Can I use AI to write my emails?

Everyone's Using AI for Email — But Almost No One's Doing It Well

Almost every team writing emails with AI today is doing it the same way — and almost none of them are doing it well. That's not a criticism of the tools; it's a gap between adoption and mastery that the data makes impossible to ignore.

According to industry statistics, 87% of businesses using AI apply it to email workflows. Yet only 6% of those organizations qualify as AI high performers. As one analysis bluntly put it: "AI works. Organizations can't absorb it. The technology arrived faster than the workflows, skills, and organizational structures needed to use it."

Part of the explanation is simply inexperience. A survey of email marketers found that 44% have used AI for just six to twelve months — most teams are still in their first year of figuring this out. And the results show it: 88% still require moderate to substantial human editing of AI output before anything goes out the door.

The numbers tell a consistent story of near-universal adoption meeting rare mastery:

• 87% of AI-using businesses apply AI to email workflows, but only 6% are high performers
• 44% of email marketers have used AI for under a year
• 88% of teams still need moderate to substantial human editing of AI drafts
• 42% of C-suite executives say AI adoption is "tearing their company apart"

That last statistic is striking. The problem isn't the technology — it's that teams have bolted AI onto old workflows instead of designing workflows around what AI does well. Most people paste a prompt, accept the first draft, and wonder why the output sounds generic. The teams getting real results treat AI as a collaborator with a defined role, not a vending machine for finished text.

Here's the reframe that matters: your question isn't whether to use AI for email — you almost certainly already do, or soon will. The question is where AI sits in your process. Should it brainstorm angles before you write? Draft the first version for you to sharpen? Catch errors and tighten tone at the end? Each placement produces different results, and the data shows the highest performers are deliberate about this choice rather than defaulting to whatever their tool suggests.

If you're curious how far this thinking extends, AI agents with the right instruction packages can handle far more than drafting — they can personalize at scale, route drafts for review, and automate the follow-up steps. SkillzBench is building a free, independent directory of these agent skills, with security scanning and benchmarking so you can see which ones actually help, plus a free MCP connector to test them with your own setup.

For now, the gap between the 87% and the 6% is where most readers live. Closing it starts with a simple question: not "Can I use AI to write my emails?" but "Which part of my email should AI touch first?"

What AI Email Writing Actually Delivers: The Evidence

The numbers on AI-assisted writing are more solid than most AI hype — and they come with honest caveats worth knowing before you hand over your inbox.

The strongest evidence comes from a peer-reviewed study by researchers at NBER, Stanford, and MIT, which found that access to an AI assistant increased worker productivity by 14% on average — and by 34% for novice and low-skilled workers, while barely moving the needle for experienced pros. As the study authors note, the AI essentially "disseminates the best practices of more able workers." Agents with two months of AI experience performed as well as colleagues with six or more months without it, per coverage of the research.

On the email marketing side, the performance gains are measurable too. According to aggregated industry statistics:

• AI-generated subject lines lift open rates by up to 22%, with 5–10% typical
• AI-driven personalization delivers a 41% revenue increase and 13% higher click-through rates
• AI-written emails achieved a 9.44% CTR versus 8.46% for human-written ones — an 11% improvement
• Average AI ROI runs $3.70 back for every $1 invested

Now the caveats — and they matter. A survey of B2B marketers found that while 87% report better productivity from AI content tools, only 58% report better content quality — and 12% say quality actually got worse. Speed gains are outrunning quality gains, which is why 88% of teams still require moderate to substantial human editing of AI output.

The practical takeaway: AI reliably makes you faster, makes emails perform better when applied to subject lines and personalization, and helps weaker writers most. What it doesn't reliably do is replace your judgment. If you want to push beyond drafting into full workflow automation — routing drafts, personalizing at scale — agent skills can handle the rest of the pipeline, and SkillzBench is building a free, independent directory with security scanning and benchmarking so you can see which skills actually help before you install them.

Where AI Works Best in Your Email Workflow

The smartest email users aren't asking AI to write for them — they're asking it to think and check alongside them. The data shows exactly where the work has moved: recent survey data shows ideation now leads AI writing use at 74%, while editing has doubled from 19% to 38% and drafting has actually fallen from 57% to 44%.

That shift makes sense. A blank page is where most of us lose time, and a second pair of eyes is where most errors hide. AI is strongest at both ends of the process, with humans owning the middle.

The highest-value tasks cluster around measurable wins. Research on AI email performance shows AI-generated subject lines can lift open rates by up to 22%, and AI-driven personalization delivers a 41% revenue increase. First drafts benefit too — a peer-reviewed NBER study found AI assistance boosts productivity by 14% on average, and 34% for novices.

