Answer engines are the gateway into all of the knowledge contained on the internet. For decades now Google, and by extension search, has been the tool within the category, however with LLMs we are starting to see new kinds of answer engines. Broadly speaking we now have three types.
Traditional Search
The one we're all familiar with – Google, Bing, DuckDuckGo. You type a question into a text box and you get a list of links to websites from across the internet that have information that can answer your question.
Pros
- Works well across across any type of question.
- Can be used to find specific references and sources of information.
Cons
- Requires manual work to find relevant information within the returned links.
- Dominated by ads.
Chatbots with Search
Chatbots like Claude, and ChatGPT have become the fastest growing answer engines competing with traditional search, and as they begin to integrate traditional search into chatbots they become much more capable.
Pros
- Summarises information into a succinct answer so you don't need to click through and read many pages.
- Can be asked to take further action with the information it's summarised eg. write documentation, or create an image.
Cons
- Is not effective at locating specific content on the internet.
- Can hallucinate and make up false information when it can't find good sources.
Generative Search
An evolution of traditional search where results feature structured information than just plain links to other sources. At the moment we think of new tools like Perplexity, Arc Search, and OpenAI's new SearchGPT, but Google (and Bing) have also been moving in this direction and offer generative search features. In Google things like the knowledge graph have existed for a long time, and now generated summaries of search results similar to Perplexity are being released as well.
Pros
- A good balance between traditional search and chatbots for information gathering.
- Links to source material allow users to verify the accuracy of generated summary information.
Cons
- Generated results are still based on LLMs, and suffer from the same hallucinations chatbots do.
- Cost to deliver generated search results are much greater than traditional search.
We're in a time of disruption and innovation at the moment, but already we're seeing these different types of answer engines converge on similar new ways of providing responses to queries. Traditional search is rapidly bolting on generative features, chatbots are crawling and summarising the open web more and more to stay up to date, and generate search tools like Perplexity are adding chat features for follow up questions. Only time will tell which platforms will emerge on top.