AI is an accelerator. Not a substitute for expertise. What really happens before AI delivers a result. (Part 2)

Junai.life: what “AI prework” looks like in practice  

Say we are preparing ads for Junai. 

In the end, the user often sees only the ad: a visual, some copy, maybe a short video. They do not see what happened before. 

They do not see the thinking. They do not see the decisions. They do not see all the questions we need to clarify before AI even starts creating. 

And that is usually where the difference starts between an ad that is merely “nice” and an ad that performs. 

First we need an idea that carries weight 

We are not just looking for an “ad idea.” We think as if we had the problem ourselves. We put ourselves in the shoes of the person the product is for. 

Example: we have the product Junai HIM, which communicates energy, strength, and recovery. Before we go to AI, we need to know: 

• Ad objective: are we aiming for awareness, consideration, or purchase. The product can be the same, the persona too, but the objective is different. 

• Funnel stage: are we addressing a cold audience that does not yet know Junai, or retargeting someone who has already visited the site. 

• Product USP: why Junai HIM and not the competition? Is the difference in ingredients, origin, certifications, formulation, trust? 

• Who is the user: a man who already does something for himself. Not a “couch potato,” not a “hardcore bodybuilder.” 

• What “hurts” him: his energy is dropping, the routine is getting harder, he feels that “it used to be easier.” 

• What he does not want: coffee or energy drink crashes, bloated marketing, “bro-science” communication. 

• What he wants: stability, strength, focus, and the feeling “I am myself again.” 

• Tone: direct, no saccharine, premium. 

• Connecting AI to existing documents: before AI even starts working, we must give it the right context. That also means brand guidelines, tone of voice, visual identity, product sheets, previous campaigns, competitor analysis, and everything else that defines who the brand is and how it communicates. 

Only when this is clear do we go to AI. 

Next comes the creative: image and/or video  

For AI to prepare an image or video well, we must give it clear inputs: 

• Scenario: what happens in the frame. 

• Mood: what feeling the creative needs to convey. 

• Style: the brand’s aesthetic and reference world. 

• Composition: where the product is, where the person is, and what is in the foreground. 

• Format: aspect ratio, dimensions, platform, and placement. 

• Do’s and don’ts: what must be included and what must not. This part is most often skipped—and most often ruins the result. 

• Reference aesthetics (descriptive): describe aesthetic directions in words or, if possible, show them visually, not with links. If we do add a link, ask it whether it can see the photo and to describe it. That way we make sure it will not miss the point. 

Then comes the copy: opening hook, main message, and call to action 

AI can write 50 opening hooks in 15 seconds. But without the right context they are useless. We need to give it clear inputs: 

• What we are testing: for example, a coffee alternative, energy without a crash, more strength after 35.

• What is the proof or reason to believe: why the reader should trust the message.

• What is prohibited: no impossible promises, no medical or non-compliant claims, no exaggeration.

• What lengths are allowed: for example, primary ad text, headline, description. Every format has its own rules. 

• Brand tone: AI does not know that Junai HIM does not speak like a generic “fitness supplement.” We must tell it explicitly: premium, no saccharine, direct. If we do not, it will very quickly slide into generic fitness copy. 

If it does not have these inputs, AI will indeed write 50 opening hooks. But they will likely be texts we would never publish. 

Next comes the part most people do not even classify as “AI”  

Once the creative is done, the part starts that is invisible yet often decisive. 

This is the technical layer. Here we are no longer talking only about whether the ad looks good. We are talking about whether the campaign is set up so it can be measured, understood, and improved. 

That means: 

• Proper campaign structure: a clear hierarchy that enables clean data and meaningful comparison of results. 

• Consistent naming: a consistent naming logic for campaigns, ad sets, and ads so results are readable and comparable. 

• UTM logic: without it we do not know precisely what drove the click, visit, or conversion. 

• GA4 events: events such as add to cart, begin checkout, and purchase must be configured correctly. 

• Consent mode and CMP: incorrect consent configuration can cut off a large share of signals and skew the data. 

• Product feed: if we use it, it must work for the right markets, the right products, and the right campaigns. 

• Linking creative to data: we must know which ad actually delivered the result, not just which got the most clicks. 

• Infrastructure and development: server, hosting, CDN, production deployments, back-end systems, databases, APIs, automations, security access, logs, performance, landing page, article, or even ad generators. 

