If you believe search marketing enjoys some kind of protected status because the public cares who—or what—does the work, Harvard just published a number that demolishes that assumption. Assistant Professor James Riley asked the American public to score how morally objectionable it would be to hand each of 940 different occupations to a machine, using a scale from 1 to 7. Search marketing strategists scored 2.31. Of the 10 occupations Harvard charted, only file clerks scored lower. Clergy scored 5.91. Childcare workers scored 5.86. Whatever is currently standing between white-collar jobs and full automation, it isn’t the public’s conscience. The real shield is a competence gap, and that gap is closing faster than most in the industry want to acknowledge.
I should reveal upfront that I love the old joke about the grocery store located between Harvard and MIT, where a student wheels a cart with 15 items into the 10-items-or-less lane. The cashier looks at the sign, looks at the student, and sighs: “You must either go to Harvard and can’t count, or go to MIT and can’t read.” It’s a joke about elite blind spots. It’s also a pretty good description of what happens when SEOs read AI research—and I nearly did it to myself with this very report. The numbers from Cambridge are unambiguous, but the interpretation requires a careful reading of what was actually measured.
What Harvard Actually Studied: The 940-Occupation Moral Scorecard
Some context first. “AI in 2026: From Adoption to Agentic” is a February 2026 roundup from HBS Working Knowledge that bundles five previously published pieces. The two that ran experiments are where the real news lives: “People Are Mostly OK With AI Taking Over Many Jobs—Up to a Point,” which cites research from James Riley’s October 2025 study, and “Who Should Approve Bank Loans: People or Algorithms?” which references Assistant Professor Elisabeth Paulson’s research co-authored with Kirk Bansak from 2024.
Riley’s occupation-scoring survey covered 940 jobs and 2,357 respondents. The 2.31 score for search marketing is only half of what he found. Based on AI’s current capabilities, the public supports fully automating roughly 30% of the occupations he tested. When Riley’s survey instead asked people to imagine a more advanced AI that outperforms humans at a lower cost, support for automation nearly doubled to 58%. A moral floor exists, but it’s narrow. Only about 12% of occupations—among them clergy, childcare workers, and athletes—drew strong moral resistance regardless of how well AI could do the job. Another 42% left people ambivalent. Riley’s own conclusion is that resistance to automation is mostly a story about whether the technology can do the job yet, not about principle. Search marketing already sits near the bottom of that moral floor. The thing standing between your job and a much higher automation number isn’t sentiment. It’s whether the tools are good enough—and that’s a much shakier position to be defending.
What is the public’s moral objection to AI taking search marketing jobs? According to Harvard data, the public finds it only slightly objectionable—a 2.31 on a 7-point scale. That places search marketing among the least morally protected occupations, directly behind file clerks. When respondents considered a hypothetical AI that outperforms humans, support for automating search marketing nearly doubled. The implication is clear: moral resistance is negligible and largely conditional on AI’s perceived competence.
The Belief Gap Behind Both Papers
Paulson’s research is a different experiment and doesn’t have an “advanced algorithm” condition the way Riley’s does, so it’s worth being precise about what it actually found. She and coauthor Kirk Bansak, an assistant professor at UC Berkeley, ran a conjoint experiment with 9,000 participants, asking them to choose a human or an algorithm to approve a loan or decide on a defendant’s pretrial release. On average, and even controlling for performance, people leaned human—by 4.3 percentage points on the loan and 7.6 points on pretrial release. Fairness, meaning equal treatment across racial groups, turned out to be the least important factor in how anyone judged either kind of decision-maker.
The more interesting number is buried in a chart on page 13 of the report, and it’s a belief split rather than a flat preference. Among respondents who already believed algorithms outperformed humans at these tasks, 56% chose the algorithm for pretrial release and 54% chose it for the loan. Among respondents who believed humans were better, 63% and 59% went with the human. Paulson said that if you can prove real accuracy gains without other metrics slipping, “that’s probably sufficient.” What her data shows is that the human preference isn’t a fixed moral stance at all. It’s downstream of a belief about who’s currently better at the job—which lines up almost exactly with Riley’s technical-feasibility argument, even though the two studies were built to test different things.
The Competence Gap Is Closing Fast
Raffaella Sadun, Karim Lakhani, and their coauthors tracked 791 product developers at Procter & Gamble, some working alone, some in teams, some with an internal GPT-4 tool and some without. Ideas ranking in the top 10% of quality were three times more likely to come from AI-assisted teams than from unassisted individuals working without it. Employees using AI also reported higher enthusiasm and energy for the work, and less anxiety and frustration, than employees who worked alone without it. That’s the exact kind of idea generation and content work search marketers get paid for, and the competence gap Riley’s data says is the only thing currently protecting the job is closing on this front in real time.
