Yes. AI email summaries can change what a prospect sees of your cold email before they open it. They do not decide whether it gets delivered.
Apple Mail can show a short AI summary under an unread email in place of its first lines. Gmail sorts mail into tabs, puts summary cards at the top of the inbox for some kinds of email, and can place an AI Overview above a thread.
All of that happens after delivery. Your domains, authentication and list still decide whether the email arrives. What inbox AI changes is the first impression, and a summary of a pitch is still a pitch.
The goal has not changed: replies that turn into calls. In one campaign my team ran for a UK firm that builds software and sells IT consulting, the cold email channel brought 41 positive replies between 17 April and 17 May 2023, and those became 13 booked calls and 2 sales.
What changed is the step between the moment your email lands and the moment someone reads it. Here is what Apple and Google now do there.
Three common assumptions, checked against the help pages:
- "Summaries push cold email toward spam." No. The Apple and Google help pages cited below describe display and sorting. They do not link summaries to spam filtering.
- "Landing in Promotions means the email failed." No. A category is a list inside the inbox.
- "You still write the preview line under your subject." Not always. For Apple Mail users with summaries on, the mail app writes it instead. Litmus puts Apple Mail and Gmail together at nearly 90% of email client market share (measured from tracked opens), so these two apps are where your prospects will meet these features first.
What is the difference between inbox AI and writing AI?
Two kinds of AI get mixed up here. Sender-side AI helps you write the email. Inbox-side AI belongs to the prospect's mail app: it sorts, summarizes and highlights what has already arrived. This article is mostly about the second kind.
Sender-side AI has published reply data. Hunter's State of Cold Email data, covering 31 million emails its users sent in 2025 (a tool vendor's own numbers), found that manually edited emails got a 5.2% reply rate, against 4.4% for fully automated ones.
Inbox-side AI has no reply data in the official documentation, only a description of behavior. Here is the map, taken from Apple's and Google's help pages:
| Mail app | Feature | What the prospect sees | What it means for a cold email |
|---|---|---|---|
| Apple Mail | Categories | Mail sorted into Primary, Transactions, Updates and Promotions | Decides which list your email sits in |
| Apple Mail | Apple Intelligence summaries | A short AI summary under each unread email, replacing the first lines | Your opening gets condensed before anyone reads it |
| Gmail | Inbox tabs | Primary, Social, Promotions, Updates and Forums in the Default inbox | Decides which tab your email sits in |
| Gmail | Summary cards | Cards at the top of the inbox for order tracking, events, travel, bills and promotions | Only relevant if your email reads like a deal |
| Gmail | Gemini AI Overview | A synthesis of key points and replies above a thread | Your whole sequence can be read as one paragraph |
Sources: Apple's guide to Mail categories, Apple Intelligence in Mail, Gmail categories, Gmail summary cards, Gemini summaries in Gmail and Google's Gemini in Gmail page.
Every summary in that table works from the text you sent. I keep every first email short and human, the way one person types to another. Inbox AI adds one more reason. A short, plain email leaves a summarizer very little room to pick the wrong point.
What should the first lines of a cold email say now?
The opening used to have a fallback: your first line sat under the subject as a preview, even if nobody opened the email. For readers with summaries on, an AI summary now sits in that spot, and the fallback is gone. The summary condenses whatever your email leads with, so the reason to reply has to sit in the first two sentences, stated plainly.
That is why the first message sells a meeting and nothing bigger. I use the first email to sell a short call, never the service itself. A stranger cannot judge scope, budget and risk from an inbox. They can judge 15 minutes.
A pitch in message one is the quickest way to get ignored. An AI summary only speeds that up, because the prospect gets the verdict "someone is selling something" without reading a word you wrote.
The summary also works only from real text. When Oracle's Digital Experience Agency tested AI email summaries in 2024, Apple Intelligence and Android AI both ignored any text inside images. Apple Intelligence sometimes did little more than a basic rewrite of the preheader, and other times ignored it, with no clear pattern. In a plain-text cold email, your words are all the summary has to work with.
What goes first:
- Why this email reached this person: their role, their market or a real fact about their company
- The problem, in their words
- The ask: a short call, and what they get from it
What stays out of the opening lines: small talk, "I hope this finds you well", a compliment nobody believes, the year you were founded, your headcount and a list of services.
When I go through a founder's draft, the company history paragraph is one of the first things I delete. If a summarizer meets it first, the most it can report is that a company introduced itself.
