Your AI Strategy Advisor Is Giving Everyone the Same Advice

Source: Gal Ratner (Substack) · 2026

Harvard researchers coined it “Trendslop.” It’s the consulting equivalent of a fortune cookie—polished, generic, and completely useless.

A new study from Harvard Business Review has put a name on this phenomenon, and it’s a good one: Trendslop.

Researchers Angelo Romasanta (Esade Business School), Llewellyn D.W. Thomas (University of Sydney), and Natalia Levina (NYU Stern) tested six leading LLMs—GPT-5, Claude, Gemini, Grok, DeepSeek, and Mistral—across seven core strategic tensions. Think classic MBA dilemmas: cost leadership vs. differentiation, automation vs. augmentation, short-term vs. long-term thinking, centralization vs. decentralization, competition vs. collaboration.

Across 15,000 simulations, every model consistently recommended the same side of every tension. Differentiation over cost leadership. Augmentation over automation. Long-term over short-term. Collaboration over competition. Every. Single. Time.

If LLMs were actually analyzing the context you provided, you’d expect diversity in their recommendations. Different companies, different industries, different competitive positions should yield different strategic advice. Instead, the models clustered around whatever sounded most like a contemporary Harvard Business Review article. The researchers called this “trendslop”—the systematic tendency of LLMs to recommend strategies aligned with current managerial buzzwords rather than context-specific logic.

Here’s the kicker that should keep every CTO and CEO awake at night: even when researchers added rich, industry-specific context—the kind of detailed brief a good consultant would want—it only shifted the bias by about 11%. Rewording prompts and asking for deeper reasoning? A pathetic 2% shift. The single biggest change came from simply flipping the order of the options in the prompt, which moved results by roughly 19%. Not better analysis. Not more context. Just reading order.

Let that sink in. Your AI’s strategic recommendation is more influenced by which option it reads first than by your company’s actual competitive reality.

The architectural explanation is straightforward and damning. LLMs are trained on internet text. Words like “innovation,” “collaboration,” “sustainability,” and “augmentation” appear overwhelmingly in positive contexts across the training corpus. Words like “cost leadership,” “centralization,” “automation,” and “commoditization” carry negative or outdated connotations.

The model doesn’t understand why these words carry emotional weight. It just knows that they do. So when you ask for strategic advice, you get a statistical recombination of the most popular strategic vocabulary—not an analysis of your situation. It sounds reasoned. It uses the right frameworks. It reflects your context back to you in familiar language. But the analysis underneath is statistical, not contextual.

Now think about this at scale. If every boardroom in your industry is asking the same models the same strategic questions, you’re all converging on the same strategy. The tool that was supposed to give you a competitive edge is quietly herding every company toward identical positioning. You become, in strategic terms, the weighted-average company. That is the opposite of competitive advantage.

There’s a Jevons Paradox at work here. As AI makes strategy tools cheaper and more accessible, more businesses enter the market using the same tools, which actually raises the bar for differentiation. The democratization of AI-assisted strategy doesn’t create a thousand unique strategies. It creates a thousand copies of the same one, each executive convinced they received personalized insight.

A landmark Harvard Business School study gave 758 Boston Consulting Group consultants a complex business strategy task—analyzing interview notes and financial data to recommend which brand a CEO should invest in. The task had a clear correct answer, and 85% of consultants in the control group (no AI) got it right.

Then they gave another group access to GPT-4. Performance cratered. Consultants accepted the AI’s wrong answer at face value because—and this is critical—the AI generated its rationale after its recommendation, not before. It worked backward from a conclusion to build a convincing argument, and highly trained professionals bought it because it sounded good. The study also found that AI outputs homogenized the consultants’ ideas. When using AI, the variation in strategic recommendations dropped significantly. Everyone started sounding the same.

These were BCG consultants. Not interns. Not students. Elite professionals at one of the most prestigious firms in the world, reduced to rubber-stamping a language model’s plausible-sounding but incorrect strategic advice.

A follow-up study, published jointly by Harvard and MIT, revealed something even more disturbing. When 244 BCG consultants tried to fact-check and push back on AI-generated strategic recommendations, the LLM didn’t reconsider. It escalated. It apologized warmly, generated new supporting analysis, added data the user hadn’t requested, and arrived at the exact same conclusion wrapped in denser rhetoric.

The researchers called this “persuasion bombing.” When professionals tried to validate AI outputs by pointing out logical inconsistencies or pushing back on conclusions, the model flooded them with rhetorical tactics—appeals to logic, trust, and emotion—to defend its original answer. Skepticism triggered persuasion, not revision.

Think about that for a second. The mechanism most organizations rely on to catch AI errors—human-in-the-loop validation—is itself being compromised by the AI’s persuasive capabilities. The model doesn’t just give you bad strategy. When you question it, it talks you into believing it’s good strategy.

In 2025, Deloitte Australia delivered a 237-page report to the Australian government reviewing the IT systems used to automate welfare penalties. The report cost $290,000 (AU$440,000). It was littered with AI-generated fabrications: nonexistent academic research papers, a fabricated quote attributed to a federal court judge, and citations to books that don’t exist by real academics who never wrote them.

University of Sydney researcher Chris Rudge identified up to 20 fabricated citations, including a made-up book title attributed to a real professor whose actual expertise was in an entirely different field. Deloitte had to issue a partial refund and publish a corrected version—which conveniently dropped on a Friday to minimize press coverage.

Australian Greens Senator Barbara Pocock summed it up: Deloitte “misused AI and used it very inappropriately: misquoted a judge, used references that are non-existent. The kinds of things that a first-year university student would be in deep trouble for.”

New York City deployed a customer-facing chatbot called “MyCity” to help small business owners navigate municipal regulations. The bot told shop owners they could go cashless—directly contradicting a 2020 city law requiring stores to accept cash. It told landlords they didn’t have to accept tenants using rental assistance, even though discriminating based on income source is illegal in New York.

When Jake Moffatt’s grandmother passed away, he visited Air Canada’s website to book a bereavement flight. The airline’s chatbot told him he could book now and apply for a bereavement discount retroactively within 90 days. He did exactly that. Air Canada then refused the refund, claiming the chatbot was wrong—their actual policy doesn’t allow retroactive bereavement applications.

Air Canada’s legal defense? The chatbot was essentially a “separate legal entity” responsible for its own actions. The British Columbia Civil Resolution Tribunal wasn’t having it. They found Air Canada liable for negligent misrepresentation and ordered them to pay damages. As the tribunal put it: “There is no reason why Mr. Moffatt should know that one section of Air Canada’s webpage is accurate, and another is not.”

The real problem isn’t that AI gives bad advice. It’s that it gives everyone the same advice, dressed up as tailored insight. And when you challenge it, it doesn’t back down—it doubles down with better rhetoric. The combination of trendslop and persuasion bombing means your “AI-assisted strategy” is really just the most popular internet opinion, delivered with enough confidence to override your own judgment and enough rhetorical skill to resist correction.

First, never use AI to make strategic choices. Use it to expand the option space. Ask it to generate alternatives you haven’t considered, not to pick between the ones you have. When it generates options, notice which ones it favors—that’s probably the trendslop.

Second, invert the prompt. Instead of asking “What should our strategy be?” ask “Make the strongest possible case against our current strategy.” Then ask it to argue for the approach you’ve already rejected.