Templates vs. AI: which one should answer that email?

Picture a contractor who sets up his own AI email assistant on a Sunday. He's proud of it, and he should be: he figured out the whole thing himself. The first message it handles Monday morning is somebody asking whether he services Middleburg.

He does. He always has. That answer has never once been different.

So a question with one permanent answer gets routed through a language model that costs money, takes a few seconds, and can occasionally decide to get creative. A saved reply would send the identical sentence instantly, forever, for free.

Meanwhile the emails that eat his evenings (a customer describing a problem across four rambling paragraphs, where the job is to figure out what they're really asking) he's still writing by hand at nine at night.

Backwards. And it's the most natural mistake to make.

The whole decision comes down to one question

Before you automate any email, ask this: does the answer change?

Not "is this email common." Not "is this email annoying." Does what I say back depend on what this specific person said?

If no — the same answer every time, no matter who's asking — that's a template. AI adds nothing to it but latency and risk.

If yes — I have to read this thing, weigh it, and decide something — that's where AI is worth having. That's the only place it's worth having.

That's the line. The rest is detail.

Templates are better than people give them credit for

A saved reply gets treated like the boring option, the thing you settle for. It has one property AI can't match: it is identical every single time.

Things that should almost always be a template:

• Hours, service area, "do you do X" — the fixed facts of your business
• "Got your message, I'll have numbers to you this week"
• Booking confirmations and what to expect on the day
• The nudge you send when a quote's been sitting for a week
• Deposit and payment instructions
• The polite no for work you don't take

A saved reply in Gmail takes about ninety seconds to set up and then never drifts. It won't invent a price. It won't get three percent friendlier at eleven at night. It won't tell somebody you're open Sundays because that seemed like a reasonable thing for a business to be.

Templates have exactly one failure mode, and I'll get to it.

Where AI earns its keep

The good uses all have the same shape: something has to be read before anything can be said.

Triage. Fifteen inquiries land in a week and roughly half are tire-kickers. Something that reads each one and pulls out the budget, the timeline, the location, and what they actually want, so you're scanning six lines instead of six paragraphs. You still decide. You decide faster.

Turning a mess into notes. A voicemail transcript from Otter, or the note you thumbed into your phone in a truck, becomes a clean list of what the job involves and what you still need to ask.

First drafts of hard replies. A real complaint has different facts every time — no template survives it. AI's value here isn't the writing, it's getting you past the blank screen while you're still irritated. You'll rewrite half of it. That's fine. Half is the part that was hard.

Tone repair. You wrote a true sentence at 6pm that reads meaner than you meant. "Make this warmer, keep it short." Two seconds, and you didn't have to sit there rewording it four times.

Catching up on a long thread. Nine messages deep with a customer, and you need to know what you actually committed to. That's reading and summarizing, which is the thing these tools are best at.

Both mistakes cost you something

Point AI at your fixed answers and you've converted a settled fact into a probability. Ninety-nine times out of a hundred it describes your service area correctly. The hundredth time it says something plausible and wrong, and you find out when somebody's standing in a driveway you don't serve. You took on risk in exchange for nothing — the answer was already written.

Point a template at a variable email and the customer can feel it. Everyone has received a canned reply to a specific complaint. It doesn't read as efficient, it reads as you didn't read my message. That template didn't save you ten minutes. It created a second, harder email you now have to send.

What I'd set up

It's less impressive than you might expect.

Eight to twelve saved replies for the questions with fixed answers, named so you can find them in a hurry. Gmail templates or a text expander — whatever's already in front of you.

One custom GPT loaded with your pricing, your service area, your FAQ, and a handful of your own past emails so it sounds like you. It handles the read-first pile. It drafts; it does not send.

And one written rule about where it isn't allowed to go: anything with a number in it, anything with a promise in it. Quotes, dates, and commitments get typed by a human. Every time.

That's the whole system. Most of the work is deciding which pile each email belongs in, which is the part only you can do.

There's a third category

Some emails shouldn't be answered by a template or a model. An angry customer. A deal that's wobbling. Anything where the next sentence decides whether the relationship survives.

Those are phone calls. Neither tool helps you there, and reaching for one is usually what turns a fixable problem into a lost client.

Which is the point of automating the rest of it. Not to answer more email. To have enough attention left for the four messages a week that deserve all of it.

If you want to see where your own inbox time goes first, the free AI Time-Leak Checklist is a good place to start.

---

Sorting your inbox into those piles is most of what an AI workflow audit is. Tell me what your week looks like and I'll give you a straight answer about whether you need AI at all or just a better set of saved replies.