Marketing

AI-generated content for businesses: what should you hand to AI, and what not?

By Equipo ZENITAR, Consulting and execution10 min read
Camera operator filming on a lit set during a brand content production

AI is fast at the repeatable work: voice-over, animation, template pieces and catalog scenes. Strategy, brand shoots and final review still belong to people. How to split the work without losing the brand.

AI-generated content is produced with AI tools for text, image, voice or animation. A person directs it: they decide what is said, to whom, and under which brand.

The useful question is not whether to use AI. It is which part of the work goes to the machine and which part stays with a person. Companies that get that split wrong end up with a lot of content and very little brand. This note sorts the split piece by piece, based on what we do at ZENITAR for clients in Costa Rica.

01

Which content tasks should you hand to AI?

Hand over the repeatable work. That includes pieces built on a template, variations of one design, voice-over for explainer videos, animation and background scenes for product photos. In these tasks quality depends on following a rule, and AI follows rules quickly and without getting tired.

In practice, AI pays off where volume is high and the judgment has already been made. These are the cases where we use it every month:

  • Highly repeatable designs: promotions, special dates and ads that change the copy but not the layout.
  • Voice-over for service videos, when the voice of someone from the company is not needed.
  • Animation of static pieces: logos, charts and photos that gain motion for social media.
  • Scenes and backgrounds for product photos, always built on the original product photo.
  • First drafts of scripts, calendars and short copy, which a person then corrects.

The real saving is not making more pieces. It is freeing team hours for what AI cannot do, which is the next section. For a company that sells in Costa Rica or abroad, those hours go to the shoot and the strategy. They also go to the customer conversations that no tool can hold.

02

What does a person still have to do?

A person decides the strategy, the message and the tone. They also direct the shoots with the company's real people and approve every piece before it goes out. AI does not know your business, is not accountable for what gets published, and cannot tell a brand error from a typo.

Some content loses all its value if a machine makes it. A testimonial, a tour of the plant or the team serving a customer work because they are real. That is why at ZENITAR the shoot with a camera operator and a script is still the core of brand content, and AI works around it.

What AI does and what a person does, piece by piece
PieceWhat AI doesWhat a person doesWhen not to use AI
Brand videoSubtitles, crops per formatScript, shoot with a camera operator, final editWhen the company's real people appear
Explainer videoVoice-over and animationScript, fact and tone reviewWhen the topic is sensitive or legal
Promotions and datesVariations of an approved templateTemplate design and approvalWhen the piece launches something new
Catalog product photoScene, surface and settingOriginal product photo and quality controlNever to regenerate the product
Organic contentIdeas and first draftsSelection, brand voice and replies to commentsDuring a crisis or a complaint
Testimonials and casesNo part of the contentEverythingAlways: they have to be real

The right-hand column is the one that protects a brand the most, and it is the one most often ignored when a team discovers how fast AI can produce. What is not handed over is exactly what makes a customer trust you.

Human review is not a formality either. A generated voice-over can mispronounce a product name, and a scene can put the logo in a color the brand never uses. They are small errors that a customer notices before anyone else. That is why every AI-made piece goes through a person who checks it against the brand book and the month's message.

03

How do you build a monthly content plan with AI, shoots and organic posts?

A monthly content plan splits the month into three layers. The first is a shoot day with a camera operator that feeds several weeks. The second is repeatable pieces made with AI on the brand's templates. The third is day-to-day organic content. Everything goes through human approval.

This is what a typical month looks like when the plan is well built:

  1. First week: the month's message is set, scripts are written and the shoot is scheduled.
  2. Shoot: one day with a camera operator produces the brand footage for the whole month.
  3. AI production: voice-over, animations and template variations are generated from that footage.
  4. Review: a person approves every piece against the brand book before scheduling it.
  5. Close: the team measures which pieces brought inquiries and adjusts next month's plan.

