The Marketing Workflows Worth Automating, and the Ones That Are Not

The more you automate, the more the editing pass matters.

I Tried to Automate a Whole Site

Not a workflow. The whole thing. SEO, AI optimization, briefs, drafts, internal linking, the lot.

A lot of it worked, and it worked well enough that I kept going longer than I should have. Output went up and publishing got faster.

The mechanical parts of SEO turned out to be genuinely mechanical.

Then I read a month of it back in one sitting, and it was competent, and it was completely anonymous. Any of a thousand sites could have published it.

What Automation Is Genuinely Good At

The tasks worth handing over share a trait: there is a correct answer, and finding it is tedious rather than difficult.

Research and gathering. Pulling search terms, clustering keywords, collecting what already ranks. This is the strongest use and it is not close.

Briefs. A brief is structure plus inputs. A machine builds one faster than you do and it will not skip a step because it is Friday.

First drafts of mechanical content. Product descriptions, spec pages, location pages, anything where the job is arranging known facts clearly.

Checking and monitoring. Broken links, missing alt text, schema errors, a page that fell off page one. Machines do not get bored, and boredom is why these go unchecked.

Formatting and repetition. Reformatting, tagging, internal linking, metadata. Pure overhead, gone.

What It Is Bad At, and Why

Automation flattens voice. Not immediately and not obviously, which is the problem. It happens one reasonable sentence at a time until everything you publish sounds like everything everyone else publishes.

The reason is structural. A model is built to produce the most probable next thing, and the most probable thing is by definition the average thing. Average is exactly what you do not want when you are trying to be chosen over a competitor.

So the list of what not to automate is short and it is all the same category:

Anything that carries an opinion. If the piece exists to argue something, you write it. That is the whole value.

Anything drawn on real experience. What actually happened on a client account, what you got wrong, what you would do differently. A model cannot have been there.

Anything a customer reads at the moment they decide. Your service pages, your pricing, your about page. That copy needs a human behind it, because people can feel when there is not one.

Your point of view on your own industry. If you outsource that, you have outsourced the reason anyone would hire you.

QA Stops Being Optional

Here is the part people underestimate when they start automating.

The more you automate, the more editing matters, and the editing gets harder rather than easier. Reviewing machine output is not proofreading. You are checking claims, checking that it still sounds like you, and checking that it says something rather than merely saying words.

Most people set up the automation and quietly stop doing that second part, because the output looks finished. Looking finished is the trap. It is grammatical, structured, confident, and hollow, and hollow passes a proofread every time.

If you cannot commit to the review, do not build the automation. You will ship more, worse, faster.

What Actually Separates You Now

Here is what I believe after all of it.

When everyone can produce competent content instantly, competent content stops being worth anything. The floor came up and it came up for your competitors at exactly the same moment it came up for you.

What is left is having a point of view. Knowing what you believe about your industry, being willing to say it plainly, and being able to explain why. That is the thing a model cannot generate for you, because it requires having been somewhere and concluded something.

Automate the tedious half. Spend the hours you get back on the half that only you can write. That is the whole strategy, and it is harder than it sounds, because the tedious half is comfortable and having an opinion in public is not.

Where to Start

Pick the two most repeated jobs in your week. Automate those, keep a real review step, and see what a month of it looks like before you automate anything else.

If you want help working out which two, that is what the first conversation is for.