Prompt Examples For Cold Outreach Emails That Did Not Get Flagged As Spam

If you’ve ever set up automated cold outreach emails inside something like Instantly or Mailshake or even just with Google Apps Script + Sheets, you probably know the pain of waking up to a weirdly silent inbox. No opens, no clicks, no bounces. Nothing. Just vibes

The last time it happened to me, it turned out my ESP (ConvertKit at the time — weird choice for cold, I know) silently moved my DNS settings into a new “WarmUp Only” domain. Didn’t even give me a specific warning. So yeah, sending prompts is hard enough. Making them feel human *and* deliver reliably is an actual dark art.

Below are the actual prompt formats that either:

– Got real responses
– Didn’t get flagged by spam filters
– Let me slide back into threads even after being ghosted once or twice 🙂

Everything here was tested via Woodpecker and Lemlist, and some also got ported into Zapier almost as a dare.

1. Cold prompts that feel like social messages with a purpose

This format almost always works right after a light LinkedIn engagement — like I’ve commented on a post, and then 2-3 days later sent them a message using something like this:

“Saw your post about scheduling chaos this week — totally relate. I work with small teams testing out async calendar flows; your setup might be missing a trick. Worth a quick nerdy chat?”

The key here is not pretending this is some generic offer. The outreach prompt *supports* something they’ve already said. Yes, you’ll have to do 3 seconds of actual scrolling. But the response rate when you anchor to *their timeline language* is way higher.

Things that broke:
– Lemlist sometimes pulled the wrong LinkedIn preview (blamed some phantom caching logic)
– If I inserted an emoji like 🧠 into the input, Woodpecker rendered it as “¿” on the output
– Gmail marked automated follow-ups as “Promotions” unless the subject line included a 1-syllable verb

Weirdly, the shorter the message, the more likely I was to get a response. One campaign that consisted of fewer than 300 characters consistently outperformed messages that had detailed offers.

2. Follow-up prompts sent after no initial response

People give up too fast. The first time my Zapier follow-up sequence fired on a five-day delay, over 40% of the replies came during follow-up 2 or 3. The magic trick? Making each prompt feel like a new angle, not a nag.

Here’s what didn’t work:
> “Just checking in on my last message…”

(Instant death)

Versus what actually got replies:
> “Quick one — not sure if this even fits your current goals. Totally fine if not. Just noticed how close your last launch dates were — how are you managing team handoffs when that happens?”

A human being reads that and thinks: okay, fair question.

You need to sound like someone dropping a new idea, not demanding time on their calendar. Also — ask *easy* questions. Save open-ended strategizing for warm leads.

One gotcha:
– Instantly had a bug last April where scheduled follow-up #3 would sometimes fire *after* follow-up #1 due to a misordered queue. Make sure your cold email tool lets you manually reorder the sequence. I now add a custom field `$seq_id` to track what’s actually firing.

3. Prompts that survive forwarding by the recipient

People *will* forward cold emails. If you’re lucky. But that’s where overly personal hooks or dynamic first lines can backfire. I had a sequence with `{Icebreaker}` that pulled from a Notes column like:

“Loved what you said on the SaaStr stream re: backend ops”

Which looked awesome — unless someone forwarded it internally. Then it was like:

> “Hey — why did this random person say something about a livestream from last year?”

So I started using prompts that don’t collapse under forwards:

> “Looping in here since this might touch both marketing and tech — we help small GTM teams kill off duplicate handoffs from lead to post-sale. Happy to ballpark how close your setup already is.”

It’s vague, yes. But in the way that feels safe to CC your ops lead into. And then I follow-up based on opens — if two different people open the same thread, I pivot to a more direct prompt:

> Just saw [Name] click in too — flagging this in case this landed on the wrong desk 🙂 Let me know who’d own it if so.

That last one works better than I’m willing to admit.

4. Prompts that play nice with GPT-powered spam filters

Here’s one of those painful things nobody talks about: some spam filters now use LLMs to classify “deceptive” prompts even if nothing is technically wrong. They look for *intent*. So subject lines like:

> “Quick question about your team”

…get flagged as phishy, even if your domain and DKIM are squeaky clean.

What actually makes it through consistently is specificity without presumption. Not:

> “Can I help grow your pipeline?”

But instead:

> “Saw your partner program pitch deck — curious if the attribution tools you’re using allow real-time error tracking?”

The LLMs see that and think: okay, this person seems like a real user with context.

Also discovered:
– Zoho’s mail filter engine flagged anything with more than three invitation-like verbs (book, schedule, reserve) even if they weren’t in a calendar context
– GMASS toned-down subject lines performed better than the same subject tested in Apollo.io — possibly due to subtle differences in how it injects the email headers
– Plaintext emails with wide margins and extra line breaks weirdly had *lower* click/open rates — suspected invisible styling being parsed as invisible text

So now I keep prompts tight, literal, and minus marketing lingo. Honestly hurts my copywriting ego a little, but hey — whatever survives the Great Filter ¯\_(ツ)_/¯

5. Cold prompts that convert better when used in comment sections

Some of the best cold prompts weren’t emails at all. They were tiny, comment-sized requests in community threads (Slack groups, Facebook posts, the occasional Reddit thread). The trick is to write *like a respectful parasite*.

Here’s one that blew up in RevOps Co-op:

> “We just ran into this *again* — post-sale conversion scheduled in HubSpot but no active service record inside Jira. Anyone found a way to kill off that ghost-sync loop?”

That thread generated 12 comments. I DM’d 4 people. 3 ended up on a call. One converted 🙂

Nobody likes a hard pitch in a help forum. But if your prompt reads like shared pain, other people lean in.

One technical thing I messed up early: I used tools like PostMagic to schedule posts with prompts like this, thinking I’d up my surface area. But Slack’s threading behavior killed that strategy — half the time my posts went into archived channels, despite PostMagic saying it had “posted successfully.”

So back to manual commenting it is. But if you’re gonna go down that route, just commit. Think like a helpful broken robot with very niche domain pain — you’ll blend right in.