How AI Is Unmasking the Hidden Truths Behind Marketing Failures

Imagine launching a holiday ad campaign only to discover your carefully crafted copy’s been mangled and your product images swapped… and you swear no soul on your marketing team touched a thing. Sounds like a nightmare, right? Well, that’s exactly what happened to a bookstore when Meta’s ad AI took it upon itself to “improve” the approved creative after launch, turning the spotlight—and the backlash—onto an unsuspecting photographer branded with the infamously delightful label “AI slop.” Now here’s the kicker: as AI tools weave themselves deeper into the marketing funnel, the age-old question, “Who’s responsible?” is starting to vanish like last year’s holiday sales. Guy Hanson from Validity nails it—the systems meant to speed us up are silently rewriting the rules of accountability, and honestly, many teams are just waking up to the chaos. So, what do you do when the tool you hired to boost your campaign actually becomes the wildcard nobody’s watching? Buckle up—this isn’t just about AI glitches; it’s about who’s holding the reins when the bots start running the show. LEARN MORE.

A bookstore’s holiday ad campaign went live with garbled text and a swapped product photo, and nobody on the marketing team had touched either one. Meta’s ad AI altered the approved creative after launch. The team only found out because the photographer whose work got rewritten started fielding messages calling it “AI slop.”

I’ve been chasing a version of this story since Google unveiled an agent that works across Ads, Analytics, Merchant Center, and Marketing Platform. Every time a marketing team hands another piece of the funnel to an automated system, somebody stops being the obvious answer to a very old question. Who’s responsible for this campaign? For decades that question had an easy answer. According to Guy Hanson, vice president of Customer Engagement at Validity, the answer is quietly disappearing, and most teams haven’t noticed yet.

Hanson has spent the past several months talking with Validity’s email marketing clients about exactly this problem, and his read on the bookstore incident is blunter than I expected going in. “The tooling made an unrequested, unflagged change to approved assets, but the process also had no verification step to compare the live creative against the approved source after publishing,” he told me. His fix isn’t complicated. Lock creative from automated modification once it’s approved, or run a scheduled audit within 24 hours of launch that checks live assets against the approved files. Neither idea is new. Print production teams have used version locks for decades. What’s new is a workflow where the thing most likely to alter your creative without asking is the same system you installed to help you move faster.

Where Ownership Actually Goes Missing

Hanson breaks a typical email campaign into three stages, strategy, building, and approval or handoff. AI has colonized the middle stage first, generating subject line variants, copy options, audience segments, send-time recommendations, and image assets. Strategy and final approval are supposed to stay human. In practice, Hanson said, approval increasingly falls “between the cracks.”

The size of the organization changes where that crack opens. Enterprise teams have longer approval chains, more vendors and tools, as well as ownership that straddles departments. So, a mistake can travel through three or four sign-offs without anyone treating it as fully theirs. Smaller teams have the opposite problem. One person often runs the whole campaign, which sounds like clearer ownership until you notice that person is stretched across strategy, execution, and quality control with no bandwidth left to catch what AI quietly changed after the fact.

Hanson pointed me to a recent conversation with Zapier’s Leah Miranda and Jarrang’s Stafford Sumner that explains why this is getting harder rather than easier. Miranda’s view was that AI can now carry a campaign most of the way, especially once it’s trained on the right brand voice, but that a layer of human judgment still matters at the end. Sumner’s addition was sharper. As AI improves, judgment becomes more valuable, not less, because knowing when to trust an output, when to challenge it, and when to throw it out is turning into its own marketing skill. Neither of them was talking about accountability directly. Both were describing the exact moment where accountability tends to slip, the point where a human is supposed to apply judgment to an AI output and either does it on autopilot or skips it entirely.

What Counts As An AI Mistake Now

The obvious failures are the ones that make headlines, garbled copy, an off-brand tone, a swapped photo. Hanson thinks the category is wider than that and getting wider fast. As consumers increasingly hand their inboxes to AI agents that sort and summarize on their behalf, failing to get an email surfaced to the right recipient starts to look like a mistake too, even if nothing in the email itself was wrong.

Subject lines are the sharpest edge of this right now. Washington State’s Commercial Electronic Mail Act (CEMA) law bars any subject line containing false or misleading information, and Hanson has watched class action filings under that law climb. Generative AI writes a large share of subject lines today, and it optimizes those lines to get opened and clicked. Nobody has proven that an AI-written subject line triggered any specific lawsuit, but the incentive structure alone is worth watching closely.

There’s a murkier question sitting just behind that one. If a mailbox provider’s AI summarizes an email and gets the contents wrong, and a recipient acts on the bad summary, then who answers for it?

The mailbox provider wrote the summary. The sender wrote the email the summary was supposed to represent. The recipient acted on neither, exactly. Hanson hasn’t seen a lawsuit test that question yet, but it’s the same shape as a problem SEJ readers already recognize from a different corner of search, an AI system compressing someone else’s content into something that no longer quite matches, with no clean line back to who’s responsible when it’s wrong.

What The Hiring Data Actually Says

Validity’s State of Email 2026 report, built from a 30-question survey of 502 marketing professionals across the US, UK, Australia, and New Zealand fielded between November 19 and December 17, 2025, puts numbers behind what Hanson is describing. I checked these against the report itself rather than taking the round figures on faith. Thirty-five percent of companies are prioritizing AI and machine learning application skills in their next round of email marketing hires. Twenty-seven percent are prioritizing marketing automation and workflow development. Compliance and data privacy expertise sits at 15%. Design, HTML and CSS template development, the skill Validity’s own 2023 report said companies were hiring for with “a singular focus,” has fallen to 14%.

