CEO Sam Altman and OpenAI executives held a Reddit AMA to answer questions, including those about ChatGPT Search, providing an inside look at how it works. Their answers offer insights into what SEO may look like in the immediate future.
The people from OpenAI answering the questions:
ChatGPT Search is not a search engine, it’s an AI chatbot with search, which means it doesn’t compete with Google as a search engine, it simply replaces it with something else that people already use for work and play. Now it has additional utility as an assistant in daily life and search.
Another advantage to ChatGPT Search is that it doesn’t show advertising nor does it follow users around the Internet. Users already trust ChatGPT with personal and business information so it’s already has goodwill with users.
What makes ChatGPT Search a threat to Google is that Users are already familiar with ChatGPT and have good feelings about it. Because it’s already in use there is no switching away from Google to break the habit of searching with Google.
In the Ask Me Anything (AMA) session on Reddit, a Redditor asked OpenAI CEO what the value of ChatGPT Search is over other search engines.
The person asked:
“My question is about the value ChatGPT Search offers compared to popular search engines. What are the unique advantages or key differentiators of ChatGPT Search that would make it worthwhile for a typical search engine user to choose it?
Sam Altman answered:
“For many queries, I find it to be a way faster/easier way to get the information I’m looking for. I think we’ll see this especially for queries that require more complex research. I also look forward to a future where a search query can dynamically render a custom web page in response!”
That bit about a “custom web page” is something to look out for because it hints at personalization based on what a user is searching for.
Altman’s response about ChatGPT Search’s handling of complex queries calls attention to an advantage over Google. ChatGPT users are accustomed to using natural language, whereas Google users habitually use keyword searches. Keyword searches disadvantages Google because it’s harder to understand those queries, which is why Google displays People Also Ask features in Search.
Natural language queries is the way users interact with ChatGPT and that is an advantage for ChatGPT Search.
The next question was about OpenAI’s progress on preventing ChatGPT from making things up (aka hallucinations) and also about how it’s going to incorporate fresh data to the index.
Both problems are generally approached with a technology and technique called Retrieval-Augmented Generation (RAG) which selects data from an up to date database like a search index or a knowledge graph and then provides that to the LLM-based chatbot to summarize and use as a base for an answer.
This is the question:
“Are hallucinations going to be a permanent feature? Why is it that even o1-preview, when approaching the end of a “thought” hallucinates more and more?
How will you handle old data (even 2-years old) that is now no longer “true”? Continuously train models or some sort of garbage collection? It’s a big issue in the truthfulness aspect.”
The answer was given by Mark Chen, SVP of Research
“We’re putting a lot of focus on decreasing hallucinations, but it’s a fundamentally hard problem – our models learn from human-written text, and humans sometimes confidently declare things they aren’t sure about.”
Mark Chen continued his answer by saying that they are getting better by the use of “grounding” which is something that Retrieval-Augmented Generation (RAG) helps large language models with. Chen also reveals that they believe that using Reinforcement Learning (RL) may help models stop hallucinating.
Reinforcement Learning (RL) is a way to teach a machine with experience, rewarding it when it’s correct and withholding the reward when it’s not correct, thus reinforcing good answers. The machine “learns” by making choices that maximizes rewards. In the context of hallucinations, a reward could be a score or signal that indicates that the answer is factual (and it could also be provided by human feedback scores).
Mark Chen continued his response:
“Our models are improving at citing, which grounds their answers in trusted sources, and we also believe that RL will help with hallucinations as well – when we can programmatically check whether models hallucinate, we can reward it for not doing so.”
The next question is about what search data does ChatGPT Search use.
The question asked:
“Is ChatGPT Search still using Bing as the search engine behind scenes?”
The answer was provided by Rinivas Narayanan, VP Engineering at OpenAI:
“We use a set of services and Bing is an important one.”
That’s an interesting answer because it’s commonly assumed that Bing is the only search engine. The answer indicated that ChatGPT Search uses multiple “services” and that Bing is the most important. What are the other services that ChatGPT might use? That’s an open question.
Someone asked the important question about how to optimize content for ChatGPT Search in order to improve rankings. The question was answered by Kevin Weill who said that they were still figuring it out, which could mean that they don’t know or that they’re still figuring out what to say about optimization.
Kevin Weill, Chief Product Officer responded:
“This is a great question—the product just launched today so there’s a lot to figure out still about where search will be similar and where it will be different in an AI world. Would love any feedback you have!”
Chief Product Officer Kevin Weill is right, these are still the early days of their search and much can still change. The OpenAI Reddit AMA offers first hints at what SEO is growing into.
Other insights:
Beyond those takeaways is the consideration that OpenAI is not directly competing against Google with a standalone search engine, it has created a completely different experience for searching the web.
Featured Image by Shutterstock/Vitor Miranda