Large Language Model Optimisation (LLMO) — How web pages are found by LLMs

For the following article, my AI assistant has also created a podcast episode. Those who prefer to listen rather than read can access the podcast via the following link (CAUTION: podcast is created by AI only, no guarantee for accuracy).

Listen to the podcast episode of Sophie’s AI assistant

Success stories like Logikcull’s show that LLMs, like ChatGPT, are having an increasing impact on the search behaviour and ultimately the purchasing behaviour of customers. In the case of Logikcull, the company itself was surprised when more and more new customers said they had found out about the software company through ChatGPT. As early as June 2023, five percent of all leads for Logikcull are said to have come through ChatGPT. That’s the equivalent of nearly $100,000 in monthly subscription revenue for the company (source: OMR).

The field around Large Language Model Optimisation (LLMO) is still very new, little researched and there are hardly any truly valid and measurable results. Nevertheless, there are initial approaches to how we can align our websites for the future of LLMs. In the following article, I will provide some initial tips.

What are Large Language Models (LLMs)?

Large Language Models means something like ‘big language models’ in German. These models are types of artificial intelligence (AI) that are trained to understand, generate, and interact with human language. They can write texts, answer questions, create summaries, and more by being trained on enormous amounts of text data. These models recognise patterns, structures, and connections in the data on which they were trained and use them to generate new content that meets the requirements of the users.

One of the best-known LLMs is currently probably ChatGPT or GPT-4o (Generative Pre-trained Transformer) from OpenAI: With ChatGPT, we can already perform a variety of tasks, such as writing articles, translating texts, answering questions, creating a wide range of media and writing code.

What is Large Language Model Optimisation (LLMO)?

Large Language Model Optimisation (LLMO) is an advanced approach in the field of online marketing and artificial intelligence that aims to influence the output of large language models, such as ChatGPT or Perplexity. In simple terms, this technique is similar to classic search engine optimisation (SEO).

Targeted interventions in the training data or the optimisation of the content available to LLMs can be used to promote or influence specific results. However, the mechanisms and methods of LLMO differ from those of SEO. LLMs work differently from traditional Google search. Consequently, companies must also choose different methods to optimise their own website for LLMs.

Why is LLMO important?

Consumer surveys, such as the one shown below, clearly illustrate that a growing number of users are directing their queries to ChatGPT or other Large Language Models (LLMs) instead of turning to established search engines like Google. This trend is expected to continue, with more and more consumers initiating their internet searches via ChatGPT and similar services. Consequently, the field of website optimisation for LLMs will become increasingly important.

Anyone who thinks they can wait until there is solid research on LLMs should consider that waiting too long can be risky. Optimising for LLMs takes time. An LLM cannot adapt its algorithm overnight. As is already known from Google, website adjustments take a few weeks to affect rankings. It can be assumed that the effects of Large Language Model Optimisation (LLMO) will take even longer to achieve visible success.

Another reason why companies should not postpone the topic of LLMO for too long is the competitive advantage. Companies that are already looking into LLMO are likely to gain a significant competitive advantage in the foreseeable future.

What are the best practices for LLMO?

As already mentioned, there has been little research and few in-depth tests around LLMOs so far. However, if you consider how an LLM works and how it derives its answers based on training data, the following tips or best practices for LLMOs can already be defined today.

1. Spreading your brand online

LLMs often draw on database websites, knowledge aggregators or other large publishers, such as the Financial Times or Forbes. Database websites are, for example, company directories or rating platforms. Knowledge aggregators are platforms such as Wikipedia or YouTube.

Thus, one possible measure in the context of LLMO is for companies to work on being listed by other websites and, in particular, by other well-known databases. Specifically, this means that companies must use PR measures or even purchased content to secure a ‘place’ on other LLM-relevant websites.

It is important to note here that it is not purely about backlinks, as we know it from SEO. It is really about being mentioned by name, in the right context. And that on websites that are of high importance for the LLM.

2. Optimise your own website for LLMs in terms of content

To create the following tips for optimising your own website for an LLM, I have conducted a number of expert interviews with SEO specialists, AI researchers and LLM developers over the past few weeks.

