- 17% of CMOs have cut marketing jobs because of AI. 36% expect to within two years (47% at companies above $20 billion revenue). [1][2]
- Of about 90 CMOs surveyed, more than three-quarters are piloting or scaling AI. Zero say their function has been transformed by it. [1]
- The real mechanism is hiring, not firing: an estimated 18% fewer net-new marketing hires in 2025-26, at a median loaded cost of $294,000 per marketer. [7]
- Every reported AI return is an input. Time efficiency 49%, cost efficiency 40%, added capacity 27%. Revenue: absent. [3]
- Outside the model's competence, GPT-4 made consultants worse: accuracy fell from 84.5% to 70.6%. [4]
Most marketing org charts drawn in the last 18 months carry the same quiet lie. The boxes look identical to last year's. The work inside them doesn't.
Spencer Stuart put the question to about 90 leading CMOs: had they cut marketing jobs in the past 12 months as a direct result of AI? Seventeen percent said yes. Asked whether they expected to cut jobs in the next 12-24 months, 36% said yes. At companies above $20 billion in revenue, 47%. [1][2]
That gap is the story. The question was never humans versus AI. It is what happens to a team that keeps every seat filled while the work underneath those seats changes hands.
Seventeen percent have cut. Thirty-six percent intend to.
| What CMOs have done | What CMOs expect |
|---|---|
| 17% cut marketing jobs because of AI [1][2] | 36% expect to within 12-24 months [1][2] |
| more than three-quarters are piloting or scaling AI [1] | 47% expect to, at firms above $20 billion revenue [1][2] |
| Zero say the function is transformed [1] | 71% of CMOs plan >$10 million a year on generative AI [6] |
Between intention and action sits a third mechanism that appears in neither column. A synthesis of LinkedIn workforce data suggests AI leverage reduced net-new marketing hires by roughly 18% across 2025 and 2026. [7] Treat that figure carefully. It is second-hand, and the underlying methodology is not public. But it points at something the layoff statistics cannot see. A widely cited but hard-to-source estimate puts the median fully loaded cost near $294,000 per marketer, so the rational move is not to fire the person you have. It is to never post the role. [7]
A layoff is an event. A hiring freeze is a climate. The mechanism is slower and more consequential than a redundancy round. Being let go is visible, finite, and legible to everyone in the building. A hiring freeze dressed up as a productivity gain has no announcement, no severance line, and no end date.
Agencies feel it first. Richard Sanderson, who leads the Spencer Stuart research, is direct about where the exposure sits.
It's the content production and likely agency-facing roles that are most vulnerable. Richard Sanderson, Spencer Stuart [2]
The gain is real. The proof is not.
Ask a CMO what AI returned and you get a list of inputs. Gartner's 2025 CMO Spend Survey found returns reported as improved time efficiency by 49% of respondents, improved cost efficiency by 40%, and increased content capacity by 27%. [3] Not one of those is revenue. Not one of them is margin.
Every bar is a way of doing the old work faster or cheaper. None of them is a bar labeled revenue, and that missing bar is the whole argument. Gartner's respondents reported the returns AI is good at proving, and stayed quiet on the one that pays for a restructuring. [3]
Meanwhile marketing budgets sit at 7.7% of company revenue. Flat, going nowhere. [3] McKinsey puts the annual productivity value of generative AI in marketing at $463 billion, and finds that 5% of marketers are actively building the capability and 4% are scaling it. [8] Boston Consulting Group (BCG) counts 10% of companies as generative-AI top performers. [5] Seventy-one percent of CMOs intend to spend more than $10 million a year on it. [6]
Hours saved are not money earned, and cost avoided is not growth. This is not a technology failure. It is a measurement failure, and it is precisely why the second wave of restructuring keeps getting postponed. You cannot fire people to bank a gain you have not proven exists.
Where the machine quietly makes you worse
The most useful number in this entire subject is not about jobs. Harvard Business School's AI Institute gave consultants a set of tasks and access to GPT-4. On work inside the model's competence, they completed 12.2% more subtasks. [4] Then researchers handed them a task deliberately constructed to sit outside it.
Accuracy in the control group, working without AI: 84.5%. Accuracy among the consultants using GPT-4: 70.6%. Among those given a briefing on prompt engineering first, it fell further, to 60%. [4]
The tool does not tell you which task you are on. It answers with identical fluency either way, and the better your prompting, the more confidently it takes you somewhere wrong. BCG's own work with the same researchers found quality gains of 40% and speed gains of 25%. [4] Real, large, and entirely conditional on the work sitting in territory the model understands.
So the boxes stay. The seats stay filled. Somewhere in the middle of the org chart a job quietly becomes a different job, and nobody signs anything.
The firms that win the next two years will not be the ones running the most AI. They will be the ones who can prove what it did.
