The short answer
Links in a first cold email hurt, but almost only with one kind of recipient: companies whose email runs on Microsoft 365. In the same campaign, in the same week, a first email with a link got about half the replies from Microsoft-hosted prospects that a first email without one did. With Google-hosted prospects, and with prospects behind security gateways like Proofpoint or Mimecast, the difference was small enough to be noise.
That split is what makes this a deliverability effect rather than a copywriting one. If the link were simply a weaker call to action, it would cost replies everywhere. It doesn't. It costs replies where Microsoft's filters decide whether the email reaches the inbox.
The data
We looked at every first-touch cold email sent through RocketSDR between May and October 2026: 532,575 emails from 1,434 campaigns run by 61 B2B companies, sent from 1,161 mailboxes on 324 sending domains. Follow-ups were left out, because a follow-up's chances depend on what happened to the first email.
Most campaigns either always include a link or never do, so comparing them would mostly compare different audiences and offers. But in 124 campaigns the AI-written first email included a link for some prospects and not for others, with at least 40 emails on each side. Those campaigns are the study: same sender, same list, same offer, same week — the link is the main thing that varies.
| All first-touch emails | Controlled comparison | |
|---|---|---|
| First-touch cold emails | 532,575 | 138,530 |
| Campaigns | 1,434 | 124 |
| Companies sending | 61 | 24 |
| Sending domains | 324 | 223 |
| Delivered (bounces removed) | — | 127,757 |
| Period | May–Oct 2026 | May–Oct 2026 |
| Recipient's mail provider | Delivered emails | Share |
|---|---|---|
| Google Workspace / Gmail | 48,876 | 38% |
| Microsoft 365 / Outlook | 39,121 | 31% |
| Security gateway or other host | 39,760 | 31% |
How we measured it
Cold email has no reliable way to see which folder an email landed in, so we used two signals that only exist if the email reached a person's mailbox:
- Human replies — someone read the email and wrote back. This depends on deliverability and on the copy.
- Automatic replies — out-of-office notices, "I've left the company" messages, auto-acknowledgements. These fire from the recipient's mailbox and don't care what the email says, so they track whether the email arrived, not whether it persuaded.
For each recipient type we compared link and no-link emails only within the same campaign and the same week, then combined those comparisons into one weighted ratio (a Mantel-Haenszel rate ratio). A ratio of 1.0 means the link made no difference; 0.5 means half as many replies. The 95% ranges come from resampling whole campaigns 400 times, so one unusual campaign can't drive the result.
Result: the penalty is Microsoft-specific
Microsoft-hosted prospects replied about half as often to first emails with a link, and sent about a third as many automatic replies. For Google-hosted prospects and security gateways, every range includes or sits close to 1.0.
| Recipient provider / reply type | With link | No link | Ratio | 95% range |
|---|---|---|---|---|
| Microsoft — any reply | 0.58% | 1.30% | 0.48 | 0.32–0.69 |
| Microsoft — human reply | 0.39% | 0.84% | 0.56 | 0.39–0.82 |
| Microsoft — automatic reply | 0.19% | 0.46% | 0.32 | 0.16–0.53 |
| Google — any reply | 3.05% | 2.34% | 0.92 | 0.77–1.09 |
| Google — human reply | 2.16% | 1.72% | 0.87 | 0.68–1.03 |
| Google — automatic reply | 0.89% | 0.62% | 1.07 | 0.76–1.63 |
| Security gateway / other — any reply | 1.88% | 1.54% | 0.88 | 0.64–1.20 |
| Gateway / other — automatic reply | 1.13% | 0.69% | 1.10 | 0.73–1.75 |
Checking the result
A single comparison can mislead, so we cut the Microsoft result several other ways.
Email length. Emails with a link were longer on average (908 characters against 721). Matching on length as well as campaign and week left the result unchanged: 0.47× for any reply, 0.24× for automatic replies.
