AI-Personalised Cold Outreach: Where It Gets a Reply and Where It Gets You Reported as Spam
AI has made cold outreach faster to write and easier to spot as fake, often in the same email. The tools got good enough to generate a personalised-sounding opener for every name on a list in seconds — which is exactly why inboxes are now full of messages that reference a "recent post" that doesn't exist, or a "shared interest" pulled from a five-year-old bio line. Used well, the same technology can genuinely save hours of real research. Used lazily, it's the fastest way to get marked as spam a business has ever had. 💖
What most people get wrong about AI-personalised outreach
- Treating "personalised" as the same thing as "true" — a sentence can be perfectly grammatical, specific-sounding and still be wrong, because the AI guessed or hallucinated the detail rather than verifying it.
- Using AI to fake volume-scale sincerity — "I really admired your recent article on X" sent identically to 300 people isn't personalisation, it's a template wearing a costume.
- Skipping the human read-through — the fastest way to send an email referencing the wrong company, the wrong job title, or a "post" that was actually written by someone else at the business.
- Confusing "AI-assisted" with "AI-written end to end" — the outreach that gets replies almost always has a human doing final judgement on tone and accuracy, with AI doing the time-consuming research legwork.
- Ignoring how easy verification has become — recipients now routinely paste a suspicious opening line into their own AI tool to check if it's templated, so a fake-personalised email is more likely to be caught than ever.
The pre-send verification checklist
Run every AI-assisted cold email through this before it goes out — it takes under two minutes per email once you're used to it:
2. Did I find this detail myself, or did the AI infer/guess it? If you can't point to where it came from, cut it.
3. Would this sentence be believable if I told the recipient an AI helped write it? If it would feel like a "gotcha" reveal, rewrite it.
4. Is the connection between their detail and my offer actually logical, or just adjacent-sounding? ("You raised a Series A" ≠ "so you need my bookkeeping services.")
5. Could I say this same line out loud to their face without cringing? If not, it's too try-hard for an inbox either.
6. Have I checked this isn't going to the same person twice from two different lists or campaigns?
Two scenarios, two very different outcomes
Why this line is getting harder to fudge
Two years ago, a vaguely-personalised cold email could plausibly be read as "they did their homework." Now that everyone knows AI can generate that same sentence in bulk, the bar for what reads as genuine has moved — recipients are more suspicious by default, and more likely to fact-check a suspicious claim in seconds using the same tools that may have generated it. That's actually good news for anyone doing real research: genuine personalisation stands out more sharply against a noisier backdrop of obviously fake versions, because it's specific in ways a template can't fake (referencing something that requires actually reading the source, not just scraping a field).
Mistakes to avoid when using AI for outreach
- Sending at a volume you can't personally verify — if you can't fact-check every "personalised" detail before it goes out, the volume is too high for the method.
- Letting the AI choose the CTA too — a generic "worth a 15-minute chat?" tacked onto a specific, well-researched opener undercuts the effort that came before it.
- Reusing the same "personalised" template structure for every industry — recipients compare notes more than businesses expect, especially within tight-knit industries.
- Not tracking reply and spam-report rates by campaign — without this, you won't notice a batch is backfiring until deliverability has already taken a hit.
Frequently asked questions
Is it always obvious when an email was written by AI?
Not always — well-edited, fact-checked AI-assisted copy can be indistinguishable from something written from scratch. The giveaway usually isn't the writing quality, it's the accuracy: recipients notice when a "personal" detail is wrong, outdated, or generic-but-dressed-up, far more than they notice smooth sentence structure.
Does using AI for outreach count as spam under Australian law?
It depends on the details, not the fact that AI was involved — unsolicited commercial email to individuals is generally covered by the Spam Act regardless of how it was written, while outreach to a business email address about a business matter has more latitude. The tool used to write it doesn't change the underlying rules around consent and identification.
How much research time should I actually spend per email?
Enough to verify every specific claim you're making, even if that's only two or three minutes — the goal isn't exhaustive research, it's ensuring nothing you send could be disproved by the recipient in ten seconds.
What's the single biggest tell that outreach was faked at scale?
A "personal" detail that's technically true but generic enough to apply to hundreds of other people — "I noticed your company is growing" or "saw you're focused on customer experience" reads as templated because it could describe almost anyone, which is exactly the kind of line AI tools default to when a real detail wasn't actually checked. 🌴
Keep reading 🤍
I help Gold Coast and Brisbane businesses grow with branding, websites and marketing that actually works.
Work with me ✦