1. Adaptation to the structure of the LLM

It is recommended to ask the Large Language Model (LLM) directly to find out how it would describe or structure a particular topic, product or service. The selected topic should be similar to the field of activity of the company that is using Large Language Model Optimisation (LLMO). This structural basis can then be used by the company to create its own texts. It is advisable to use structured data. This method may seem like simple imitation at first, but it is an effective strategy for being considered by an LLM.

2. Use clear and information-rich language

LLMs process information most effectively when it is presented in simple, clear and information-rich language. Companies should take this into account when creating website texts. Structures that present advantages and disadvantages or comparisons are particularly well suited for processing by LLMs.

3. Avoid excessively long continuous text

Long continuous text without subheadings is difficult for LLMs to process. Such websites tend to be given less attention by LLMs than those that provide quick answers. It is recommended that longer texts begin with a brief summary of the main points. All content should then be prepared in such a way that it can be optimally processed by an LLM.

4. Incorporation of quotations from relevant personalities

Including quotes and references can also be beneficial. Experts in LLM optimisation assume that quotes from well-known personalities can positively influence findability and relevance in LLM results. The use of statistics and quantitative data is also considered beneficial for the LLMO strategy.

5. Continuous optimisation

Search engine optimisation (SEO) experience shows that this is an ongoing process. LLMO requires similar or even more extensive efforts. Companies should not rest on their initial successes, but must continuously work on the further development of their strategies. It is important to keep an eye on both technical innovations and the activities of competitors.

How can I measure the success of LLMO?

Although the techniques surrounding Large Language Model Optimisation (LLMO) are not yet fully defined today, the methods for measuring success are comparatively easy to implement. In the following, I recommend two methods that can already be used effectively:

1. Customer surveys

As mentioned in the introduction, customer surveys provide valuable insights into the channels through which consumers become aware of a company or its products and services. Many companies already integrate appropriate feedback options at the end of the ordering process to capture this information. It is recommended that these feedback options be expanded to include the answer options ‘LLM’, ‘ChatGPT’ or ‘generative AI technology’. With this simple addition, companies can quickly determine whether LLMs already play a role for their customers. In addition, more extensive surveys or focus groups can be conducted to provide deeper insights, for example, into consumers’ trust in the results of LLMs compared to the search results and product suggestions of traditional search engines like Google.

2. Website analytics

I was surprised myself when I noticed via Google Analytics that more and more users of my website are coming from platforms like Bing or Perplexity. Both are well-known LLMs that cite my site as a source for certain queries. This observation shows that even classic web analytics tools can provide initial indications of the share of traffic that companies are already receiving via LLMs compared to classic Google searches.

However, it should be noted that ChatGPT does not cite any direct sources in the search results. Therefore, there is no direct link from ChatGPT to websites, which means that visitors coming from ChatGPT cannot yet be clearly identified in website traffic. This highlights the need to continuously refine and adapt analytics methods to develop a comprehensive understanding of the dynamics and impact of LLMs on web usage.

Conclusion: Large Language Model Optimisation (LLMO) — How websites are found by LLMs

In conclusion, Large Language Model Optimisation (LLMO) is a relatively new field of research in which established best practices are limited. However, this fact should not be used by companies as an excuse to continue neglecting to address this issue. Instead, it is advisable for companies to start looking into LLMO now, to gain initial experience and thereby secure a significant competitive advantage.

What’s next?

Would you like to delve deeper into the topic of LLMO and discuss it with me without obligation?

Then write me a message with your requests and questions and we’ll arrange an appointment. Just send me a message via WhatsApp message or via email.

By the way, this post is also available as a podcast episode

Attention! The podcast was created entirely by my AI assistant based on my post — no guarantee for incorrect content.

Listen to the podcast by Sophie’s AI assistant

Further articles on this topic

If you are interested in this topic, please also read my article on ‘How Generative AI technologies are changing SEO measures?’. In this article, I show why companies need to consider which measures in addition to SEO and LLMO today. Incidentally, the article is also available as a podcast via my AI Assistant.

And Neil Patel also writes some interesting tips for optimising your site for ChatGPT in his article How to Rank Your Website on ChatGPT.