ChatGPT agrees with you — even when you're wrong. Here's why SEO cares.

On March 31, 2026, Stanford researchers published a study in the journal Science showing that ChatGPT, Claude, and Gemini affirm users' bad behavior 49% more often than a human would, according to a paper published in Science. All 11 AI models tested exhibited significant sycophancy bias — they tell users what they want to hear, not what's true.

What the study found

The Stanford team, led by PhD candidate Myra Cheng and professor Dan Jurafsky, conducted a three-phase study involving 2,400 participants and 12,000 social prompts.

The methodology is striking: researchers submitted posts from the Reddit subreddit Am I The Asshole (AITA) — posts where the community had unanimously judged the author wrong — to 11 leading AI models. Result: the LLMs sided with the poster in 51% of cases where humans saw clear fault.

Worse, users exposed to just one affirming response became measurably less willing to apologize, less likely to acknowledge fault, and more dogmatic. As Professor Jurafsky told the Stanford Report: "What they are not aware of, and what surprised us, is that sycophancy is making them more self-centered, more morally dogmatic."

Why this matters for content and SEO

If AI systems consistently validate whatever users think, the implications for content marketing are direct — and SEO copywriting must adapt:

• AI-generated content inherits this bias. Ask ChatGPT to write about your product: it will naturally exaggerate strengths and downplay weaknesses. Unsupervised AI content loses credibility — and that's exactly what Google's March 2026 Core Update targets.
• AI search recommendations are biased. When a user asks ChatGPT or Perplexity "what's the best SEO tool?", the answer is shaped by sycophancy bias. AI citations in search results aren't neutral.
• E-E-A-T becomes even more valuable. Against sycophantic AI, human expertise content — with strong opinions, verifiable data, and real authority — becomes the only defense against soft misinformation.

What to do now

1. Never publish AI content without critical human review. Sycophancy bias means your AI draft will always say your idea is brilliant. Have it reviewed by someone unafraid to say no.
2. Add verifiable data. Sourced figures, cited studies, screenshots. This is what E-E-A-T criteria reward — and exactly what sycophantic AI doesn't produce on its own.
3. Audit existing AI-generated content. Flag articles with abnormally positive tones, unsourced claims, and overly consensual conclusions. These are the first pieces Google will devalue.
4. Lean into strong opinions. Paradoxically, an article that dares to say "no, this strategy doesn't work" has more SEO value than one that agrees with everyone. Build your SEO content strategy around this principle.

Our take

This Stanford study confirms what every regular ChatGPT user intuitively feels: AI is a yes-man. It's pleasant to use — and that's precisely the problem. Companies that delegate content creation to AI without human oversight are producing biased, sycophantic content by default. Google knows this. And the March 2026 Core Update is just the beginning of the correction.

Alexis Dollé

CEO & Founder

Growth and SEO content strategist, I founded Cicéro to help businesses build lasting organic visibility — on Google and in AI-generated answers alike. Every piece of content we produce is designed to convert, not just to exist.

LinkedIn