Sources
- Spencer Stuart · The AI reckoning: why marketers think 2026 is a make-or-break year
- Richard Sanderson, Spencer Stuart · discussion of the 2026 CMO survey findings
- Gartner · 2025 CMO Spend Survey: marketing budgets have flatlined at 7.7% of company revenue
- Harvard Business School AI Institute · Back to the beginnings of AI at work
- BCG · Bold AI ambition for B2B marketing, sales and service
- BCG · CMO confidence in GenAI is higher than ever
- Digital Applied · Marketing team structure 2026: headcount benchmarkssynthesizing LinkedIn Workforce Report and CMO Council data. Figures are second-hand
- McKinsey, via Marketing Dive · Generative AI's $463 billion marketing productivity prize
Frequently asked questions
Is AI actually replacing marketing jobs in 2026?
Not yet, and not at scale. Spencer Stuart's 2026 survey of about 90 leading CMOs found that 17% had cut marketing jobs directly because of AI in the previous 12 months, while 36% expected to do so in the next 12-24 months. Among companies above $20 billion in revenue, that expectation rises to 47%. The dominant mechanism today is slower hiring, not redundancy.
Which marketing tasks is AI taking over first?
Execution-heavy work: content production, campaign assembly, reporting, and routine analysis. Richard Sanderson of Spencer Stuart identifies content production and agency-facing roles as the most vulnerable. Strategy, brand direction, and creative judgment have not moved, because they are not the parts AI is good at.
Why do agencies feel this before in-house teams?
Because agency work is disproportionately execution work, and execution is exactly what the models got good at. When an in-house team absorbs more throughput without hiring, the external production budget is the first line a chief financial officer cuts. It is a spending decision, not a headcount decision, which is why it does not show up in layoff statistics.
Does AI reliably make marketers better at their jobs?
No. Harvard Business School's study of consultants using GPT-4 found a 12.2% increase in subtasks completed on work inside the model's competence. On a task deliberately placed outside it, accuracy fell from 84.5% in the control group to 70.6% among GPT-4 users. The tool does not signal which task you are on.
What should a marketing leader actually do about this?
Build the measurement layer before the automation layer. Gartner's 2025 CMO Spend Survey found CMOs reporting AI returns as time efficiency (49%), cost efficiency (40%), and added capacity (27%): all inputs, none of them revenue. Until you can show incremental lift rather than hours saved, you cannot justify further automation or the restructuring that follows it.
Does 'humans with AI' mean nothing changes?
The opposite. It means the org chart stays intact while the work inside it changes hands. Execution roles quietly become supervision and quality-assurance roles, and the skills that get rewarded shift from producing output to specifying it, validating it, and being accountable for it.
If layoffs are rare, where is AI actually affecting marketing headcount?
In hiring, not firing. A synthesis of LinkedIn workforce data suggests AI leverage cut net-new marketing hires by roughly 18% across 2025 and 2026, and with the median fully loaded cost of a marketer estimated near $294,000, the rational move is to never post the role rather than cut the person you have. That figure is second-hand, so treat it carefully, but a hiring freeze leaves no announcement, no severance, and no end date, which is why it stays invisible in layoff statistics.
Are CMOs actually spending real money on AI yet?
Yes, heavily. More than three-quarters of the roughly 90 CMOs surveyed are piloting or scaling AI, and 71% intend to spend more than $10 million a year on generative AI. Yet zero say their function has been transformed by it, which is the gap between spend and proof at the center of the story.
How big is generative AI's productivity value in marketing?
Large, and almost entirely unclaimed. McKinsey puts the annual productivity value of generative AI in marketing at $463 billion, yet finds only 5% of marketers actively building the capability and 4% scaling it. The prize is real, but the number of teams turning it into a proven result is still tiny, which is the same spend-ahead-of-proof pattern the rest of the data shows.
How many companies are actually getting results from generative AI?
A small minority. BCG counts about 10% of companies as generative-AI top performers, and McKinsey finds 5% of marketers building the capability with 4% scaling it. Adoption is broad and spend is high, but the group converting it into a measurable outcome remains narrow, which is why the second wave of restructuring keeps getting postponed.
Does prompt-engineering training make AI safer to use?
Not on the wrong task. In Harvard Business School's study, consultants given a prompt-engineering briefing before a task built to sit outside the model's competence did worse, with accuracy falling to 60%, below both the control group's 84.5% and the untrained GPT-4 group's 70.6%. Better prompting made a confident wrong answer more likely, not less, which is why accountability for the output matters more than fluency with the tool.
Mikołaj Salecki
Writes about media, tech, and AI business for people who actually run digital. Former agency lead. Skeptic of frameworks that read better than they perform.
Get the next issue by email.
One letter, once a week. Sharp coverage of media, tech, and AI business. No filler.