The rest of the email should read the same way, and here is how to write cold email that does not sound like AI.
What does Apple Mail do to a cold email?
Apple's Mail guide says that with Apple Intelligence on, a short summary shows up beneath every unread email in the inbox. A reader can also click Summarize at the top of an open email to get a summary above it. For a threaded email, Apple condenses the whole conversation.
Users who dislike this can switch summary previews off. Then Mail goes back to showing the opening of the latest message.
For a cold email, that means:
- For anyone with summaries on, the preview you wrote is replaced by Apple's version of it
- When a reader summarizes a follow-up, they get your whole thread condensed, so a sequence that repeats itself reads as one long pitch
- Some prospects turn summaries off and still see your real first lines, so those lines have to work both as text and as summary input
The same guide covers Priority Messages, a view Apple Intelligence adds to the inbox: time-sensitive emails move to the top, and Apple's examples are a same-day dinner invitation and a boarding pass.
Do not try to write your way into it with fake urgency. A stranger's email that pretends to be urgent reads as pressure, and it gives the reader no reason to trust you.
Does cold email land in Promotions on iPhone?
Apple does not say that sales outreach goes to Promotions. Its iCloud Mail guide says Mail can sort messages into Primary, Transactions, Updates and Promotions. Transactions holds confirmations, receipts and shipping notices. Promotions holds marketing, coupons and sale messages.
A one-to-one note asking an operations director for a call is none of those. An email built like a newsletter, with banners, prices and a "limited offer" line, looks a lot like one.
Two details on that page matter for outbound:
- A message in Transactions, Updates or Promotions that carries time-sensitive information also appears in the Primary list
- When a reader moves a sender to another category, the choice covers that sender's current and future messages, so one move by the prospect decides where the rest of your sequence lands
Readers can also turn categories off, and then new mail is not sorted at all.
Being in Promotions is not the same as being junked. Apple presents categories as a way to find and manage mail faster, and nothing on the page links a category to spam filtering. The cost is attention. Your email sits in a list built for offers and coupons, which the prospect probably scans with that in mind.
What do Gmail tabs, summary cards and Gemini overviews change?
Gmail tabs, summary cards and Gemini overviews get lumped together, but they do different jobs.
Tabs
With the Default inbox type, Gmail sorts mail into Primary, Social, Promotions, Updates and Forums to cut clutter. Promotions is for deals, offers and other promotional email.
When a user moves an email to another tab, Gmail learns from it over time. Users can also set filters that send one sender's mail to a chosen category. As with Apple, a tab decides where a delivered email sits.
Summary cards
Summary cards show at the top of the inbox on the web and in the Gmail apps for Android, iPhone and iPad. Google lists five kinds: order tracking, events, travel, bills and promotions.
The cards are generated from the information in the email, and a "Based on [x] email" link takes the reader back to the source. A note asking one person for a call fits none of those five kinds.
Gemini AI Overviews
Gemini can put an AI Overview above a Gmail thread, pulling together its main points and the replies in it. Google says the feature is available to Gmail users worldwide with a work, school or personal account, supports only some languages, and needs smart features turned on. Google's Gemini page also describes summaries of long emails and contextual smart replies.
For outbound, the overview is the feature to write against. Once you send a follow-up, there is a thread, and a thread is what the overview condenses.
If every step repeats the first email, the overview will say something like "the sender keeps offering the same service." If each step adds a new angle, a proof point or a short question, it shows a sender with something new each time.
In our campaigns, the third follow-up, with real time between the steps, is usually the one that gets the reply, so that is the step that most needs a new point.
Smart replies point to one more habit: make the ask answerable in one line, such as a choice between two days. A short "Tuesday works" is easy to send whether the prospect types it or taps it.
Which cold email openers survive a summary?
A summary-proof opener passes this checklist:
- One audience: the email is clearly written to one role in one kind of company
- One problem, named in the reader's words
- One reason to talk now
- One proof cue: a result with a number, or the type of client you did it for
- One small ask, with what the reader gets from the call
- Plain text in short paragraphs, with no banners or price blocks
Here is an invented example for a firm selling scheduling software to civil engineering contractors.
Weak opener:
Hi Jana, I hope your week is going well. BuildTrack was founded in 2012 and is a leading provider of digital solutions for construction, from scheduling and reporting to document control and mobile apps.
Stronger opener:
Hi Jana, contractors your size usually rebuild site schedules by hand every time a delivery slips. We built a way to cut that to minutes. Worth 15 minutes next Tuesday to see if it fits your sites?