Sending provider. The penalty showed up whether the email was sent from Google Workspace or Microsoft 365 mailboxes. It was steeper from Microsoft 365 senders, though both ranges overlap.
| Sent from | With link | No link | Ratio (any reply) | 95% range | Ratio (automatic) |
|---|---|---|---|---|---|
| Google Workspace mailboxes | 0.68% | 1.31% | 0.60 | 0.44–0.88 | 0.38 |
| Microsoft 365 mailboxes | 0.48% | 1.38% | 0.24 | 0.08–0.55 | 0.28 |
| Cut | With link | No link | Ratio |
|---|---|---|---|
| Main comparison (campaign + week) | 0.58% | 1.30% | 0.48 |
| Also matched on email length | 0.60% | 1.31% | 0.47 |
| Single-link emails only | 0.51% | 1.30% | 0.45 |
| Links to a website only | 0.55% | 1.29% | 0.45 |
| Month | With link | No link | Ratio |
|---|---|---|---|
| June 2026 | 0.76% | 3.87% | 0.32 |
| July 2026 | 1.07% | 1.62% | 0.61 |
| August 2026 | 0.70% | 1.18% | 0.56 |
| September 2026 | 0.51% | 0.91% | 1.03 |
| October 2026 | 0.39% | 1.33% | 0.22 |
Campaign by campaign
Looking at campaigns one at a time is noisier, because each has only a handful of replies, but it shows the effect isn't coming from one or two big senders. Of the 27 campaigns with at least 60 emails to Microsoft-hosted prospects on each side, 16 got fewer replies with the link, 6 got more, and 5 tied.
The same count for Google-hosted prospects also leans towards fewer replies with a link (26 campaigns lower, 10 higher). So links may cost a little everywhere — a weaker call to action, perhaps. The difference is size: within noise at Google, about half at Microsoft.
| Recipients | Campaigns compared | Fewer replies with link | More with link | Tied |
|---|---|---|---|---|
| Microsoft-hosted | 27 | 16 | 6 | 5 |
| Google-hosted | 36 | 26 | 10 | 0 |
Which links?
Almost every link in the sample pointed to a website — usually the sender's company site — sent from a separate cold-email domain. 16,252 of the 16,594 website-link emails to Microsoft-hosted prospects linked to a domain other than the one they were sent from, and those carried the full penalty (0.44×).
That leaves questions we can't answer yet:
- Calendar links. Only 978 comparable emails carried a booking link alone — too few to say anything.
- Linking to your own sending domain. Only 342 comparable emails did; the difference was small, but so was the sample.
- Why Microsoft reacts this way. A link to a different domain from the sender's is a classic phishing pattern, and Microsoft's filters are built to catch phishing. That fits the data, but it's our reading, not something this study proves.
What this study can't tell you
- It measures replies, not folders. We infer inbox placement from human and automatic replies. A direct seed-box test would measure placement itself; we're running one next.
- The AI decided when to add a link. It may have added links more often for some kinds of prospects. Matching on campaign, week and length removes most of that, and a prospect-type effect would show up at Google too — it mostly doesn't — but it can't be ruled out completely.
- One platform, B2B outbound, May–October 2026. Filters change. We'll re-run this every quarter.
- Mostly website links. See above.
What to do with this
- Send the first touch without links. Ask a question instead of pointing to a page. If the prospect is interested, they'll ask for the link — and a link in a reply thread is a different situation from a link in a cold email.
- Move links to later touches or replies. Case studies, decks and booking links can wait for step two or the reply.
- Know how much of your list is on Microsoft. About 31% of the delivered emails in this study went to Microsoft-hosted companies. Look up your prospects' MX records; if most of your market runs on Microsoft, this matters more for you.
- Report results by recipient provider. Compare reply rates from Microsoft-hosted and Google-hosted prospects. A big gap at similar audiences usually means Microsoft is filtering you.
Where RocketSDR fits
RocketSDR looks up every prospect's mail provider before sending and routes each email to the mailbox most likely to land with that provider. This study is part of how we decide what those rules should be. If you want to see how your own outbound splits between Microsoft and Google prospects, book a call and we'll run the numbers on your campaigns.
Frequently asked questions
Do links in cold emails hurt deliverability?
In our data, yes — for recipients on Microsoft 365. Within the same campaign and week, first emails with a link got 0.48× the replies from Microsoft-hosted prospects. For Google-hosted prospects the effect was within noise (0.92×).
Is it the link or the call to action?
Mostly the link. Automatic replies (out-of-office and similar), which don't depend on the copy, fell to 0.32× at Microsoft-hosted prospects but not at Google-hosted ones (1.07×). A weaker call to action would cost replies everywhere.
Are calendar links safe in cold emails?
We can't say yet: only 978 comparable emails carried a booking link alone. Nearly all links in our sample went to company websites.
Does it matter whether I send from Google Workspace or Microsoft 365?
The penalty appeared from both. It was steeper from Microsoft 365 mailboxes (0.24×) than from Google Workspace mailboxes (0.60×), though the ranges overlap.
How do I know if a prospect uses Microsoft 365?
Look up the MX record of their email domain. Microsoft 365 domains route mail through hosts ending in mail.protection.outlook.com.