Using AI in Advertising: How to Dodge the Pitfalls

A while back, I came across a news story that stopped me in my tracks for a few seconds.

In 2024, Google's chatbot Gemini, while generating images, produced things that never existed in history: ethnically diverse Nazi soldiers, non-white American Founding Fathers.

Ethnically diverse Nazis. Sit with that image for a second.

Worse, this image technology was embedded in Google's advertising tools. When things went wrong, Google urgently paused Gemini's ability to generate images of people.

My first reaction to the news was: of all the money the ad industry has saved thanks to AI, how much of it came without paying a price?

Today, let's talk seriously about this: when using AI in advertising, how do you claim the upside and dodge the pitfalls?

First, Something Counterintuitive

Many people assume AI entered the ad industry only in the last couple of years.

Actually, no.

Programmatic advertising, automated bidding, algorithmic optimization — the technology behind these has been in use in advertising for over a decade. AI has long been the invisible employee in advertising's back office.

The truly new variable is generative AI.

What is generative AI? It's the kind of AI that can write copy on its own, draw images on its own, edit videos on its own. Unlike the AI of the past that hid in the back office crunching probabilities, it walks straight onto the stage and starts doing creative work.

And its momentum only keeps building. In 2025, the US private sector announced up to $500 billion in AI infrastructure investment; the policy winds, too, have shifted toward embracing AI.

Money is flowing in, tools are spreading, competitors are sprinting ahead.

So here's the question: what exactly are the decision-makers at ad agencies and brands facing right now?

In one line: on one side, the temptation of efficiency; on the other, the bill for risk.

Use AI, and you can cut costs, gain efficiency, keep pace. Use it badly, and you'll wreck your brand, lose your budget, and end up in court.

How do you strike that balance? My view: see the pitfalls clearly first. Once you've seen them, then talk about how to use it.

Which pitfalls? Three.

Pitfall #1: AI Will Confidently Make Things Up

Generative AI has a famous flaw called hallucination.

What is hallucination? It doesn't know what it doesn't know — so it invents nonexistent facts with total confidence.

How bad does this get in advertising? A survey of industry practitioners asked how people view generative AI's brand-safety and misinformation risks. The result: 100% of respondents saw a risk, and 88.7% rated it at least moderate.

One hundred percent. Not a single person thought it was fine.

The Gemini incident at the top of this piece was hallucination and bias rolled into one. Its image model had been set up to avoid racial bias, but it overcorrected so hard it flipped into a new failure, generating images that never existed in history. Ironic, isn't it? Trying to dodge bias, it manufactured fresh errors.

So should we just stop using AI for creative work?

Not quite. Molly Marshall, Partner, Client Strategy and Insights at Basis, has a view I fully agree with: AI, for now, cannot replicate the core of creativity — starting from a genuine insight and making content that truly moves the target consumer. So AI-generated assets should be used to supplement and iterate on existing creative strategies. Building strategy from scratch is, for now, not AI's job.

In plain terms: AI can do the work for you, but it can't do the insight for you.

Pitfall #2: Chatbots Are Wrong Half the Time

The second place generative AI goes wrong is customer service.

Lots of brands have bolted AI chatbots onto their websites, hoping to save manpower and personalize service at once. Good idea. And the reality?

Someone ran a real-world test on the AI support bots of TurboTax and H&R Block. The finding: at least half the time, these two bots were giving users wrong information.

Half the time. Imagine it: if one out of every two of your service reps was making things up, how would you rate that brand?

Marshall is clear-eyed about this too: chatbots are genuinely an opportunity, especially now that physical stores are dwindling and customer service is moving online. But the damage a bot can do by talking nonsense may, for certain brands, far outweigh the costs it saves.

What you save is labor; what you may lose is trust. Do this math in advance.

So every brand has to answer one question: which parts of the business are you willing to hand to AI to face customers directly, and which must keep a human review in the loop?

There's no standard answer. But every good answer contains one consensus: every point where AI faces customers directly needs a human standing guard.

Pitfall #3: Your Ad Budget May Be Flowing into AI-Made Junkyards

This pitfall is more hidden, and more expensive.

Open Google Search today, and the share of AI-generated content in the top 20 results has climbed from 5.6% when ChatGPT first launched to 19% by early 2025.

What does 19% mean? Roughly one in every five search results is written by AI.

As AI-generated content multiplies, bad actors have an easier time too. They mass-produce low-quality pages riding on trending search keywords and stuff content farms full of misinformation — what the industry calls MFA ("Made for Advertising") sites. Their only reason to exist is to trick your programmatic advertising into spending money there.

How expensive does this get? Industry research has tallied it: 15% of programmatic advertising budgets flow into MFA sites.

Fifteen percent. The campaigns you painstakingly optimized — fifteen out of every hundred dollars are feeding the garbage dump.

Making matters worse, regulation is loosening and platforms are retreating.

In 2025, the new US administration signed an executive order revoking existing policies seen as obstacles to AI innovation. Meta also scrapped its fact-checking program, switched to user-reported Community Notes, and relaxed rules that once banned controversial content.

Regulators loosened. Platforms stepped back. Low-quality AI content will only multiply.

So what should advertisers do? Three things: use MFA blocklists to keep budgets outside the junkyard; build long-term partnerships with quality, credible publishers; monitor campaigns in real time and block misleading sites and keywords the moment they appear.