This last part is usually built by a developer. But we must be able to specify what the system needs to measure, how it should operate, and what the objective is. 

Without this, a developer can build a system that appears correct at first glance. The problem shows up later, when we realize it measures the wrong things, the data are not connected, or the results cannot be interpreted reliably. 

When we get to the technical layer, the difference becomes clear between a “nice ad” and a system that actually sells. 

Without tracking and data, an ad is just a poster. 

This is where results are decided. 

Time can save you money 

Just the prework for a single campaign, without production and without ad setup, can take 4–8 hours or more of serious work. 

Most of this work is invisible. The client does not see it, the ad does not show it, analytics does not measure it directly. Yet it is precisely this part that determines whether the campaign will work or not. 

So when someone says: “AI will do this in five minutes,” it is probably true. But those five minutes do not include the prework.  

To move beyond the abstract, let’s look at how this plays out in practice.

Concrete example: poor prompt vs. excellent prompt 

Let’s look at the difference on a concrete example. Say we want to prepare an AI image for an ad for Junai HIM.

Example 1: AI image for an ad (Junai HIM) 

Poor prompt 

“Create a modern image of a man with a lot of energy, promoting the Junai HIM dietary supplement. Make it premium.” 

What will AI most likely do? 

• a generic “stock” man without a story,

• a generic “premium” aesthetic—i.e., what AI itself interprets as premium,

• an image without clear context and composition, 

• the classic cliché: dark background, blue lighting, and randomly emphasized muscles. 

At first glance, the result may look solid. But it will not necessarily connect to the brand, persona, product, or the actual reason why a user would click. 

Excellent prompt (enhanced for branding + product sharpness) 

“Create a photorealistic ad visual in a vertical 4:5 format (1080 × 1350) for Meta Ads. 

Brand: Junai—that means premium, minimalist, European quality, with a clean and confident feel. 

Product: Junai HIM bottle, a dietary supplement for male energy, strength, and recovery. 

Mandatory: the product must be the central element of the visual and the sharpest element in the frame, as in a professional product ad. The bottle must have high sharpness, clear details, and good micro-contrast. 

The label must be fully legible: text must be clear, without distortion, blur, or wrong characters. 

The bottle must not have generic logos, fake textures, or elements that are not consistent with the brand. It should feel realistic, premium, and visually refined.” 

Typography, fonts, and branding on the label (strict) 

• Label: typography on the label must follow Junai brand guidelines. Gabriela is used for headlines and display elements, and Montserrat for supporting text and descriptions. Gabriela must take the lead in the frame, while Montserrat functions as the supporting element: visibly smaller, lighter, and less dominant. No substitute fonts, no improvisation, no approximations. Only these two fonts may be used, in this hierarchy. 

• Alignment and proportions: the wording on the label must be aligned, visually balanced, and set with sufficient white space so the label retains a premium feel. 

• Label colors: the label should follow the Junai brand identity: dark green base and cream or white text with high contrast and a premium look. 

• Style: the visual language should feel like a European premium nutraceutical, not a generic “sports supplement.”

Scene (real and commercial, not artificial) 

Morning in a tidy, minimalist apartment. 

• A man aged 30–50, active and well-groomed, but not a bodybuilder. Dressed in neutral clothing, e.g., white/gray. 

• He stands by a kitchen counter, holding a glass of water. 

• The product sits on the counter in the foreground. The hand and the product are in focus, while the face is slightly out of focus so the visual does not feel like an influencer selfie. 

Composition (ad logic and safe area) 

• Place the product in the lower third or bottom right, with a “premium” safe area. 

• Leave a clean empty area in the top third (for potential ad copy), but do not add text onto the visual. 

• Perspective: camera roughly at eye level, with a natural 35 mm or 50 mm lens look, without extreme angles. 

Lighting and textures  

• Lighting should feel natural and soft, like morning light coming from the side through a large window. Tone should be slightly warm, with very subtle cinematic grain. 

• Materials: stone/wood (no visual clutter). 

• Add subtle realistic details: a slight reflection on the bottle, a natural shadow, and realistic depth of field (DOF). 

Prohibitions (to stay Junai) 

• No neon effects, tech glow, “gym bro” aesthetics, or overly emphasized muscles. 

• No aggressive visual language, screaming colors, or too many props. 

• No distorted, nonsensical, or artificially generated text on the label. 

