Generative AI Can Make Video. But Can It Make Advertising?

A Video Experiment with Generative AI

I asked AI to make an advertisement for my own company.

That sentence would have sounded absurd a few years ago. Today, it sounds almost routine. Open an AI video platform, provide a prompt, wait a little while, and out comes something with stock footage, voiceover, music, graphics, transitions, and a logo reveal. In other words: something that looks like an ad. Here’s the video: 

For this experiment, I used a GPT inside ChatGPT called InVideo to create a short promotional spot for Metamer Studios, our creative production company. The goal was straightforward: generate a professional-looking advertisement that could communicate who we are and what we do.

The result was fascinating.

Not because it was a disaster. It was more useful than that.

The video had the recognizable shape of a modern marketing spot. It used cinematic imagery, fast pacing, production-related visuals, a musical underscore, a voiceover, and a dramatic brand reveal. At a glance, it looked like an advertisement. But the closer I watched, the clearer the problem became: generative AI is now very good at creating ad-shaped objects.

That is not the same thing as creating advertising.

And for companies trying to understand where AI fits into their marketing or video workflows, that is an important distinction.

The ad looked polished — until you actually watched it

The finished spot was roughly 39 seconds long and built around the familiar visual grammar of a video production company: cameras, editing timelines, monitors, sound meters, production environments, green screen talent, and dramatic animated brand moments.

There were flashes of polish throughout. Some of the shots looked professional in isolation. The pacing had energy. The overall structure followed a recognizable arc: hook the viewer, show production-related imagery, build toward a call to action, and finish with the company logo.

That surface-level competence is exactly what makes these tools so interesting. They can generate something quickly that appears to belong in the world of marketing. For internal ideation, mood boarding, or early-stage concept exploration, that has value.

But once the video is judged by the standards of actual advertising — not just content generation — the weaknesses become impossible to ignore.

AI understands the shape of an ad, not the strategy behind one

The biggest limitation was not technical. It was strategic.

The AI seemed to understand that an ad for a video company should include cameras, editing software, production imagery, and a confident brand reveal. But it did not understand why those things mattered or how they should connect to a specific message.

A real advertisement needs to answer basic strategic questions:

Who is this for?
What problem does the company solve?
Why should the viewer care?
What makes this company different?
What should the viewer remember after watching?

The AI-generated ad gestures toward all of those things without really answering any of them. It communicates “video production company” in a generic sense, but it does not communicate Metamer Studios in a meaningful sense.

That is the gap. AI can mimic the conventions of advertising — the pacing, the camera shots, the bold voiceover, the music, the logo treatment — but advertising is not just a sequence of conventions. It is strategy made visible.

Without that strategic foundation, the result feels like a collage of marketing signals rather than a persuasive piece of communication.

The voiceover made the problem worse

The voiceover was one of the clearest giveaways that the spot was AI-generated.

It sounded robotic, with very little tonal fluctuation. It had the cadence of a synthetic narrator reading promotional copy rather than a human being trying to connect with an audience. In advertising, that’s important. Voiceover is not just narration; it is performance. It carries attitude, confidence, warmth, rhythm, and emotional intent.

Even more practically, the tool struggled to pronounce our company name correctly. Attempts to correct the pronunciation did not fully solve the issue.

In a brand advertisement, the company name is not a small detail. If an ad cannot reliably say the name of the company it is promoting, it immediately becomes commercially questionable. A human voice actor, producer, or editor would catch and resolve that quickly. In this workflow, it became a repeated friction point.

The issue was not only that the voice sounded artificial. The issue was that it lacked direction. A good voiceover is shaped by choices: Should this feel premium? Playful? Urgent? Human? Sophisticated? Warm? The AI voice delivered the words, but it did not interpret them.

That is a major limitation for marketing content, where tone is often the difference between sounding credible and sounding forgettable.

The 3D logos were treated like heroic characters — and it was absurd

One of the strangest creative choices was the use of 3D logo imagery.

The video treated logo assets almost like heroic characters. They appeared with dramatic weight, staged as if the audience should emotionally respond to them simply because they were large, shiny, and moving through space.

That is a very AI kind of mistake.

The tool understood that brand imagery should feel important, but it did not understand how normal human viewers actually respond to branded assets. A logo can be powerful at the right moment. It can punctuate a story, reinforce recognition, or create a polished finish. But when a logo is over-dramatized without the emotional or narrative foundation to support it, the result feels disconnected.

A human audience does not automatically care about a floating 3D logo. The audience cares about what the company does, what it makes possible, and why it matters to them.

