Flaws discovered in ChatGPT search engine

As the ChatGPT search tool opens new avenues in digital marketing, recent testing has revealed some surprising vulnerabilities. These flaws allow third parties to manipulate its responses using techniques reminiscent of old SEO practices, such as keyword stuffing or hidden texts.

A new era, but familiar manipulation techniques

SEO professionals remember the years when practices like satellite pages, keyword stuffing or the insertion of invisible texts were commonplace to artificially improve the positioning of sites in search results. These strategies have pushed Google to roll out major updates, such as Panda and Penguin, in order to combat this abuse.

A recent Guardian investigation highlighted potential flaws in OpenAI’s ChatGPT search tool, better known as SearchGPT. This tool, which promises to revolutionize online research, however presents worrying vulnerabilities. Tests revealed that hidden content on web pages can manipulate AI responsesrecalling the historical practices of black hat SEO. These manipulations cast doubt on the reliability of the information generated by AI tools in the long term.

When “prompt injection” meets SEO and digital marketing

The concept of prompt injection reminds security specialists of another well-known technique:injection SQL. In the same way that a hacker manipulates a database by inserting malicious commands, hidden or carefully crafted content can hijack the responses of an AI model like ChatGPT.

The implications for digital marketing are significant. Imagine a site built specifically to influence the responses of a tool used by millions of consumers. These manipulations could give rise to practices similar to black hat SEO, aimed at deceiving not search engines, but AI interpreting and summarizing content.

SearchGPT - the waiting page to register as a tester — SearchGPT – the waiting page to register as a tester

The issue of reliability for AI tools and marketers

For marketers, these findings pose a central question: can data-based tools be trusted? language models to provide reliable answers? The example of a cryptocurrency developer, having lost $2,500 following the use of code generated by ChatGPT (to access the Solana blockchain), is one of the perfect examples of erroneous or manipulated information .

Furthermore, these incidents are a reminder that the relationship between AI and research is only just beginning. Just as Google has had to refine its algorithm over the years to counter abuse, OpenAI will have to overcome similar technical challenges to secure its SearchGPT.

A promising technology, with its share of challenges

Digital marketing has always evolved according to technological innovations and the limits they impose. If ChatGPT Search represents a significant step forward, current vulnerabilities show that it will have to, like Google in its time, learn to defend itself against abusive practices. Problems that will also be encountered by other technologies around AI and online search, such as Perplexity.ai.

For marketers, this highlights the importance of adopting an ethical approach and monitoring the evolution of tools to maximize their effectiveness while minimizing the risks of manipulation.