Where AI earns its keep in email:

• Subject lines — the highest measured lift, up to 22% on open rates
• Personalization — 41% revenue increase when AI tailors content
• First drafts — biggest gains for less experienced writers
• Proofreading — catching errors traditional tools miss

That last point is worth a closer look. In hands-on testing by the Jisc National Centre for AI, ChatGPT caught 5 out of 5 deliberate writing errors, while Microsoft Word caught only 2. But the same test surfaced a real quirk: ChatGPT tends to rewrite your text even when explicitly told not to. The fix is prompt discipline — instruct it to suggest changes only, and give it context about audience and purpose.

One caution before you worry about your emails "sounding like AI": detectors won't save you. Testing shows humans correctly identified only 57% of AI-written texts, so AI detectors are unreliable judges of quality. Judge each email on its own merits with human review — which is what 88% of teams already require.

The pattern is collaboration, not delegation. If you want to push further, agent skills can automate the surrounding workflow — routing drafts, personalizing at scale, and prepping review queues. SkillzBench is building a free directory of security-scanned, benchmarked agent skills plus a free MCP connector, so you can test which ones actually help before trusting them.

The Risks Nobody's Talking About: Shadow AI and Sensitive Data

That polished AI-drafted email might be the least of your company's worries — what you pasted into the prompt to get it could be the real problem. While most conversations about AI email focus on tone and productivity, the governance picture is quietly alarming.

According to Zylo's workplace AI research, 97% of organizations reported an AI-related security incident in the past year. This isn't a fringe risk reserved for careless companies; it's the baseline experience of organizations adopting AI at any scale. Email sits at the center of that exposure because it's where sensitive information naturally lives.

The core driver is what practitioners call "shadow AI" — unvetted tools employees bring in themselves. The same research found that 78% of AI-using knowledge workers bring their own AI tools to work, and 60% of IT leaders admit they can't see every generative AI tool in use across their organization. Your security team can't protect what it can't see, and a quick paste into a personal chatbot account is nearly invisible.

The concern is well-founded. 43% of IT leaders name sensitive data exposure as their top AI concern, and email drafting is a prime vector. When someone asks a public model to "make this client email sound better," they may be sharing contract terms, customer names, pricing, or internal strategy — content that leaves the company's controlled environment entirely.

So what should never go into a public AI model? A short list covers most of the real-world incidents:

• Customer or client names combined with account, pricing, or dispute details
• Contract terms, negotiation positions, or unreleased financial figures
• Employee personal data — anything HR-related, performance reviews, or disciplinary matters
• Credentials, internal system names, or anything covered by NDA or regulation

The fix isn't banning AI — the productivity evidence is too strong, and peer-reviewed research shows real gains from AI assistance. The fix is sanctioned, vetted tooling: enterprise versions with data protections, a clear policy on what may be shared, and tools your IT team actually knows about. As Zylo's guidance puts it, treat AI output as a first draft — and treat AI inputs as a governance decision, not just a writing task.

The same vetting logic applies to the AI agents and skills increasingly automating email workflows. SkillzBench is building a free, independent directory of agent skills with security scanning, benchmarking, and a free MCP connector — so the tools touching your communications are ones someone has actually checked.

Your AI Email Playbook: From Prompts to Automated Workflows

Knowing AI can write emails is one thing; knowing how to use it well is where most teams stumble. The gap is real: 87% of AI-using businesses apply AI to email workflows, yet only 6% qualify as high performers, according to industry data. A simple playbook closes much of that gap.

Start by treating every AI draft as exactly that — a draft. 88% of teams still require moderate to substantial human editing of AI output, and governance guidance from Zylo recommends treating AI output as "a first draft rather than a finished decision." Your edit pass is where tone, judgment, and accountability live.

Prompting matters more than most people expect. Hands-on testing by the Jisc National Centre for AI found that ChatGPT rewrote text even when asked not to — so be explicit about what you want:

• Say "suggest changes only, do not rewrite" when you want edits, not a rewrite
• Give audience context — who reads it, why it matters, what the desired action is
• Specify the language variant (British vs. American English) and register (formal, warm, concise)

Then assign a human owner after each AI step. As one analysis of AI writing adoption puts it, check that someone with subject knowledge owns the stage after every AI step — a rule that keeps quality from quietly eroding as speed increases. The data shows why: 87% of B2B marketers report better productivity from AI tools, but only 58% report better quality, and 12% say quality got worse.

Once your drafting process works, agent skills take it further. AI agents with the right skills — instruction packages for tools like Claude Code, Cursor, and Codex — can automate the rest of the workflow: personalized drafts at scale, subject-line variants (which lift open rates up to 22% per campaign research), and routing drafts for human review. The catch is trust: published research has found roughly 26% of skills carry at least one vulnerability, and with 97% of organizations reporting an AI-related security incident in the past year, unvetted tooling is a real risk.

That's where SkillzBench comes in. It's building a free, independent directory of agent skills with security scanning, version-pinned validation, and sandboxed benchmarking that measures whether a skill actually improves output — plus a free MCP connector so your agent only sees the skills you choose. The pre-launch waitlist is live now, so you can be first to browse, test, and equip your email workflow with skills you can actually trust.