Hi everyone, this is Neo.

Let me start with a question that may determine how secure you feel for the next three years: do you think the public actually cares whether SEO work is done by humans?

For years, the whole industry — myself included — has operated on a default assumption: Google keeps rewarding named human bylines and E-E-A-T (Experience, Expertise, Authoritativeness, Trust) signals because the public holds some moral preference for content created by humans — people don’t feel right about letting machines write articles and run marketing. So we marketed “human-made,” “human authors” as our moat.

A batch of Harvard research just dragged that assumption across the pavement: the public has almost no moral objection to AI taking over search marketing work.

1. A moral test across 940 occupations: search marketing is second from the bottom

First, the data trail. The discussion starts with a working paper released in October 2025 — “Performance or Principle: Resistance to AI in the U.S. Labor Market” — by Harvard Business School Assistant Professor James Riley and postdoctoral fellow Simon Friis. HBS’s Working Knowledge ran a feature on it in February 2026, and Search Engine Journal columnist Greg Jarboe pulled it back into the spotlight on September 7 to shake the SEO industry awake.

How was the study built? Riley’s team asked 2,357 American respondents to score, on a 1-to-7 scale, how morally objectionable it would be to hand each of 940 different occupations to a machine — roughly 23,570 occupation ratings in total.

The results are brutal for SEO:

• Search marketing strategists scored 2.31 — in the ten occupations Harvard charted, only file clerks ranked lower
• For contrast: clergy scored 5.91 and childcare workers 5.86 — professions the public treats as nearly untouchable
• Based on AI’s current capabilities, the public supports fully automating roughly 30% of occupations
• When the question shifted to “imagine a future AI that outperforms humans at lower cost” — support nearly doubled, to 58%
• About 42% of occupations sat in an ambivalent middle: not strongly opposed, just “depends”
• Only about 12% of occupations drew strong moral resistance no matter how capable the AI — funeral directors, athletes, artists and the like

Watch that near-doubling: 30% to 58%. Riley’s own conclusion — resistance to automation is mostly about whether the technology can do the job yet, not about principle. Plainly: people aren’t stopping AI from taking your job; they just think the AI can’t take it yet.

And here’s one number worth extra attention: 94% of respondents favor using current-day AI to augment human work (rising to 96% for a hypothetical advanced AI). The public has zero problem with “AI helping me work faster.” It also has surprisingly little moral baggage about “AI replacing you” — as long as it genuinely does the job better and cheaper.

2. Experiment two: human preference is a belief, not a stance

Riley’s experiment measured attitudes. The one by HBS professor Elisabeth Paulson and UC Berkeley’s Kirk Bansak measured behavior — and it stings just as much.

They ran a conjoint experiment with 9,000 participants: choose between a human and an algorithm to approve a loan or decide a defendant’s pretrial release. On average, holding performance equal, people did lean human — by 4.3 percentage points for the loan and 7.6 points for pretrial release.

Sounds like a win for humans? Hold on. Look at the belief split buried in a chart on page 13 of the report:

• Among respondents who already believed algorithms outperformed humans: 56% chose the algorithm for pretrial release, 54% for the loan
• Among respondents who believed humans were better: 63% and 59% went with the human

See the pattern? Choosing human or algorithm isn’t a moral position — it’s downstream of a belief about who is currently better at the job. And the kicker: “fairness” (equal treatment across racial groups) turned out to be the least important factor in how anyone judged either decision-maker.

Paulson’s own read: if you can demonstrate real accuracy gains without other metrics slipping, “that’s probably sufficient.” Her data shows the human preference is not a moral fortress — it’s an output of perceived competence. Once the AI is demonstrably better, the preference flips. That lines up almost exactly with Riley’s feasibility argument, even though the two studies were designed to test completely different things.

3. Experiment three: 791 developers at P&G — AI-assisted teams crush it

Think the first two studies are too far from SEO? This third one tests exactly the kind of work we get paid for: ideas and creative output.

