Bospar has published a survey on how well Americans understand artificial intelligence, finding widespread confusion about how AI systems generate answers and assess brands.
The poll of 1,021 U.S. adults examined public understanding of AI hallucinations, the sources AI models use to form responses, the role of media coverage in brand visibility, and the risks tied to dependence on large language model providers.
One of the starkest findings centred on hallucinations, when an AI system produces an incorrect answer with confidence. Only 23% of respondents correctly identified that kind of error as a hallucination.
Most chose other explanations instead. Some 26% blamed bad data sources, while 30% pointed to coding errors or a filtering rule, suggesting many people struggle to distinguish between flawed source material and the predictive nature of generative models.
Brand visibility
The poll also pointed to limited understanding of how companies appear in AI-generated answers. Just 29% of respondents understood that consistent coverage in high-authority media is the main factor affecting a company's visibility in AI responses.
The same share, 29%, recognised what Bospar described as the main long-term effect of AI on public relations: changing how brands build credibility online. Meanwhile, 41% identified authoritative coverage in credible publications as the most effective long-term way to improve AI search visibility, and 46% said a footprint across authoritative publications helps a brand become more visible over time.
Another finding suggested confusion over the sources AI systems rely on when discussing brands. The survey found that 72% of respondents did not know AI systems draw on earned media and third-party journalism to synthesise information about a company.
Views were mixed on individual platforms. While 51% recognised LinkedIn as an important source for verified corporate identities, only 35% correctly identified that online forums are crawled based on contextual relevance.
Newsroom use
The survey found somewhat greater awareness of how newsrooms use AI than of how AI models work. A total of 58% of respondents correctly recognised that newsrooms mainly use AI for research, summarisation, and drafting with human oversight.
Questions were drafted with the help of several AI models, then checked across systems including ChatGPT, Claude, Gemini, Copilot, DeepSeek, Meta, and Mistral before the survey was fielded.
The poll also tested awareness of concentration risk in the AI market. Bospar found that 58% of respondents did not recall a warning by Microsoft Chief Executive Officer Satya Nadella that AI dominance could hollow out industries by allowing a small number of frontier models to absorb a sector's knowledge.
Only 43% correctly identified loss of institutional knowledge and unique value creation as the main risk of outsourcing AI infrastructure to a small group of large language model providers, according to the findings.
Curtis Sparrer, Principal, Bospar, said the results showed a gap between public use of AI tools and public understanding of the systems behind them.
"The findings expose a serious disconnect for brands," said Curtis Sparrer, Principal, Bospar. "While frontier and foundational models rely heavily on earned media and third-party validation to synthesize answers, the vast majority of Americans have no idea how this process works."
The study arrives as companies across sectors reassess how they appear not only in traditional search results, but also in answers produced by chatbots and AI search tools. That shift has raised new questions for communications teams about the role of published reporting, external references, and digital reputation in shaping machine-generated summaries.
Bospar framed the issue as Generative Engine Optimisation, or GEO, a term used to describe efforts to influence how brands are surfaced in AI-generated outputs. The survey suggests the concept remains poorly understood by the broader public, even as AI tools become more common in everyday information gathering.
Methodology details released with the findings said the online survey was conducted among U.S. adults aged 18 and over. Respondents were drawn from an online panel and targeted by demographic characteristics, with a stated margin of sampling error of plus or minus 3 percentage points at a 95% confidence level.
Sparrer said the findings should prompt a rethink of how companies approach trust and visibility in AI systems. "This data is a brutal wake-up call for communications professionals and brand executives alike. When nearly three-quarters of the American public can't correctly identify how brands earn AI visibility, it tells us that the entire conversation around AI and marketing has been happening at the wrong altitude. Investing in authoritative earned media is essential to your brand's visibility. AI doesn't index your ad spend or your owned content. It indexes trust. And trust, in the age of generative AI, is built one credible byline, one media mention, one third-party validation at a time," he said.