A summary of the first can only say a company introduced its products. The second keeps the problem, the promise and the ask intact, because those are all it contains. Add one real result from your own clients and it carries the proof cue too.
One quick test before you send: delete your first sentence. If the email loses nothing, that sentence was too vague to deserve the top spot.
Will the summary help, hurt or do nothing?
| Your email | What the summary has to work with | Likely effect |
|---|---|---|
| Reason and ask in the first two sentences, kept short overall | Your own point | Neutral to helpful: the summary repeats it |
| Opens with a greeting, a compliment or company history | An introduction | Hurts: the prospect learns only that someone introduced themselves |
| Lists several services or several asks | Many competing points | Likely hurts: the summary has to pick one point and may pick the wrong one |
| Follow-up that repeats step one | The same pitch twice | Hurts in a thread overview: the sequence reads as one pitch |
| Follow-up with a new angle, proof or question | A sender adding something each time | Helps in a thread overview |
| Built like a newsletter, with offers, prices or banners | A promotion | Hurts: it matches how both providers describe promotional mail, so it is likely to be sorted as one |
Do summaries change reply rates or only the preview?
Neither Apple nor Google publishes anything about reply rates for summarized email. Their pages describe how mail is displayed. A summary can shift a first impression, but a prospect still has to need what you sell.
So keep expectations tied to real ranges. In our campaigns, reply rates usually sit around 1%. Once volume runs into thousands of emails a day, it is closer to 0.5%. Usually 10% to 50% of those replies become booked meetings, and those averages fall as volume rises.
Replies and booked calls are the layer you can actually measure. For one software development client, 392 replies turned into 29 booked calls in 30 days and 5 contracts worth EUR 246,000. A rewritten first line will not rescue a weak list or a weak offer, which is why the cold email work we run starts with the list and the offer.
The summary is also not where anyone decides to buy. They buy after they look you up, and that happens away from the inbox.
Measure the same way you always should have: positive replies and booked calls. Opens were already a shaky number, and summaries make them shakier, because a prospect can read Apple's summary of your email without ever opening it. That is why open rate is the wrong number to judge a campaign.
Before your next send, read your last first email as if all you could see was a one-line summary of it. If that line would not make you reply, rewrite the first two sentences before you touch the subject line.
If your cold email results shifted once inbox summaries arrived and you want another opinion on why, book a call.
Questions founders ask about inbox AI and outbound
What about Outlook?
I found no official Outlook documentation on this to cite here, so check it yourself. Mail apps change their preview and assistant features often, so write as if any client may condense your message.
Send a test from your outreach domain to an Outlook mailbox you control, then look at it on desktop and mobile. If the text beside the subject drops your reason for writing, rewrite the opening. One sharp opening works across every client, so you do not need a separate fix for each one.
Does Yahoo Mail summarize emails too?
Probably. I found no Yahoo help page to cite, so this rests on one secondary source. Salesforce notes that Apple, Gmail and Yahoo pull snippets and summaries of messages into inbox previews. If your prospect list includes Yahoo addresses, the same advice holds for them. Add a Yahoo account to your test inboxes so you can see how its preview treats your opening before the campaign goes live.
Can an AI summary get my cold email wrong?
Yes, and you cannot correct it after sending. Litmus warns that AI-generated summaries may not always be accurate, so the only thing you control is the input. Avoid sarcasm, irony and negative framing such as "not another agency", which a model can misread. Keep each figure in the same sentence as the result it measures, so a summarizer cannot attach the number to a different claim.
Are click rates still a reliable signal for outbound?
Only with care. Salesforce points out that summaries let someone read an email without technically opening it, so opens and clicks may no longer mean what they used to. For a first touch, the simplest fix is to leave links out completely. A plain note with no tracked link also looks less like marketing. Treat a click as a hint about interest. The real signal is the reply.
How can I see what the summary will say before a prospect does?
Set up a few test mailboxes you own. Use one Apple device with Apple Intelligence switched on and one Gmail account with smart features enabled. Send the whole sequence to them, every step, and reply once so a thread forms.
Then read every generated line and check whether it states who you help, the problem and the ask. Wherever it misses one of those, edit the source email and test again.
Is it wrong to use AI to write cold emails?
No, as long as a person stands behind every word that goes out. The real risks are invented facts about the prospect, fake personal details and claims your company cannot back up. A workable rule: AI produces a draft, a human checks each fact, the proof and the ask, and the email goes out under the name of someone who will actually answer the reply.