Two studies, one promotion filter
If you manage SEO, paid media, or the broader digital mix, the next career jump may not be decided by your channel wins alone. New research from MIT Technology Review and Harvard Business School points to a sharper filter: the managers who reach director and executive roles will be the ones who can show that AI actually pays while keeping their teams intact.
The MIT side looks at the economics of AI. The Harvard side looks at how job postings and required skills are already shifting. Together they describe a manageable but demanding promotion case.
MIT: the productivity bill is coming due
In the MIT Technology Review, David Rotman analyzes the bet behind AI tools. Hyperscalers are expected to spend around $750 billion on data centers this year. Reported AI revenue is far lower: Gary Gensler, former SEC chair and now at MIT Sloan, puts it at $150 billion to $200 billion. Jessica Wachter of Wharton asks how fast earnings would need to grow to justify roughly $1.1 trillion in spending through 2027. Her answer: a factor of 2.7 by 2030 just to break even. If the productivity does not arrive, she and her coauthor call the buildout the “largest misallocation of capital in history.”
The signal for search marketers is direct. Alphabet reported nearly $120 billion in revenue last quarter but still posted a free cash deficit of about $5.9 billion—its first since going public in 2004. That kind of pressure means AI answers, citations, and SERP features will keep shifting as platforms chase returns.
So far the productivity lift is not visible in the data. A survey of roughly 6,000 executives across four countries found that around 90% reported no productivity gain from AI after three years. Workday’s research, as cited in the piece, shows that for every 10 hours AI saves, companies give back about four hours fixing weak output. BetterUp Labs and Stanford Social Media Lab found that 41% of workers received AI “workslop” in the previous month, and each instance took nearly two hours to sort out.
Harvard: repetitive tasks are exposed, judgment work is enhanced
Harvard Business School professor Suraj Srinivasan and coauthors reviewed U.S. job postings from 2019 through March 2025, using ChatGPT to sort more than 19,000 tasks across more than 900 occupations. After ChatGPT’s public launch, postings for occupations heavy on structured, repetitive tasks fell 13%. Postings for analytical, technical, or creative work, the kind AI can enhance, grew 20%.
The skill requirements moved too. Automation-prone roles listed 7% fewer skills. Augmentation-heavy roles asked more often for AI skills such as prompt writing and using AI tools. Finance and technology saw the largest declines. Srinivasan’s advice to companies is to treat generative AI as an “augmentation tool rather than merely a cost-cutting measure” and to fund reskilling.
For our world, the repetitive parts—recurring reports, bulk title tag and ad copy production, bid adjustments, search term cleanup—sit on the exposed side. The judgment work—strategy, test design, measurement, and cross-team persuasion—is where a manager earns the director title.
Winning the budget conversation
Put the studies side by side and you get the tension in the room. Executives want productivity gains, often by growing sales while reducing staff. The Harvard research says firms get more from AI when they build human-AI collaboration and retrain people. Your job is to hold both.
Here is the practical playbook:
- SEO and AI search: keep an AI hours ledger per workflow. Log time spent prompting, checking, and fixing output next to rankings, citations, and assisted conversions. Compare your rework rate against Workday’s four giveback hours per ten saved, and take the gap to your director as a finding.
- Paid media: ask platforms and agencies which AI features are bundled or discounted today, and get pricing terms in writing. Judge automated campaigns on incrementality with a holdout test, not just cost per result.
- Broader digital marketing: build workflows so a cheaper model can be swapped in. Keep prompts and test tasks in files you own, run one smaller model against your current tool on a real task, and build a one-page AI scorecard with hours saved, hours given back, the outcome metric that moved, cost per workflow, and the fallback model.
- Team plan: add a reskilling line to your budget request covering AI literacy and working alongside AI, because Harvard’s data shows augmentation-prone roles are increasingly asking for those skills.
The promotion case is no longer “we adopted AI.” It is “we proved AI pays, and we have a plan for the team.” That case works whether the AI bubble deflates or the productivity finally shows up.
Source: Search Engine Journal
