With the rapid development of AI technology, search engines, as a key tool for obtaining information, are also undergoing profound changes. However, most of the existing AI search engines focus on the single scenario of "finding answers", which is significantly different from the usage habits of traditional search engines. This article will discuss the functional characteristics of search engines in the AI era, including generating questions based on keywords, distinguishing request types and providing appropriate presentation methods, pre-reading summaries and blocking irrelevant content, and reverse tracing and verification of the source of ideas.

When we use search engines, they can usually be divided into three situations:

First, find the answer.

What year was Michael Jordan born? There is a single correct answer to this question; What is Messi's career best goal? There may not be a single answer to this one, but there are certainly a few of the most popular ones.

The second is to find resources.

Most typically, look for pirated movies or novels. Or, look for the investor section of a listed company's official website, in which case the user may be looking for a specific URL or a set of files available for download.

The third is to find inspiration.

I heard that Musk is a very powerful person, and I don't know much about him, so I searched for it. I don't know exactly what I'm looking for, it could be his personal introduction, his romantic history with a female star, or one of his TV talk shows - in any case, I'll find out what I'm looking for in the process.

Recall that all of your search engine usage history can be classified into one of the three scenarios above?

Then, when AI comes, we will find that all products under the banner of AI search engines, whether it is Perplexity or ChatGPT's search function, they basically focus on the first use case, which is to find answers.

You ask questions, and it gives you answers. The answers can be so in-depth that you need to read a report the size of an academic paper, but ultimately find the answers.

Then we soon found that there are many differences between these search engines and the search engines we are used to:

• Traditional search engines use keywords as input, but these AIs are more inclined to let you ask complete questions;
• Traditional search engines return link addresses and content summaries, and these AIs tend to only give you a conclusion that they summarize.
• Traditional search engines can easily find a bunch of useless garbage, AI does not, but AI will make up nonsense in a serious way, and open its eyes to talk nonsense.

In short, they also claim to be search engines, but they are completely different things to use.

As a result, to date, none of them have been able to replace traditional search engines, or even prove themselves to be possible.

So, in the era of AI, what should a so-called AI search engine look like that can perfectly adapt to the usage habits of traditional search engines and meet the needs of users more efficiently than traditional search engines?

I think the following features should be a must:

First, the ability to generate questions based on keyword input

Most users are accustomed to the keyword-based use of traditional search engines, and having them organize their language to give a precise question is a high barrier to entry in itself. A better approach would be that users still ask questions with keywords, and the search engine gradually converts the keywords into specific questions through multiple rounds of interaction (note that users must be asked to do multiple-choice questions rather than essay questions). For example:

User input: Jordan

Search engine:

I guess you want to know:

1. Who is Michael Jordan?

2. What honors has Jordan won in his career?

3. What has Jordan been doing lately?

4. Are Jordan sneakers related to American basketball superstars?

5. Others

The most common questions are generated by AI, and then users are asked to click on the questions they want to know in these lists, and if one round fails, it is divided into two or three rounds, which is the process of generating questions based on keywords. (In this regard, chatGPT's in-depth research function is relatively good, and it will actively ask questions about the unspecified parts of the user's needs)

Second, distinguish between different request types and give them an appropriate presentation

As mentioned earlier, there are three common use scenarios for search engines: finding answers, finding resources, and finding inspiration. The AI search engine should be able to distinguish which scenario the user is currently in, and give the corresponding way to present the results:

If the user is looking for answers, then the ideal way to present them is to first give a clear answer/conclusion (Jordan's personal introduction), and then give the content source/reference address that supports these conclusions (Baidu Encyclopedia, Wiki, NBA official website, etc....... ), and in order of credibility;

If the user is looking for a resource, then the ideal way to present the URL is to divide the URL into categories according to specific criteria, and then present it to the user in categories. For example, if I search for a free fiction website, then the search engine may divide the comprehensive category into a group, the immortal cultivation fantasy group, and the urban life group...... The most popular TOP10 is given in each group, which is very convenient;

If the user is looking for inspiration, then the AI search engine may generate something similar to the mind map style for the user, and then fill all the relevant content into a certain position of the mind guide...... Whether you're looking for a serious academic career or glamorous gossip, you'll find your target at a glance and won't be boggling around in a maze of information......

In short, according to the user's usage scenario and the characteristics of the content, you can decide what kind of presentation you want to use, rather than sticking to a fixed display template.

Third, the ability to pre-read, summarize and block irrelevant content on behalf of users

In fact, it is to let the AI quickly go through the destination URL for you, find out which content is most relevant to the current problem, and then block those spammy destination pages.

Traditional search engines,Essentially, the search results are sorted based on PR value,This gives a variety of SEO an opportunity,But in the AI era,I can rely on the advantage of computing power,Read the 20 pages you find as a whole,Skim off 0 rubbing traffic,Then delete 0 full of ads that lack substance,And then remove those that have a small relationship with the current content,Leave only a few pages that are really related to the current results and the best reading experience,This is basically equivalent to letting an assistant/ The secretary will help you check the search results again in advance.

Of course, this part of the work, the current traditional search engines are also doing, and some have also achieved good results, but it is far from the extent of covering all the return results, in fact, the most perfect situation should be for each returned search results page, in which all the links are pre-checked by AI.

Fourth, the ability to reverse trace and verify the source of any opinion

We all know that AI is easy to generate hallucinations, so we need to conduct a random reverse retrospective of any summary point of view given by AI, and find out the source, although many AI large language models have provided support in this regard (use it themselves, ingots are better at this aspect), but there are always some limitations, in the most ideal case, I can choose a line of words in the search engine return results, and I can reverse trace this point of view/opinion, Look at where the earliest source of this statement is, and search engines will also take the initiative to attach reference links to the source in some of the more important and critical places (i.e., Yuanbao and ChatGPT) to increase the credibility of the content.

In fact, the four functional supports I listed above are not too new in themselves, and some AI products often do well in one or two of them, but I have not seen products that do well in all four at the same time, and these products are often not developed and customized based on the user habits of traditional search engines.

If AI manufacturers want to truly replace traditional search engines and eat this tempting cake, then what they need to do is to use the new generation of AI technology in their hands to optimize the experience and efficiency of the original search engine, so that users can get a much better user experience than before without changing the original usage habits, so as to attract users to move to their own products. Rather than wishful thinking expecting users to change the habits they have developed over the years, forcing everyone to adapt to your new product logic, even if your logic is more AI, more advanced, and more ......

But at the end of the day, as a business, you should serve the user, not let the user adapt to you. This truth is still true even in today's era when AI is in power.