Order matters. If AI produces before the month's message exists, the result is volume without direction. In how a marketing agency works we explain which metrics that monthly close is reviewed against.

Whiteboard with colored sticky notes used to plan a monthly content calendar
The month's plan is decided before AI produces a single piece.

04

How do you design an online catalog with hundreds of products?

An online catalog with hundreds of products is designed like a production line. The original photo of each product is kept, AI generates the scene and the surface, and a fixed process checks every image before delivery. Without that process, the catalog ends up with mixed styles and distorted products.

Online stores with large catalogs have a scale problem: every product needs an image that sells, and a photo shoot per product is not viable. For an online perfume store, ZENITAR built an AI editing farm that produces the catalog images in batches of 5 pieces, over a first run of 200.

The farm's rules show where the human value sits:

  • The product is always the original photo. AI never regenerates or redraws it.
  • AI generates the scene, the surface and the ingredients that surround the product.
  • Surfaces rotate: never the same type in two consecutive pieces, or the same material within 12.
  • Every piece comes out at 1024 by 1024 and passes a position check before delivery.
  • No image is regenerated without a person's order.

The result is a consistent catalog that grows at the pace of the inventory. If the catalog also has to live inside the website, with product pages, search and an inquiry button, that part belongs to web development and online stores.

05

Why does a brand book keep AI from drifting?

A brand book gathers a brand's logo, colors, typefaces, tone and allowed uses. With AI it becomes more necessary, because it is the written rule every generated piece is checked against. Without it, each tool interprets the brand its own way.

AI does not remember your brand from one month to the next. Every time it produces, it starts from what it is told. That is why a brand book does two jobs at once: it guides the people who design, and it serves as a checklist for what the machine generates.

A brand book that works with AI includes, at a minimum:

  • Colors with their exact code and the proportion in which each one is used.
  • Typefaces for headings, body and secondary text.
  • Correct and incorrect uses of the logo, with examples.
  • Tone of voice: how the brand speaks to customers and which words it never uses.
  • Photography and illustration style, which is the direct guide for generated scenes.

On the marketing agency page you can browse a complete brand book, exactly as a client receives it.

06

Does Google penalize AI-generated content?

No. Google stated in February 2023 that appropriate use of AI is not against its guidelines. What it penalizes is content produced to manipulate search results, whether it is made with AI or without it. The criterion is the same as always: the content has to be useful to a person.

The position is in Google Search Central's post on AI-generated content. Its guide to helpful, reliable content adds a practical question: if a reader might wonder how a piece was made, it is worth saying so.

Google's guide to its generative AI features is even more direct. It asks site owners not to recycle what others have already published, or what any AI model could easily produce. Mass production aimed at manipulating results falls under its scaled content abuse policy.

For a company in Costa Rica, the reading is clear. Any tool can write a generic article on "the benefits of digital marketing" in seconds, so it outranks no one. A note with the prices your sector actually handles, a photo of your warehouse or a client's measured result cannot be copied by anyone.

The conclusion for a business is concrete. AI content, yes, but with a fact, a photo or an experience that only your company has.

07

How do you measure whether AI content is selling?

Measure it with the same business metrics as any content: inquiries received, qualified leads, attributable sales and cost per new customer. The number of pieces published is not a metric. If AI doubled production and inquiries stayed flat, the problem is the message.

There is one more reason to measure well. Google reported in 2026 that its AI Overviews have more than 2.5 billion monthly users. More and more people read a generated answer before visiting a site, and every page competes to be one of the few sources that answer cites.

Three questions are enough to review the month:

  • Which pieces generated real inquiries, and which only generated reach?
  • How much of the budget went to production and how much to distribution?
  • Did this month's content repeat what competitors say, or show something of your own?

If you want to know what it costs to sustain a plan like this, what a marketing agency costs in Costa Rica explains what sets the price. And if you would rather review your case directly, the ZENITAR team builds the proposal from what your company already has.