Harvard’s Raffaella Sadun, Karim Lakhani and co-authors tracked 791 product developers at Procter & Gamble — some working alone, some in teams, some with an internal GPT-4 tool, some without. The results:

• Ideas ranking in the top 10% of quality were three times more likely to come from AI-assisted teams than from unassisted individuals
• Employees using AI also reported higher enthusiasm and energy, and less anxiety and frustration, than those working alone without it

Read that slowly. Coming up with ideas in product development is the same class of cognitive work as the content ideas and strategy SEO sells — and the data shows human-plus-AI teams are visibly pulling away from human-alone. Riley’s data says a competence gap is the only thing currently protecting your job; the P&G study says that gap is closing in real time. Harvard’s Tsedal Neeley and Expedia’s Ritcha Ranjan wrote the technical note pointing to what comes next: agentic AI acting as chief of staff, competitive intelligence analyst, and executive coach, running with minimal human oversight once set up. Neeley’s advice to leaders: start deployment on the “no-joy” work — the repetitive tasks nobody wants — before handing over anything high-stakes. That’s the classic automation on-ramp: prove yourself on the boring stuff, then creep upward.

4. Google and the AI engines are running the same test

The sharpest thing in Jarboe’s column is connecting all this research to our daily reality. His core argument:

The industry has assumed Google keeps rewarding named human bylines and E-E-A-T because the public has some moral stake in SEO staying human work. Harvard’s data says that stake doesn’t exist.

So why does Google still emphasize E-E-A-T? Because Google’s ranking systems — and increasingly, the citation behavior of AI answer engines — are running the same test Paulson’s respondents ran: comparing the human-produced version against the machine-produced version, and asking which is still demonstrably better. Human preference tracks perceived competence; so do ranking systems. Once the machine version overtakes on quality and cost, the preference flips — not because the moral wall gives way, but because the data already crossed the line.

5. So what do we do? Three survival strategies

Jarboe closes with three recommendations. Let me translate them into executable moves for cross-border content teams:

First, put a real, checkable human name behind anything AI touches before it goes external. Not a ghost “Editorial Team” byline — a real person with a LinkedIn profile, credentials, and a verifiable track record that readers (and crawlers) can cross-check. Paulson’s belief-split data says preference follows perceivable competence. A byline isn’t a moral gesture — it’s the anchor that lets machines and readers attach a competence signal to your work.

Second, publish your performance record, not just your process. If you or your content/SEO program has produced measurable results, put the receipts inside the piece itself — case data, test results, before-and-afters. That’s the accuracy demonstration Paulson says actually moves people; your own track record can do in reverse exactly what the AI is doing.

Third, reserve full automation for the boring, repeatable “no-joy” work, and keep a named human on anything that touches trust or money. Internal link audits, meta-description drafts, log-file triage — automate freely. Anything touching reader trust (bylined opinions, deep content) or a client’s money (strategy, budget recommendations) keeps a human accountable. Riley’s data says that’s the one line the public still won’t fully cross — but it’s a far narrower line than most SEOs assume, and right now it’s the only one left worth defending.

Neo’s take

I read this article twice. The first time as industry news, the second time as a mirror.

For practitioners, the sobering part is this: stop marketing “human-made” or “human-crafted” as a moral moat. That moat doesn’t exist in the data — the public doesn’t care, and honestly, neither do clients. They care about one thing: where you’re actually better than the AI, and what the evidence is. I keep meeting SEOs and agencies whose business card says “100% human-written original content,” as if that were a premium badge. In the AI era that phrase is quietly mutating from “quality guarantee” into “capability confession” — it’s telling clients you don’t plan to use AI to raise your own output ceiling.

The only real moat has never changed: use AI to produce work at a quality level pure humans can’t reach — and be able to prove it. The P&G numbers are the perfect footnote: the best ideas don’t come from pure humans; they come from AI-assisted human teams. Tools don’t eliminate the people who use them — they eliminate the people who don’t. And with 94% of the public welcoming AI as an augmentation tool, the market has already pointed the way: stop competing on the “replacement” track and take the “augmentation” track.

For independent site owners and content leads, there’s an even more practical layer: E-E-A-T is not about looking human — it’s about proving a human exists. Google and the AI engines are running Paulson-style tests: attach a verifiable human identity to the byline, back the content with checkable performance evidence, push automation to the back-office grunt work, and keep real people in the front-office trust positions. This playbook doesn’t require you to lose sleep over whether AI will replace you. It requires answering one question: when the machine runs its “who’s better” test, what evidence does your site have to show?

One last honest, uncomfortable point: Harvard gave search marketing a 2.31 not because the public dislikes the profession — but because the public doesn’t care who does the work. It only cares about the outcome. So stop arguing that the work should stay human. Argue instead who can do it better — and then be that answer. That’s the one thing AI can’t take from you.