What to Automate, and What to Always Keep Human First

Here is the line, and it decides every automation question you will ever face:

Automate the machinery. Keep the meaning.

The test is one question: are these words someone’s genuine expression? If they are functional — a caption, a hashtag, an announcement — hand them to the machine. If they are genuine — a customer’s testimonial, a post you sincerely mean — they stay human.

Machinery to the machine. Meaning stays yours.

Every “should I automate this?” resolves to that one distinction. You never have to agonise, and you never have to choose between “all automated” (soulless) or “all by hand” (unsustainable). You sort each thing by whether it carries genuine meaning.

The two categories

Everything you post falls into one of two bins:

Machinery — functional content and mechanical work. It needs to be correct and done, not felt:

• Captions (your voice, but functional)
• Hashtags
• Subtitles / transcription
• Formatting for platforms
• Distribution
• Routine announcements (hours, specials)

Meaning — content whose entire value is that it is genuine:

• A customer’s testimonial (their genuine words)
• A sincere free-text post (your genuine words)
• The capture itself (a real moment)
• The judgment (who to ask, what is worth posting)

The machinery bin goes to the machine, completely and without guilt — that is what frees you to run the business. The meaning bin stays human, absolutely — because meaning cannot be automated without being destroyed.

Why meaning cannot be automated

The reason is not sentimental. It is mechanical.

Genuine words work because they are genuine — a stranger can feel that a real customer meant her testimonial, that you meant your thank-you. The moment a machine generates them, that quality is gone, and gone invisibly: it still looks like a testimonial, still reads like a heartfelt post, but the thing that made it persuasive — that a real person genuinely expressed it — is absent.

So automating meaning does not give you a slightly-worse version. It gives you a hollow one that looks fine and works not at all. A testimonial that reads better than the customer speaks is a fake one; a heartfelt post an algorithm wrote is a lie about your feelings. Both are automation crossing the line from machinery into meaning, and both fail precisely where it matters.

The test, applied

Run any piece of content through the question — is this someone’s genuine expression?:

Content	Genuine expression?	Automate?
The caption	No, functional	Yes
The customer’s testimonial	Yes, hers	Never
Hashtags	No, mechanical	Yes
A sincere thank-you post	Yes, yours	Never
Subtitles of what she said	Her words	Transcribe, never rewrite
Holiday-hours announcement	No, functional	Yes
Your apology after a mistake	Yes, yours	Never
Formatting and distribution	No, mechanical	Yes

The pattern is clear: functional → machine, genuine → human. You never guess; you ask whether the words are someone genuinely expressing something.

The subtle case: your own genuine words

Most people grasp “keep the customer’s testimonial human.” The under-appreciated half is that your own genuine words are equally protected.

A caption is yours and functional — automate it. But a post you mean — the bereavement notice, the real apology, the heartfelt thank-you — is yours and genuine, and it must be written by hand and published verbatim, exactly as you wrote it. The machine does not tidy it, because the tidying would remove the sincerity that was the whole point.

So “human” is not only about the customer. It is about genuine expression, whoever’s it is — the customer’s words and your meant words, both kept human, both published as spoken or written. That is the Pen Rule pointing in both directions.

A Saturday at a florist, sorted

Picture a florist on a busy Saturday. A regular collects a bouquet, pauses at the counter, and says — half to herself — “I don’t really know flowers, but yours are the only ones that last me a fortnight.” It comes out a bit clumsy. That clumsiness is meaning. It is hers, and the small hesitation is exactly what makes a stranger believe it. If she wants to share it, she asks with a short script that hands the words back, records it, and posts it as spoken. She does not tidy “I don’t really know flowers” into “Stunning quality, always fresh.” The tidy line reads better and persuades no one, because nobody actually talks like that.

An hour later she photographs the week’s new peonies for a Sunday post. The photo is genuine — she chose it — but the caption under it, “Fresh peonies in, £18 a bunch,” is machinery. Let the machine write it, tag it, size it for each platform, and schedule it. Her judgment went into which flower was worth a picture; the caption is just plumbing.

Same shop, same afternoon, two pieces of content, sorted by one question. The customer’s clumsy line stayed human because it was genuine. The peony caption went to the machine because it was functional. Neither call took a second of agonising.

Where AI fits inside this

AI is a tool that sits on the machinery side — it may write captions, hashtags, promotional copy, all the functional content, and it is welcome to. It may never write, rewrite, or improve genuine expression: not the testimonial, not the sincere post.

The line is not “AI good” or “AI bad”. It is the same machinery/meaning line: AI may do the machinery, never the meaning. Point it at the functional and it saves your evenings; point it at the genuine and it destroys your credibility. Same tool, opposite outcomes, decided entirely by which side of the line you aim it at.

The line is also the product

This is not just marketing advice; it is the principle a whole product is built on, so it is worth stating plainly.

A business is made trustworthy by real things — real customers, real words, real feeling. Automation that handles the machinery lets a small business be consistent and present without a marketing team. Automation that reaches into the meaning — faking testimonials, generating sincerity, inventing reviews — quietly dismantles the trust that was the entire asset.

So the machine writes the caption and spares your evening. It never writes the testimonial, because that would spare you the one thing you could not afford to lose. Keep the machinery automated and the meaning human, and you get the efficiency without the hollowing-out.

But automating the testimonial would be faster

Of course it would. You could have an AI write ten glowing testimonials before lunch, and they would all read beautifully. That is exactly the trap. The speed is real; the value is gone. A generated testimonial costs nothing to make and earns nothing to post, because the one ingredient it needed — that a real customer meant it — is the one ingredient a machine cannot supply. A made-up testimonial is not a fast testimonial. It is a fake one, and a stranger’s gut reads fake faster than you would like.

There is a quiet mercy hiding in the slower, human way. When a customer would rather not be quoted, that “no” is the filter working: the weak testimonial simply never gets made. Automate the ask and you override that filter — you start manufacturing the exact thing it exists to stop. Keep the capture human and every “no” keeps quietly protecting you.

Sort your content by the one question

For everything you post, ask: is this someone’s genuine expression?

If no — caption, hashtag, format, distribute, announce — automate it, freely, and get your evenings back. If yes — the customer’s words, the moment, the thing you mean — keep it human, always, and publish it exactly as spoken or written.

Machinery to the machine. Meaning stays yours. That single line answers every automation question there is.

How automation gives you the time back — social media as a by-product, not a job — is the piece beside this.