• No watermarks, additional logos, or elements that are not part of the Junai brand. 

Final look 

• The look should feel like a premium editorial product ad, with a sense of calm strength, similar to a luxury wellness or technology ad. 

• The visual should be photorealistic, well executed, with a clean composition where the product is the central element. 

Why is this better? 

Because AI is not guessing. It gets clear context, objective, persona, composition, aesthetics, technical requirements, prohibitions, and format. It no longer creates by feel, but by guidance tied to the brand, product, and the ad’s purpose. That is why the result is not just a “nice picture.” 

The result is a visual with a clear reason for why it looks the way it does. It knows whom it addresses, what it must show, what it must avoid, and how it must perform to stay consistent with the Junai brand. 

That is the difference between a generic supplement ad and an ad that actually looks like Junai. 

And this is precisely the essence of AI prework: AI does not deliver a better result because we wrote a longer prompt. It delivers a better result because we gave it better inputs. 

What makes an output “wow,” and why is that not a coincidence? 

An excellent AI output always has the same feel: as if someone truly understood the brand, as if someone understood the person, as if someone knew the goal.  
 
And that does not happen by itself. There is work behind it and a defined standard. Not just “I like it” or “I don’t like it.” You need to know whether: 

• it is relevant to the persona?

• it follows the brand’s voice?

• the format is used correctly?

• the idea is strong enough to stop the scroll?

• it can be tested (more than one angle)?

• it is compliant with what you are allowed to claim? 

AI can produce 100 variants. But it is the human who must know what is good, brand-aligned, and meaningful for the campaign. 

The value is not in the number of variants, but in choosing the right ones. 

The technical layer: without it, AI is just nice text 

And now the part many ignore because it is not “sexy.” Yet this is exactly where the main reason lies why some projects feel professional and others like improvisation. 

When we ship an ad, the work is only beginning: where does the click lead, is the landing page fast, does tracking really measure purchase (purchase), begin checkout (begin_checkout), and add to cart (add_to_cart), does the consent mode cut off half the signals, are the UTMs correct, are audience signals meaningful, does the product feed, if we use it, work for the right markets—and above all: can we explain the results and act on them. 

And this is where the “invisible” power comes in: infrastructure programming. Servers, hosting, content delivery network (CDN), deploy pipeline, back-end systems, databases, APIs, automations, security access, logs, performance optimization, landing-page or content generators — this is the engine that makes marketing measurable and scalable in the first place. 

AI can help generate the event specification, write the naming logic for GTM and GA4, suggest campaign structure, prepare copy variations or text modules for the landing page. 

At Humanfrog we use AI exactly this way: as an accelerator for the technical work, not a substitute for expertise. 

However, without the fundamentals, AI will not solve the technical part. It will just write something that sounds correct, until we discover nothing is being measured or it is measured incorrectly. No AI can replace technical knowledge and experience. Because AI does not know what it does not know. It does not know our data, does not understand our system, and does not see where the signal is lost. 

This is exactly where Humanfrog enters the picture. We know which questions to ask, where the data get lost, and what the right objective is. AI is used as an assistant, not as the decision-maker. And that balance is what separates a campaign that sells from a campaign that merely exists. 

The truth is simple: AI can accelerate the technical work, but it cannot replace it. In infrastructure, it is not enough that code exists. It must be correct, secure, fast, measurable, and built around the right objective. 

Clarity between pilot and system 

The best AI results have a common denominator: clarity. 
Clear instructions. Clear objectives. Clear understanding of who we are. Clear understanding of what we do not want. 

If the communication between the pilot (you) and the system (AI) is fog, the output will be fog too. 

And then we wonder: 

“Why is this generic?” 
Because we gave AI generic instructions. 

“Why is this odd?” 
Because we left too much room for interpretation. 

“Why are some pages ‘wow,’ and others poor?” 
Because in the first case there was a pilot at the controls. In the second, a tourist. 

AI is not a shortcut. AI is an accelerator. 

AI can save us hours in production. But it cannot save us the thinking, the strategy, the standards, the brand intuition, the audience understanding, the platform understanding, and the technical knowledge. AI does not make a project professional. The process is what is professional. And within that process, AI is just the airplane. We are the pilot. You are the pilot. 

That is exactly why AI prework is so important. When the cockpit is set up correctly, the result is “wow.” When it is not, it can be awkward. And in advertising, “awkward” is expensive.