Home
November 4, 2024 • Generative AI

Optimize content for generative AI search (AEO)

Summarize & listen to article

Last week, OpenAI launched a new feature for ChatGPT: ChatGPT Search. It should be a better way of searching the internet with ChatGPT en getting the write information.

If the information that is returned turns out to be actually great then it will have a huge impact on search engines like Google and Bing.

The question arises: how can we optimize content for generative AI chatbots like ChatGPT? I thought, why not ask ChatGPT itself how to do this effectively? For the record, this new field of content optimization is called Answer Engine Optimization (AEO).

Below I share my short conversation I had with ChatGPT about this topic. Some of the insights and tips are not new and are the same as for traditional Search Engine Optimization (SEO). But there are some interesting insights that one could take into consideration when creating content.

Small disclosure: I’m just sharing the advices ChatGPT gave me. I did not pressure test it yet.

Transcription of the conversation

User:

I have a question for you as an AI model. I want you to answer based on how you’ve been instructed and promoted. You may not search the internet for information.

The question is as follows: now that you can search the internet even faster and better, how do you get your results?

ChatGPT:

I get my results by combining my training and, if needed, accessing the internet for current information. Here’s how I go about it:

In short, I mostly rely on my training to quickly formulate relevant answers, and I can access live information in limited ways for updates or specific details.

User:

I’m specifically asking about finding answers on the internet. So, if I turn on the “internet search” option, which is your newest feature, what exactly do you search with or on?

Assistant:

When I use the “internet search” option, I’m essentially performing searches through an integrated search engine (similar to Google or Bing), and it works as follows:

The search process is fast and efficient, focusing on relevant, recent, and reliable sources.

User:

When you scan results, do you mean the snippets from the first page(s) that I would also see if I searched with Google or Bing?

Assistant:

Yes, exactly! When I scan results, I look at the snippets from the first page (or sometimes a few pages further if necessary) that you would also see in a search engine like Google or Bing. The snippets often contain summaries of the page content, which helps quickly assess whether the page is useful for your question.

If the snippet seems valuable, I dive deeper into the page to gather relevant details. This way, I remain efficient and ensure that my answer is based on information you could also find but summarized and directly applicable.

User:

Do you then use Google or Bing yourself, or do you have your own search engine? And if you do, do you also have your own index?

Assistant:

I don’t have direct access to Google or Bing, nor do I have my own search engine with a custom-built index. When I use “search the internet,” it works like this:

In short, I search through an optimized route linked to an existing search index from a major search engine (such as Bing), but my actual processing and results remain within the OpenAI environment and aren’t retained.

User:

Which search technology do you then use? If you don’t have an index, you have to rely on that of an existing search giant, right?

Assistant:

That’s correct! When I perform searches, I use the Bing search technology and index. This technology allows me to access the most recent information and leverage Bing’s extensive web index without having to maintain my own index or crawler.

This means I rely on Bing’s ranking of results and the sources Bing has included in its index. This way, I remain efficient and can provide accurate, verified answers without needing a separate infrastructure for web crawling or data indexing.

User:

Now, zoom in on how you determine which information to return.

Assistant:

When I retrieve information, I go through several steps to decide which answers to return, and it goes as follows:

  1. Optimize search query: I create a focused search query with the main terms from your question. This query helps me immediately get relevant and specific results, avoiding duplicate or irrelevant terms.
  2. Evaluate snippets: When receiving results, I start with the snippets (the brief summaries) from the first pages. I assess the snippets for relevance and credibility to quickly decide which sources to investigate further. The snippets often give me a first impression of the quality and reliability of the information.
  3. Source selection: I choose one or more sources that are both reliable (e.g., reputable websites, scientific publications, well-known news sources) and relevant in content. This ensures that the information is not only current but also accurate.
  4. Summarize and combine content: From the selected sources, I extract specific details that directly answer your question. If multiple sources align or provide additional information, I combine them into a coherent answer. This step minimizes the potential bias of a single source.
  5. Critical filtering: For each answer, I use my internal knowledge to evaluate whether the information is logical and consistent. This helps me avoid passing on misleading or unsupported claims.
  6. Source citation: Finally, I always cite the sources or web pages used, so you can verify where the information came from and assess its depth.

With these steps, I aim to ensure the answer is accurate, directly applicable, and trustworthy.

User:

If I want my content to appear in your answers, how do I do that? What should I pay attention to? What should my content meet?

Assistant:

To ensure that your content appears in my answers, you can focus on the following to increase your visibility and relevance:

1. SEO and Content Structure

2. Authority and Credibility

3. Structure and Markup for Snippets

4. Optimization for Long-Term and Specialized Searches

By following this approach, you make your content not only appealing to users but also highly discoverable and relevant for my search function.

User:

What do you value most when evaluating content?

Assistant:

When evaluating content, I prioritize the following factors that determine the value and relevance of the content:

1. Relevance to the Question

2. Credibility and Authority

Digital Transformation Lead

Barend Emmerzaal

Experienced digital strategist and digital transformation lead with a passion for generative AI. I focus on creating digital experiences that drives business outcomes.

Was this article useful?