The AI skipped too quickly to brand worship. It treated the logo as the story instead of using the logo to conclude the story.

In strong advertising, brand elements are integrated into a larger emotional or strategic arc. In this AI-generated spot, the 3D logo moments felt like spectacle without meaning.

Generic footage creates generic brands

The imagery had another major problem: it was broadly relevant but not specifically meaningful.

Much of the spot relied on stock-style footage associated with media production: cameras, editing screens, talent on set, technical interfaces, production monitors, and abstract creative environments. None of those images are wrong for a company like Metamer. But they are also not especially ownable.

They could belong to almost any production company, agency, content studio, SaaS platform, creative consultancy, or media brand.

That is one of the hidden risks of AI video tools. They can make something look professional by drawing from the visual clichés of a category. But category clichés are not brand identity.

A stronger Metamer advertisement would need to show something more specific: our actual work, our point of view, our production process, our standards, our clients, our humor, our taste, our ability to solve creative problems. It would need to communicate not just “we make video,” but why the way we make video is valuable.

The AI-generated ad defaulted to the visual average of a production company. Good advertising usually requires the opposite: a specific point of view.

The repeated imagery exposed the lack of editorial judgment

Even though many assets were available to the tool, the AI reused the same or similar imagery in multiple places.

That is a small thing that becomes a big thing in a short ad. When a spot is only 30 or 40 seconds long, every shot has to earn its place. Repeating imagery can be useful when it is intentional — for rhythm, emphasis, recall, or structure. But when repetition appears accidental, it makes the piece feel thin.

In this case, the repeated visuals suggested that the AI was not really making editorial decisions. It was filling time with assets that were close enough to the prompt.

A good editor would likely avoid that. Or, if they reused an image, they would do it for a reason.

Editing is not just assembly. Editing is taste, rhythm, hierarchy, and restraint. It is the process of deciding what the viewer should see, when they should see it, and how each moment changes their understanding of the message.

The AI could assemble. It could not truly edit.

The transitions were another tell

The video used a lot of cross-fade transitions.

Too many.

Cross-fades have their place, but they are generally used sparingly in advertising and filmmaking. When overused, they can feel sophomoric — like a default editing choice rather than a motivated creative decision.

In a polished commercial spot, transitions are part of the visual language. A hard cut, match cut, whip pan, dissolve, speed ramp, or graphic transition should serve the tone and pacing of the piece. They should help create momentum or meaning.

Here, the cross-fades felt like a crutch. Instead of using cuts to create energy, contrast, or precision, the video often softened the movement from one generic image to the next. The result was less dynamic than it wanted to be.

This is another example of AI understanding what editing software can do without understanding what editing is for.

The commercial usability problems showed up quickly

Some of the most important issues were not creative at all. They were practical.

The generated video included visible third-party watermarks from stock sources. It also included a platform watermark. These are of course solvable by purchasing licensed footage and paying the platform for its hard work, but there was no easy way to replace or license the underlying footage, even through more prompting.

There was also an unknown musical underscore. The AI included music, but there was no clear indication of where the track came from, what rights were attached to it, or whether it could be safely used in a commercial context.

That is a major issue.

In marketing and advertising, rights are not an afterthought. Footage, music, imagery, fonts, logos, likenesses, and voice assets all need to be cleared. A video can look finished while still being completely unusable from a licensing standpoint.

This is where the promise of speed becomes complicated. A tool may generate a video quickly, but if the result requires extensive legal review, asset replacement, music clearance, watermark removal, and reconstruction, the apparent efficiency starts to disappear.

A video that takes minutes to generate but hours or days to make usable is not necessarily a shortcut.

The workflow itself became a deal breaker

The workflow created another important limitation.

This video was created using a GPT inside ChatGPT called InVideo. Apparently, GPT-based workflows can use temporary chats that cannot be retrieved afterward. For this test, that created a serious problem: we could not go back and review the exact prompting history used to generate the video.

That has several consequences.

First, it makes the process harder to evaluate. If we cannot revisit the prompts, we cannot easily understand what worked, what failed, or what should be adjusted next time.

Second, it limits repeatability. A professional creative workflow depends on iteration. You need to be able to return to prior versions, compare changes, document decisions, and refine the direction.

Third, it creates a very practical asset-management problem. As mentioned in the previous section, this is a huge issue and massive time-suck. If the tool pulls in stock footage and the chat history disappears, it becomes much harder to identify which footage was used, where it came from, and how to purchase or license the clean versions. In this case, that made watermark removal and proper licensing much more difficult.