I think that gap is the real story buried in this report, more than the AI adoption numbers everyone will quote out of it. Companies are staffing up for the parts of the AI transition that are visible and immediately profitable, and staffing down for the part that only pays off when nothing goes wrong. Hanson put the economics behind that instinct more precisely than I would have. Lifecycle automations, he told me, generate 41% of total email revenue while making up roughly 5% of a typical program’s sending volume, which makes the investment case for automation easy to defend in a budget meeting. Nobody has ever walked into a budget meeting and gotten applause for the lawsuit that didn’t happen.

That same imbalance shows up outside email, in the exact search and ads infrastructure this publication covers every day. Google’s cross-platform agent doesn’t stop at email. It reaches across Ads, Analytics, Merchant Center, and Marketing Platform, which means the accountability gap Hanson is describing in one channel is about to open up across four at once, under a single system that almost nobody has fully mapped yet. An AI Overview that misattributes a claim to your brand and an AI agent that reallocates ad spend without a sign-off are the same failure wearing different clothes. Somebody approved the tool. Nobody defined what happens when the tool goes past what it was approved to do.

From Specialist To Orchestrator

The hiring shift away from design specialists isn’t really about design. Hanson’s explanation of it is the clearest I’ve heard. Email designers used to be valuable because very few people could build layouts that reinforced brand identity and rendered correctly across every inbox and device combination. AI can now generate multiple on-brand variations of that same asset faster than a human ever could, which is why companies have pulled back on hiring design specialists over the past three years.

What replaced them isn’t nothing. It’s a generalist role Hanson calls an orchestrator, someone who gives AI the context it needs to produce usable work, checks the output for errors, adjusts it, and figures out how the final version fits the rest of the campaign. That job needs email fundamentals plus prompt engineering, quality control, and campaign execution all at once. Nobody has formalized it as a standalone title yet. Hanson expects it will eventually report directly to the marketing director, mostly because that’s the person who ends up owning AI accountability once something breaks badly enough that leadership starts asking who was watching.

Until that role gets formalized, Hanson’s read on where accountability actually lands is worth ground truthing. Teams keep the credit when AI produces a win. When it produces a failure, the same teams tend to reframe it as a tooling or vendor issue rather than their own. Marketing leadership usually stays accountable on paper. The blame, in practice, gets routed to procurement, IT, or whichever vendor sold the tool. That’s not a new instinct. It’s the oldest instinct any organization has when facing a mistake nobody wants to own, applied to a system that’s unusually good at absorbing blame because it genuinely did make the decision, even when a human was supposed to be watching it happen.

The Compliance Blind Spot Nobody’s Budgeting For

Hanson doesn’t think the design-to-orchestrator shift is the risky part. The risky part is compliance and data privacy expertise sitting at 15% of hiring priorities at the exact moment AI agents are making more autonomous decisions with customer data than they ever have before.

He pointed to an active example. The UK’s Information Commissioner’s Office has opened an investigation into X over whether Grok unlawfully used personal data to generate nonconsensual deepfake imagery. Most marketers will never come close to a case at that scale, but Hanson’s point wasn’t about scale. It was about the underlying question every marketing team using AI on customer data now has to answer for itself, whether the systems it deployed to move faster are actually operating inside the legal basis the company obtained consent for in the first place.

His prescription is three steps any team can start this quarter. Audit existing processes and run a risk assessment on what your AI use might actually expose. Check whether the legal basis and consent you already have from customers covers the new things AI is doing with their data, not just the things it was doing when you first got that consent. Then update the privacy policy to match, since most companies’ policies still describe a pre-AI process. He’s adamant this isn’t a competing priority against AI investment. The more a team automates, he says, the stronger its safeguards need to become, and one of the more efficient moves is using AI itself to flag potential compliance issues before they reach a customer, rather than after.

What To Actually Do About It

Hanson’s advice for closing this gap in the next quarter, not the next fiscal year, comes down to three concrete moves any SEO or marketing lead can start on directly.

First, name a human owner for every AI agent capable of directly affecting a customer. That doesn’t mean matching one employee to every agent. It means someone can answer, without hesitation, who defined what the agent is allowed to do, what data it can use, and what has to stop and wait for a human before it happens. If a real consequence for a customer or the brand is on the table and nobody can answer that question on the spot, the team doesn’t have an owner. It has hope.

Second, extend the pre-send QA process past the moment of approval. Most teams already check a campaign before it goes live. Almost none have defined what an AI system is allowed to change after that sign-off, or built a scheduled audit to catch it if it does anyway. That gap is exactly what let the bookstore’s creative slip through.

Third, stop treating AI investment and compliance investment as competing line items in the same budget conversation. The more a team automates, the more its safeguards need to scale alongside it rather than lag a cycle behind. Auditing AI use cases against the actual legal basis for processing customer data, then checking whether the privacy policy has caught up to what the tools are actually doing, isn’t a defensive move. It’s the cheapest insurance a marketing team can buy against becoming the next case study.

I keep coming back to the fact that nobody planned for the bookstore’s photographer to become the public face of a mistake she didn’t make. That’s what happens when accountability isn’t assigned in advance. It gets assigned after the fact, by whoever’s name is easiest to find, and that’s rarely the person who actually owned the decision. My take is that the teams that pull ahead in 2026 won’t be the ones running the most sophisticated AI stack. They’ll be the ones who can name, in one sentence, who’s responsible when it’s wrong.

More Resources:

Featured Image: Roman Samborskyi/Shutterstock