We believe that LLMs can also be exploited to verify if your online presence accurately communicates the concepts that are most important to you. Whether it's through articles about your work, your website, social media, or blog, you want to ensure you've conveyed your key concepts effectively. Sometimes, you might assume something is understood when it's not well communicated. Therefore, a simple interaction with an LLM can be a new way to check your indexing and communication skills.
Here's a two-question test for ChatGPT4:
THE CHAT
Me: I have a company with a product for personalizing LLMs. The company is called Memori and the product AIsuru. First, search everything you can about this topic so you know every feature, vision, and everything about our product.
ChatGPT4 (after a search through the function): Your company, Memori, has developed AIsuru, a platform focused on managing and personalizing language models (LLMs), like ChatGPT or Mistral. AIsuru stands out for its ability to create customized artificial intelligences that understand and integrate into various business and creative processes, enhancing interaction and effectiveness. It offers a unique approach by developing conversational digital twins that learn and develop personalities based on interactions, experiences, and contexts. A notable feature of AIsuru is its integration with Unity applications, allowing the creation of digital twins within Unity3D games or applications. The software development kit (SDK) is designed to be user-friendly, supporting multiple functionalities like session management, multilingual communication, and sharing media or links with these digital twins. Moreover, AIsuru’s technology is not limited to digital twins. It also has capabilities like the Board of Experts, a feature ensuring targeted and high-quality AI responses by directing questions to the most specialized AI in a particular field. Additionally, AIsuru offers Open Badges for skill certification, and its flexible sharing options extend to various platforms, including the web, physical devices, and the metaverse. In summary, Memori's AIsuru is a complete and versatile AI platform that combines advanced AI capabilities with user-friendly applications, making it suitable for both companies and individual professionals. The emphasis on making AI interactions more human-like and the wide range of features position AIsuru as an innovative solution in the field of artificial intelligence and digital interaction.
Me: Search information about Deep Thought and AIsuru.
ChatGPT4 (after a second search): Deep Thought is an innovative feature offered by AIsuru, Memori's platform, representing a significant advancement in conversational artificial intelligences. This feature goes beyond traditional prompt-based interactions and immediate responses, allowing the development of a genuine personal relationship with the AI. The uniqueness of Deep Thought lies in its ability to equip AI with memory, enabling it to remember and reflect on past conversations. This transforms each interaction into a personalized experience, significantly differentiating from traditional chatbots that were limited to basic functions and predetermined paths without true understanding or contextualization of individual needs. A particularly relevant function of Deep Thought is natural language marketing segmentation. This tool allows businesses to segment their audience using only natural language, for example, identifying user groups based on specific experiences or expressed sentiments, thus facilitating highly personalized marketing campaigns. Additionally, privacy and control are central elements in the design of Deep Thought. Users have exclusive access to their memories and can control and delete them at any time, with frequent notifications about AI conversation recordings, thus ensuring user privacy protection. It should be noted that, although ChatGPT has recently introduced a memory feature, it is not comparable to the use cases of Deep Thought. Unlike ChatGPT, AIsuru is a business tool that remembers conversations and uses memories for marketing purposes, integrating with various external services like CRM, email marketing, and other API-exposing services. Deep Thought marks a paradigm shift in conversational AIs, allowing for deeper understanding, personalization, and greater engagement in digital marketing strategies, while maintaining strong user privacy control.
CONSIDERATIONS ON A NEW SEO
The two answers are perfect in terms of content. They are derived from two articles and posts between LinkedIn and the website. So the communication is correct. However, we noticed that Deep Thought emerged from the second interaction, prompting us to review our content to clarify how significant this feature actually is, changing the game from chatbot to digital twin with an active role in processes and social interactions.
We are often asked if there's a new SEO to consider now that people search more on an LLM than a search engine.
The fundamental question we posed is: is there a way to add content to a GPT4 LLM.
Example: I want my website content to be easily understood by OpenAI. How do I do this?
A technician might argue: isn't an import from the website content sufficient?
It's not exactly an import that we're doubting, but we're keen to explore how a company in 2024 can write content on its sites hoping to be properly read by an AI Company while it gathers data in building the new updated version of its LLM (e.g., GPT6)?
The blunt answer from AI experts is: there's not much that can be done. The opposite is possible (i.e., blocking crawlers) with robots.txt. So, first of all, you need a permissive robots.txt.
Digging deeper, we can say that the large text datasets we know of, like Common Crawl, carry the whole site with them, then it's up to those implementing information extraction mechanisms to filter the content and extract the relevant part. How OpenAI does this is not known. We add that the world is moving in the opposite direction, i.e., preventing the contents of one's site from being literally remembered by GPT, so OpenAI works hard to eliminate copyrighted content, people's names, phone numbers, contacts, etc. It should not be considered a search engine.
This is the first article on the topic we found, but as you can see, they are all classic SEO techniques:
https://www.digitalmogli.com/how-to-show-your-content-on-chat-gpt/
Intuitively, we can say that GPT, with its ability to generalize, will remember your content if it's applicable to many cases: if your site is the showcase of a company, it's already a miracle if it remembers that it exists, but if your site talks about techniques for doing something, it's possible that GPT will remember the technique, maybe even attributing it to the company (but it's not certain).
We also wondered if constructing content in a question-and-answer style would help. The answer is NO. An LLM is not trained this way. Only fine-tuning is done this way (exactly as we do in AISURU with the EXPORT JSONL tab).
We then did a test with a search on a partner/client company (large but not huge). In short:
- we asked a question designed to make it respond "XXX" (name of the company), indeed if we ask the same question to Google, it comes up first
- instead of that, it hallucinated (ARIA does not exist, XXX has no competitors in Italy, there are other companies called ARIA but they do different things)
- indeed, when we asked for the website it did a Bing search and couldn't give us a link
- then we asked if it knows XXX, and yes it knows it well, but it has the old company name (YYY, since 2022 it's called XXX Srl)
- we asked for the link and it found it immediately
XXX's website is an old-fashioned site, with really nothing special: a few demos, many documents, a blog link, etc. there is no FAQ or anything like that, yet it knows it very well, also because it is mentioned on Wikipedia (page managed by the same XXX) so, what can we deduce? That normal SEO techniques are fine, but most importantly, you need to have something to say.
Perhaps the key point here is that XXX has a Wikipedia page, which is certainly one of the first texts it's trained on (and therefore has more weight than subsequent ones). However, despite knowing it well, when asked to bring it up, it hallucinated.
So, at the current stage, focus on things to say so that, as you read at the beginning of this article, the concepts you care about come out, how you do certain things, and the purposes they serve. This will help you understand how to better value what you think is important or less known about you, your work, and your products.