AI as a Production Partner

When growth is often measured at the last click, you’re paying to compete for demand that was created somewhere else.

Reach people in the purchase planning phase before your competitors know these customers even exist.

With high-intent Pinterest signals on Performance TV you can reach audiences earlier where they watch the most.

Get an advertising advantage

Start Your Campaign

The last edition was about using AI to think. This one is about using AI to produce, which is the use everyone already reaches for, and most teams are getting wrong.

Here is the problem, stated honestly. Audiences in 2026 have learned to detect AI-generated content. They have learned the patterns. The smooth, confident, slightly hollow tone. The tidy structure. The complete absence of any genuine specificity or any real point of view. And when they detect it, they disengage. We covered the data in an earlier edition. A majority of consumers reduce trust and engagement when they recognise content as AI-generated.

So most marketing teams are now in a trap. They adopted AI to produce more content. They are producing more content. And the content is performing worse, because more of it reads as generic, and audiences are routing around generic.

The answer is not to stop using AI for production. The answer is to use it correctly, in a specific way that produces work that does not read as AI. There is a real method here. Here it is.

The single biggest cause of AI slop is asking AI to generate content from nothing. "Write a LinkedIn post about customer retention." With no input, the model has nothing to work from except the average of everything ever written about customer retention, which is the definition of generic.

The fix is to never let AI invent the substance. AI should only ever be transforming raw material that you supplied.

The workflow. Before you ask AI to produce anything, you produce the raw material. This can be messy. It can be a voice memo. It can be five bullet points of what you actually think. It can be a transcript of you talking through an idea for three minutes. It can be the pattern analysis from last edition's customer data exercise. The point is that the substance, the actual insight, the specific example, the genuine opinion, comes from a human and from real experience.

Then the prompt becomes:

"Here is a rough three-minute transcript of me talking through an idea about customer retention, including a specific example from our own data. Turn this into a LinkedIn post. Keep my specific example and my actual opinion intact. Do not add generic advice. Do not smooth out the point of view. Match the tone of the writing samples I am about to give you."

Now AI is doing what it is genuinely good at, shaping, structuring and tightening. It is not doing what it is bad at, which is having something to say. The substance is yours. The labour of formatting is the AI's. The output does not read as slop because the slop comes from substance-free generation, and you removed that step.

Teams tell the AI to write "in a friendly, professional, conversational tone" and then wonder why everything sounds the same. Adjectives do not carry voice. Voice lives in sentence rhythm, in specific word choices, in how you open and close, in what you refuse to say.

The fix is to never describe your voice. Show it.

The workflow. Collect your best work. Five to ten pieces, your highest-performing posts, your best newsletter sections, the emails that actually got replies. Paste them into the prompt and instruct:

"Below are eight examples of my writing. Study the sentence rhythm, the typical openings, the level of formality, the way arguments are structured, and the words and phrases that recur. Do not summarise the style back to me. From now on, when you draft, match this voice precisely."

Better still, if your AI tool supports saved projects or custom instructions, store these samples permanently so every draft inherits the voice. We covered this in the prompting edition. The good setup is built once and reused.

The difference this makes is large. AI given adjectives produces the generic average. AI, given twenty real sentences of your actual writing, produces something a reader would plausibly believe you wrote.

Not all production is created equal. Some of it requires taste and judgment. A lot of it is purely mechanical, and the mechanical parts are where AI should carry the full load with no guilt.

Mechanical production tasks, hand these fully to AI. Resizing one piece of content into platform-specific formats. Generating fifteen subject line variations to test. Turning a long-form piece into a structured outline for a different format. Producing first-draft alt text. Reformatting a transcript into clean paragraphs. Creating the first pass of a meta description. These tasks have a correct answer or a clear spec. There is no voice or taste being lost. AI should own them entirely.

Judgment production tasks, AI assists but does not lead. The hook of a piece. The core argument. The specific example that makes a point land. The opinion. The decision about what not to say. These are where the value of the content actually lives, and these should stay human-led, with AI used only to pressure-test or offer alternatives, never to decide.

The teams that get production right are ruthless about this split. They do not feel they must use AI for everything, and they do not refuse to use it for anything. They use it fully for the mechanical 60 per cent and barely for the judgment 40 per cent, and that division is what keeps the output from sliding into slop.

Every piece of AI-assisted content needs a human edit before it ships. But "edit it" is too vague to be useful. The edit has a specific job, and the job is to put back in the things AI strips out.

AI drafts tend to be smooth, complete, and slightly dead. The edit is where you re-inject life. Concretely, the editing pass should do four things. Add one specific detail the draft is missing, a real number, a real name, a real moment, something only your company could have written. Sharpen the opening so it does not sound like every other opening. Cut the one sentence that is pure filler; the AI draft always has at least one. And read it aloud, because anything that you would not actually say out loud is the part that reads as AI, and the read-aloud test catches it instantly.

This edit takes five to ten minutes. It is the difference between content that ships as forgettable and content that ships as genuinely yours. Skipping it is the single most common reason good teams still produce slop.

Take your next piece of content and produce it the right way.

Do not start by prompting AI. Start by recording yourself, a voice memo, for three minutes, talking through what you actually want to say, including one specific real example. Then give AI that transcript plus eight samples of your best previous writing, and ask it to shape, not invent. Then run the four-point human edit. Then ship it.

Compare it honestly to the last piece you produced by asking AI to write from a blank prompt. The difference will be obvious, and once it is obvious, you will not want to go back.

The principle underneath all of this. AI is an extraordinary production partner and a terrible production replacement. It should transform your raw material, never originate the substance. Used that way, it makes you faster without making you generic. Used the other way, it makes you faster at being forgettable.

Next edition, the final one in the series. We move from producing individual pieces to running the whole marketing function. How to use AI to build the operational systems and workflows that run in the background, so the team spends its time on judgment, and the machine handles the rest.

See you at the next edition, Arindam