That is not necessarily an unsolvable problem for all AI video workflows. It is a workflow issue that could be addressed in future projects by using persistent project files, documented prompts, asset logs, and clearer source tracking.

But for this test, it was a major deal breaker.

Advertising production needs a paper trail. AI tools that generate assets without preserving the decision trail, source trail, or prompt trail are not yet aligned with the way commercial creative work actually has to function.

It also was not as fast as expected

One of the assumptions about generative AI is that it saves time.

Sometimes it does. But in this case, it still took a couple of hours to get the video to the point where it was even somewhat usable as a case study.

That is important because a good human editor may have been able to create a better video in the same amount of time — especially if they had access to existing Metamer assets, brand guidelines, licensed music, and a clear creative brief.

This does not mean AI has no value. But it does challenge the simplistic idea that AI-generated video is automatically faster or cheaper.

The relevant comparison is not “AI versus doing nothing.” The relevant comparison is “AI versus a skilled creative person using the same amount of time.”

And in this case, a skilled editor would likely have made better choices across the board: stronger pacing, cleaner transitions, more varied imagery, better brand alignment, licensed assets, intentional music, and a more coherent story.

Speed is only valuable if the output moves you closer to the goal.

AI video is useful — but maybe not where people think

Despite all of these limitations, the experiment was not a failure.

The AI-generated ad was useful. It just was not useful as a finished advertisement.

It was useful as a creative diagnostic. It showed what the generic version of our category looks like. It revealed clichés we would want to avoid. It surfaced workflow risks. It gave us a quick way to evaluate tone, pacing, imagery, and messaging. It created an object we could react to.

That may be one of the best near-term uses for AI video in marketing: not replacing production, but accelerating pre-production.

AI can help generate mood boards, rough animatics, placeholder edits, visual territories, sample scripts, and early concept directions. It can help teams get from blank page to conversation faster. It can make abstract ideas more tangible.

But that is different from saying it can produce final advertising.

At least in this test, the AI was most valuable before the real creative work began.

What a human creative team would have done differently

A human-led process would not start by grabbing production-adjacent visuals. It would start with strategy.

Who are we trying to reach? What do they already believe? What do they need? What kind of production partner are they looking for? What makes Metamer credible? What kind of feeling should the spot leave behind?

From there, the creative choices would be built around a message.

A better Metamer Studios ad might show actual work. It might show the scale and energy of our productions. It might highlight the kinds of clients we serve. It might focus on the experience of working with a team that can take an idea from concept to finished film. It might lean into craft, humor, precision, speed, emotion, or trust.

The visuals would not just say “video.” They would say something specific about Metamer.

The voiceover would be directed. The music would be licensed. The transitions would be intentional. The edit would build. The logo would appear when the viewer had been given a reason to care.

That is the difference between generating assets and making an ad.

The real limitation is not output. It is judgment.

The most interesting thing about this experiment is that the AI did produce video. It produced motion, imagery, music, voiceover, graphics, pacing, and a brand reveal.

But advertising is not simply the presence of those ingredients.

Advertising requires judgment. It requires taste. It requires knowing when something is too much, too generic, too slow, too slick, too empty, or too disconnected from the audience. It requires understanding not just what can be placed on screen, but what should be placed on screen.

Generative AI is getting better at output. But output is only one part of creative work.

The harder part is deciding what the work is supposed to do.

Conclusion: AI made an ad-shaped object

The Metamer Studios test ad was impressive in one sense: it showed how quickly AI can create something that resembles a professional video advertisement.

But it also exposed the current limits of that resemblance.

The voiceover lacked human nuance. The company name was mispronounced. The 3D logos were over-dramatized. The transitions felt amateur. Imagery was reused. Stock footage was watermarked. Music licensing was unclear. The workflow did not preserve the prompt history or make it easy to locate the original assets. And after a couple of hours of effort, the result was still far behind what a good editor could likely produce in the same amount of time.

That does not mean AI video tools are useless. Far from it. They can be powerful for ideation, experimentation, previsualization, and fast creative exploration.

But if the goal is advertising — real advertising, built to persuade human audiences and survive the practical demands of commercial use — then AI still needs direction, oversight, taste, and a clear production workflow.

The AI made an ad-shaped object.

The distance between that and a working advertisement is where the real creative work still lives.

Looking for a video partner to help work through the best video content strategy for your organization? We’re here to help. Reach out through our contact form or send a message to info@metamer.io. We can’t wait to hear from you.