Marshall puts it bluntly: advertisers must be able to block misleading sites and keywords in real time.

In plain terms: the best way to save money isn't spending less on ads — it's not spending your ads on garbage.

Worse Still, Consumers Aren't Buying It

The pitfalls above are all internal to the industry. But what truly decides AI advertising's fate is how consumers see it.

And consumers' attitude, honestly, is a bit cold.

In 2024, the Edelman Trust Institute published a report: over the past five years, US consumers' trust in AI fell from 50% to 35%, a 15-percentage-point drop.

Meanwhile, nearly two-thirds of US adults said AI-generated ads make them uncomfortable.

Here comes the interesting contrast: 77% of industry practitioners believe generative AI will have a positive impact on marketing and advertising.

Practitioners are mostly optimistic; consumers are mostly skeptical.

That temperature gap is the tuition brands will end up paying.

So what to do — give up on AI?

No need. As noted, AI has been working in advertising's back office for over a decade — bidding optimization, budget allocation — and consumers long ago accepted it in the places where they never notice it. What they resent is AI content shoved right in their faces with no acknowledgment.

So the way out comes down to one word: transparency.

Tell consumers plainly where AI was used in your marketing. Build a dedicated disclosure page on your website, or label it directly on the content.

You might think: admitting it's AI-made — isn't that shooting yourself in the foot?

The data says the opposite. Research has found that when advertisers proactively disclose AI-generated content, the ad's appeal rises 47%, its credibility rises 73%, and trust in the brand itself rises 96%.

Ninety-six percent. The more candid you are, the more consumers trust you.

Data privacy works on the same logic. 71% of US consumers worry about security risks from their digital activity; 81% believe companies will use AI to collect and analyze their personal information — in ways that make them uncomfortable.

How to break through? Explain your data-protection measures clearly, partner with privacy-focused collaborators, and earn certifications like SOC 2 (an independent audit standard for how organizations manage and secure data).

Transparency isn't a cost — it's an asset. In an era when everyone is guessing "was this made by AI?", the ones who volunteer the truth are the ones who win trust.

The Third Hurdle: The Law

Enough about the market; let's talk law. This part is drier, but stepping into it hurts more.

First, data privacy.

Everyone in advertising knows third-party cookies are exiting and signals are drying up. AI happens to be a sharp tool for replenishing signals: use first-party data to build lookalike audiences and predictive audiences, and win back the lost precision. Many advertisers are already using this toolkit to hedge against the death of cookies.

But note: AI tools themselves eat data too. They train models on personal data, and where different tools get their data, where it's stored, and who can access it is often unclear. More subtle still, some tools can infer your location, health conditions, even political and religious leanings from your data.

If any of this goes wrong, it's a compliance incident.

So before procuring any data-touching AI tool, interrogate it end to end: how is data collected, processed, stored — and is it compliant? At the same time, your own data systems must be compliant too. Don't count on regulation staying loose forever; even if certain policies are relaxed right now, the data-privacy line will only be drawn finer and finer over the long run.

Next, deceptive advertising.

The FTC (Federal Trade Commission) has been watching something for years called dark patterns.

What are dark patterns? The little mechanisms that use design to induce you to order, to induce you to surrender personal data. Like making the opt-out button minuscule, or setting the default option to consent. The arrival of AI makes these mechanisms easier to mass-produce and polish.

The US federal level is watching, and states aren't idle either: the Colorado Privacy Act and California's CPRA (California Privacy Rights Act) both write dark-pattern oversight into law.

Some will say the new FTC chair's style may be looser than his predecessor's — won't regulatory pressure ease up?

Don't bet on it. The FTC's core mission of protecting consumers from misleading advertising and harmful practices is written into its DNA; it doesn't vanish because one person is replaced.

Finally, copyright. This one is still in flux.

In January 2025, the U.S. Copyright Office released a report on AI and copyright whose core conclusion, in one sentence: AI-generated content can be protected by copyright only when a human author has determined enough expressive elements.

What are "enough expressive elements"? In plain terms: pressing a button and letting AI spit out an image doesn't make it your work. You have to genuinely take part — plan, edit, polish — and leave your own creative judgment in it.

What's more, the boundary of "enough" is still blurry, and the future will most likely be decided case by case.

So what should ad teams do? Make the creative process solid. Keep written records at every step: who used AI, where it was used, and what changes a human made. For assets you intend to register for copyright, the records need to be even finer. And keep a reliable legal counsel on call.

This boundary is still moving. Your job is to make sure your important assets aren't standing outside it when it settles.

Closing Thoughts

Back to that opening image of "ethnically diverse Nazis."

That absurd picture reminds us: AI is an amplifier. It amplifies your efficiency — and it amplifies your mistakes.

Bringing AI into advertising: the direction isn't wrong. What's wrong is treating AI as magic that wins the moment you turn it on. The truly smart teams are the ones that think about risk first: audit AI's output, guard the exits that face customers, block junk traffic, be candid with consumers, and do the legal homework in full.

The pioneers aren't the ones who never fell — they're the ones who fell and know why they fell.

The sooner you learn to hit the gas with a foot ready on the brake, the farther you'll go.

Here's to eating the AI dividend — and keeping your name intact.