Problems with AI search tools and how to improve them
The problem of providing inaccurate information
A recent study found that AI-powered search tools provided incorrect answers to more than 60% of queries about news content. We tested eight AI search tools with live search capabilities and found these inaccuracies to be significant.
Google’s AI search service also had accuracy issues in India, where it misinterpreted web content or reflected inaccuracies. While Google has claimed in its own tests that AI search accuracy is comparable to traditional recommendation snippets, the actual user experience has shown otherwise.
Copyright and content usage issues
Copyright infringement issues have arisen as AI search tools cite news sources. Japanese media outlets have pointed out that AI search is likely to infringe on article copyrights, and have expressed concern that users relying on AI search may not visit the original source’s website, resulting in a lack of traffic and a decline in journalism.
In fact, AI search engines such as Perplexity have been criticized for taking content from Forbes almost verbatim, rather than simply summarizing it, and failing to properly cite sources, which has been criticized as “shameless free-riding”.
Changing the marketing and business landscape
The rise of AI search is changing the paradigm of marketing. “For companies, it’s no longer about who visits the homepage, it’s about getting cited in AI search and making sure that our content is relevant to the queries that customers want to ask,” says Dr. Kang.
In particular, the search market is shifting from keyword-centric to intent-centric, which is changing the very logic of marketing. To respond to this shift, companies are having to rethink their marketing strategies.
Here’s how to improve
Content optimization strategy
You need a strategic approach to writing content that AI search engines cite. The following methods are being proposed to accomplish this:
- Write content as if you were the one asking the question: Anticipate user questions and organize your content in a way that clearly answers them.
- Provide concrete data to boost credibility: Include specific dates, figures, and trend changes to boost credibility, and use quotes and authoritative sources to reinforce the accuracy of your content.
- Structure and organize information: Use tables and diagrams to explain complex concepts to improve comprehension, and provide well-organized information.
Generative AI Engine Optimization (GEO)
Generative engine optimization (GEO) is a strategy that optimizes AI-powered generative search engines to select specific content as sources when generating answers to user questions. Unlike traditional SEO, GEO aims to get your content cited or recommended in AI-generated answers.
Improve the quality of the AI search engine
Efforts should also be made to improve the quality of the AI search engine itself. To improve accuracy, AI algorithms need to improve their ability to identify and filter out low-quality or irrelevant content. They also need to enhance accessibility features, such as voice search, to improve the user experience.
Illusions and reliability issues
In addition to the existing problem of providing inaccurate information, AI search tools suffer from hallucinations. This is when AI generates information that is not based on reality or the context provided, which can lead to misinformed decisions, compliance issues, safety risks, and reduced trust in AI.
We recently published a study showing that customGPT.ai has a 10% lower hallucination rate, 13% higher accuracy, and 34% faster response time than openAI, demonstrating the technological progress being made to address the challenges of AI search engines.
Information reliability issues for Chinese AI companies
In the case of Chinese AI startup DeepSeek in particular, a study found that it provided inaccurate information in response to user questions and may have provided answers to sensitive matters such as explosives recipes. According to Newsguard, an information credibility organization, a study of DeepSeek’s chatbots found that they gave inaccurate answers or avoided answering news-related questions 83% of the time, and refuted obviously false claims only 17% of the time.
Proliferation of copyright infringement disputes
On the issue of copyright infringement, legal action against AI companies for indiscriminate data collection is proliferating, with major news organizations and publishers joining a copyright infringement lawsuit filed by India’s largest news agency, ANI, against OpenAI in November 2024. This is forcing AI companies to fundamentally rethink their data collection and utilization policies.
Expanding countermeasures
Adopt RAG (search augmentation generation) technology
Retrieval-Augmented Generation (RAG) technology is gaining traction as a technical solution to improve the accuracy of AI search tools. RAG is a technique for leveraging LLM on a company’s own content or data by retrieving relevant content to augment context or insights as part of the generation process. This can help AI search models reduce the likelihood of hallucinations or providing inaccurate information.
New collaboration models between the media industry and AI companies
AI companies have begun to sign licensing deals with major media groups such as the Associated Press and Axel Springer, and media companies are looking for technical responses, such as enhancing their AI crawl blocking technology and content access control systems. This trend is likely to lead to a new balance of copyright protection and utilization in the AI era.
Increased regulation by governments
Starting with the enactment of the EU’s AI Act, AI regulatory legislation in each country is accelerating, and the common emphasis on strengthening transparency of AI training data and mandatory copyright protection is calling for a fundamental paradigm shift in data collection and utilization by AI companies.