at who holds this territory. Search for AI and the inbox and you’ll get deliverability vendors, almost without exception, framing it as a spam-filter and sender-reputation story. Fix your authentication, warm your IPs, watch your complaint rate.
That advice isn’t wrong. It’s just answering a different question from the one a CRM lead is actually being asked.
The strategic consequence is this: if something reads and summarises your message before the human does, you have two readers, and only one of them is a customer. The first one decides what the second one sees. That changes what “capture attention” means, it changes what a subject line is for, and it changes which part of your email is load-bearing. None of that is a deliverability problem, and no amount of SPF records addresses it.
What it’s done to your measurement, which is the part with teeth
Open rate has been unreliable for years, and it’s now unreliable in a new way as well.
Validity’s own engagement data shows Gmail image loading activity, tracking pixels included, dropped by roughly a third in late November 2025. Their published analysis attributes it to reduced image prefetching, and they’re straight about that being informed speculation rather than a measured cause. They don’t state a sample size. Treat the figure as directional and the explanation as a hypothesis, which is how they present it.
Either way, the practical effect is that a third of a signal disappeared without anybody’s programme getting worse. If your 2026 reporting showed a drop around then and somebody went looking for a campaign to blame, that’s likely what you were looking at.
Clicks are heading the same way, for a different reason. Security systems follow the links in an email to check them before a person sees the message, so a share of what your platform counts as clicks was never a human. The tell I look for is a click landing within a second or two of delivery, or one contact clicking several unrelated links at once.
The honest position is that the two numbers most CRM reports lead on are both measuring less than they used to, and neither decline says anything about whether your programme is working. That’s an argument for measuring progression instead: first to second purchase, repeat rate, revenue per customer over a window. Those don’t care what the inbox does.
Four things worth doing
1. Read your own emails the way a machine does. Paste one into two different AI assistants and ask each for a five-bullet summary. Where they agree, your message is clear. Where they wander off in different directions, your email is carrying too many jobs. It takes five minutes, it costs nothing, and it’s the single most useful exercise I’ve given a team this year.
2. Put the point in the first two lines. Not the offer’s history, not the seasonal framing. If your value proposition sits halfway down, assume it may never be seen, because a summary pulls from what it judges meaningful and it judges early.
3. Treat structure as content. Clear hierarchy, short separated sections, real headings, white space, and clean code underneath so that hierarchy survives. A summary built from a well-structured email resembles the email. One built from a wall of copy is a guess.
4. Read your small print. A summary doesn’t know your terms were written for the lawyers. There are already cases of a summary lifting a rate or an exclusion out of the footer and presenting it to the reader as part of the message. Read the bottom of your next promotion and ask whether you’d want any of it quoted back to a customer.
Where this leaves you
The brands that handle this well won’t be the ones that bought a tool. They’ll be the ones whose emails already had one clear purpose, because that’s the email that survives being compressed by something that doesn’t care about your campaign calendar.
Which is the same thing good CRM has always rewarded. The machine has just made it harder to get away with the alternative.
If you want to know how your programme scores on the capabilities this touches, the CRM Priority Check takes about five minutes and tells you which capability to fix first.