Marketing labor costs rose despite the AI rollout, and two studies blame a skills gap, not a headcount gap

Marketing budgets are putting more money into labor this year, not less, even as AI writing and research tools spread through nearly every software marketing team. Gartner's 2026 CMO Spend Survey, fielded among 401 marketing leaders, found that labor's share of the average marketing budget climbed from 21.9% in 2025 to 24.5% in 2026, a move in the opposite direction from what AI vendors have been promising buyers for two years.

The same survey points to why. Thirty-eight percent of CMOs ranked a shortage of internal AI talent among their top three barriers to AI-driven efficiency, and 19% called it their single biggest obstacle. Buying the tools turned out to be the easy part. Building a team that can use them well is the part still unsolved, and it is now showing up as a line item rather than a footnote.

AI created a new job: checking AI's work

A separate study commissioned by Optimizely and run by the research firm Savanta across 2,003 marketing leaders in seven countries puts a number on where that unsolved problem lives day to day. Seventy-six percent of marketers now spend three hours or more each week cleaning up, verifying and rewriting what the AI handed them, and 48% say catching hallucinations and wrong facts is the single biggest chunk of that extra workload, ahead of the time lost shuttling data between disconnected systems.

"AI was supposed to give marketers room to think. What most teams got instead was more to manage." Tara Corey, SVP of Marketing, Optimizely

The same research found a gap in how that burden is perceived inside the building. Fifty-four percent of respondents said leadership underestimates the human effort it takes to make AI output usable, and 46% worry that heavy reliance on AI is eroding the skills of junior marketers, the people who would otherwise grow into doing that review work themselves.

The study also found what happens when that review time runs out. A quarter of marketers admitted that, under deadline pressure, they routinely ship machine-written material even when they can tell it misses the brand's voice, and 30% said they routinely pass that same machine-written work off as something a person wrote. Neither number describes a tooling failure. Both describe teams asked to move faster than their review capacity allows, which is a staffing and training question before it is a software question.

A skills gap and a headcount gap aren't the same purchase

Writing on MarTech, Digital Mindshare principal Gene De Libero argues that both surveys point to a diagnosis most marketing leaders skip before approving another AI seat or posting a new requisition: is the team short on skill, short on hours, or both. A skills gap means the team cannot do the work well even with the tools it has. A headcount gap means it can do the work but does not have enough hours in the week to get to all of it. Buying AI to fix a skills gap, his argument goes, just produces more of the wrong work, faster, and hiring into one does the same thing at a higher payroll cost.

He lays out three checks a marketing leader can run before the next AI request goes to finance:

• Look at what gets corrected in review: judgment errors point to a skills gap, rushed detail points to an hours gap
• Look at where unfinished work backs up: stuck with one person signals a concentrated skills gap, stuck with a busy but capable team signals an hours gap
• Pull one recurring project off the team's plate for a month and watch what happens to the rest of its output

That distinction lands harder on software marketing teams than on most, because the AI line item here is rarely a one-time purchase. It behaves like a headcount decision wearing a tooling budget, and the Gartner and Optimizely numbers both describe an industry currently answering that decision by buying more tools and assuming the skill to use them will follow on its own.

What to change before the next AI request

Software marketing teams are usually the first to pilot a new AI tool internally, the first to publish AI-assisted copy at scale, and the ones reporting ROI back to a CMO now watching labor costs rise instead of fall. That makes the fix less about whether to adopt AI and more about what gets funded alongside it. De Libero's own earlier benchmark, cited in the same piece, sets training spend at 15% to 20% of a tool's annual license cost, so a $100,000 AI license should carry a $15,000 to $20,000 training line next to it, not zero.