Tsedal Neeley and Expedia Group’s Ritcha Ranjan’s technical note describes where that competence is headed next. Their vision has agentic AI acting as a chief of staff, a competitive intelligence analyst, and an executive coach, running with minimal human oversight once it’s set up. Neeley’s advice to leaders adopting it is to start with what she calls the “no-joy” work—the repetitive tasks nobody wants—before handing over anything higher stakes. That’s a sensible on-ramp. It’s also a description of exactly how automation tends to creep upward once the technology proves itself on the boring stuff first.
Why This Matters For SEO: The Moral Cover Is a Mirage
My take, and I’ve only driven by the Harvard Business School on my way to the airport, is that the industry has been assuming Google keeps rewarding named human bylines and E-E-A-T signals because the public has some residual moral stake in SEO staying human work. Harvard’s own data says that stake doesn’t exist. What’s protecting search marketing right now is a competence gap, not a conscience—and competence gaps close. Google’s systems, and increasingly the citation behavior of AI answer engines, are running the same test Paulson’s respondents ran on loan officers and judges. They’re asking whether the human-produced version is still demonstrably better, and the moment that answer flips, so does the preference. The P&G study and the Neeley technical note both suggest that moment is closer than most of us in this industry want to admit.
The belief-split data from Paulson reinforces this point. When people already believe algorithms outperform humans, they choose the algorithm by a majority. The search marketing industry’s primary defense has been to assert human superiority without proving it in a way that registers with decision-makers or crawlers. That strategy works only as long as no one runs the comparison—and competitors, AI vendors, and Google’s own evaluation systems are all running it right now.
The Narrow Line of Moral Resistance
Riley’s research identified only about 12% of occupations where the public maintains strong moral resistance regardless of AI performance. Search marketing is not in that group. The occupations that do enjoy that protection—clergy, childcare workers, athletes—share a common thread: they involve deep interpersonal trust, spiritual guidance, or physical performance where human presence is intrinsic to the value. SEO work, by contrast, is judged as a functional output. If the output quality matches or exceeds human work, the public sees no reason to insist on a human being behind it.
What This Means For Your Strategy
First, put a real, checkable human name behind anything AI touches before it goes external. Not a generic “Editorial Team” byline. A person with a LinkedIn profile, credentials, and a track record a reader—or a crawler—can verify against other work. Paulson’s belief-split data says the preference tracks perceived competence, so give yours a competence signal to attach to, not just a name.
Second, publish your performance record, not just your process. If your content or your SEO program has produced measurable outcomes, put the receipts in the piece itself. That’s the accuracy demonstration Paulson’s data says actually moves people from the human column to the algorithm column—and there’s no reason your own track record can’t do the same work in reverse.
Third, reserve full automation for the boring, repeatable, no-joy tasks Neeley describes—things like internal link audits, meta description drafts, and log file triage—and keep a named human on anything that touches a reader’s trust or a client’s money. Riley’s data says that’s the one line the public still won’t fully cross regardless of performance, but it’s a narrower line than most SEOs assume, and it’s the only one left to hold.
How to Operationalize the Competence Signal
The practical implication of both studies is that proof of performance is the only durable differentiator. For an SEO agency or in-house team, that means building a system to track and publish comparative metrics: content engagement rates before and after AI assistance, conversion uplift from human-edited versus fully automated outputs, and accuracy comparisons on tasks like keyword research or competitive analysis. This is not about creating marketing fluff—it’s about creating the kind of evidence that influences the belief split Paulson documented. If you can show that your human-in-the-loop process consistently outperforms pure AI, you give audiences and search engines a reason to prefer you.
But the warning from the Harvard data is symmetrical: if you cannot demonstrate that advantage, the default preference will shift toward the cheaper, faster algorithm as soon as its output reaches parity. The P&G study shows that AI-assisted teams already produce higher-quality ideas than unassisted individuals. The question for SEO is whether the industry is still unassisted, or whether competitors are already using tools to widen the gap.
The Real Error in the Harvard-MIT Joke
The kid with the overloaded cart wasn’t wrong about the math. He just couldn’t read the sign. Harvard handed our industry both halves of that problem in the same report: a hard number on how little moral cover we actually have, and a fairly precise description of the one thing still buying us time. Get the count and the read right, or we’ll be the ones getting rung up as the error.
The count is a 2.31 on a 7-point scale—negligible moral objection. The read is that only transparent, verifiable performance can sustain the human preference. The research from Riley, Paulson, Sadun, Lakhani, and Neeley collectively tells a coherent story: the public does not care who does the work as long as the work is done well. The window for SEO to prove it still does it better is closing, and the only honest strategy is to make that proof undeniable, measurable, and publicly available.
Featured Image: Andrey_Popov/Shutterstock
