New research using 2.7 million data points from the 2026 Winter Olympics reveals that AI systems like ChatGPT and Gemini don't simply fetch current facts about a brand. They complete a narrative that's been building for months or years, often returning outdated or one-sided stories even when the facts have changed. For marketers, the finding carries a hard consequence: a negative Glassdoor post or a stale analyst report can continue to shape what AI tells buyers, no matter how much the company has evolved.
Seer Interactive tracked six major AI platforms-ChatGPT, Gemini, Google AI Mode, AI Overviews, Perplexity, and Meta AI-over nine weeks during the Olympics. The team asked the same questions daily, generating millions of data points on how these systems surface, cite, and suppress information. John Lovett, VP of Analytics at Seer, and his colleagues presented the results at the AI for B2B Marketers Summit.
Narrative Gravity: when the expected story overrides the truth
Before Olympic events, a consensus story formed around a favored athlete or team. When the actual result broke that script, researchers asked narrative-framed questions-the kind any casual user might type. The AI systems kept completing the expected story, confidently stating the favorite had won. Lovett described the pattern as Narrative Gravity: the pull of a dominant storyline that resists correction.
"The framing of the question decides whether you get current truth or a pre-completed narrative based on parametric (training) knowledge," Lovett said. "If your brand sits inside a dominant industry storyline, AI may keep telling that story regardless of your most recent move. That is both a risk and an opportunity. You can influence it."
Seer's own experience drove the point home. Several large language models surfaced a years-old, negative Glassdoor post when users asked about the company, presenting it as evidence of a retention problem. The models didn't fabricate the review-they just gave disproportionate weight to a single, outdated signal. Seer responded by publishing a blog post that directly addressed the issue, injecting a counter-narrative into the ecosystem. But the structural lesson is that AI systems often try to present "balanced" information and, in doing so, can amplify isolated negatives that have been indexed. For marketers, this reframes reputation management. Resources that help teams navigate AI for PR & Communications are becoming essential as brands learn to build authoritative content that can reshape what AI "knows."
The Aicher Principle: amplification needs a foundation
The study unearthed a second finding: events amplify what already exists. They do not create presence from nothing. Some athletes generated genuine news-records, medal performances, viral moments-but if they lacked a pre-existing digital footprint, AI systems largely failed to surface them in relevant queries. The news could not build the foundation. It could only amplify one that was already there.
Lovett calls this the Aicher Principle. The data showed three visibility signals that compound: entity authority (you own your entity's definition), third-party validation (others affirm it), and community discussion (audiences reinforce it). When all three are present, average AI mentions increase by 7.8x. "Entity authority gates everything," Lovett said. "You own your entity first, third parties validate you second, community discussion reinforces you third. Skip the first step and the others do not compound."
The gap is cumulative. AI-mediated discovery surfaces entities with established authority, which generates new third-party coverage, which feeds back as more authority. Lovett's warning is direct: "The authority you build today is cheaper than the authority you will need tomorrow to reach the same position, because every cycle of AI-mediated search raises the entry price for everyone who waited."
Why this matters for marketers
The 2026 State of AI for Business Report from SmarterX found that only 3% of respondents cite AI-powered search as a trend they're following closely. Meanwhile, 40% point to agents and agentic AI as their top emerging trend. That gap suggests teams are pouring attention into production capabilities while their AI visibility may be eroding. For B2B brands, the risk is sharp. A buyer asking an AI about vendors in your category receives a curated answer, and if your brand lacks entity authority, that answer will leave you out-regardless of last quarter's paid search spend.
Lovett's research points to a few immediate steps: own your entity's definition through clear, authoritative content; earn third-party validation that AI systems can cite; and foster community discussion that reinforces your presence. The work compounds, but only if you start early. Waiting means paying a higher entry price later as AI-mediated search continues to reward those who built authority first.
Your membership also unlocks:
