Lexlab director Alfie Lagos
We use an enterprise subscription to ChatGPT and unfortunately the test group is only being offered to personal accounts for now. So whilst I am on the list, I only got there later as I had to manually do this via my old non business account.
The alignment of Search and Large Language Models (LLM) to give more insightful answers makes total sense, and I suspect that just like the battle between Yahoo, Google, Alta Vista (and more) in the early ‘00s where Google eventually rose to the top via their superior results and algorithm to dominate the market, we’ll see something very similar happen with GPT based searches.
A recent test we did (included below) shows it already rivals Google for insight and accuracy of results. And the reality is that since the early 2000s, search (from an advertising perspective) hasn't really evolved that much outside of adding featured snippets, videos, imagery in the results. Sure, it's now using ML and much more sophisticated algorithms and inputs to calculate the optimal relevant results, but from the consumer perspective it hasn't changed.
The same thing that made Google dominant will be the same reason it will lose its share if it doesn't come up with an evolved alternative quickly!. It will simply be easier for the consumer to get results using a GPT powered search engine.
Outside of Perplexity.ai and Bing, it's hard to find a viable alternative that gives accurate GPT fuelled responses that you can interact with. We recently did a test (below) that runs through some of the current players. Devv.ai, despite looking quite ugly, gave the only results that could remotely compete with Perplexity.ai.
- Their ability to remove the ambiguity of Bing search and their CoPilot/GPT based searches, and potentially a bridge to replace Bing to become the leader in this new generation of search.
- Normalise more consultative search responses, or ‘clarifying questions’ to avoid the hallucination behaviour we typically see when a GPT isn't confident in a response
- A legal need for better fact checking, governance and transparency of the source data.
- Potentially a less ad cluttered search environment where the top search results aren’t paid spots
In terms of challenges, the biggest one is accuracy. Ultimately the results are based on data from websites and we already live in a world where people don’t do enough of their own research, so if the masses start deferring their thoughts and answers to a GPT to decide results, we as a community may become less critical and informed as a result.
Often, the lack of a clear answer (or Featured Snippet) from Google usually was for a reason. That reason being that it wasn't a clear cut, black and white answer. Even in my below innocuous football-based test, where my follow up question of ‘how tall is the Geelong forward’ doesn't have a clear cut answer given Geelong has multiple key forwards. ‘Perplexity.ai’ answered it best by assuming I was talking about one of them but that’s a harmless assumption. What will it assume if you ask it who you should vote for, or whether a politician's claim is true or not?
From an advertising perspective, the BIG challenge will be how search engines monetise results. Currently you have to trawl down up to four paid results to get to the organic result which gives viable (clickable) options.
Maybe Google knows this, and their proven ‘do nothing and wait *cough delay*’ style will ultimately make them more money.
Making it easier for the consumer (aka searcher) will ultimately drive the direction of GPT powered search, and ultimately SearchGPT. As a result the consultative nature of talking with a GPT to get answers will dominate.
I’ll take both Google's paralysis to address the cookie deprecation and their confusingly bad competitor to ChatGPT, Gemini, as an example of where I think this will go. Google is petrified of losing their search revenue and could wait too long to roll out a commercially equivalent Gemini-powered search alternative. Meanwhile SearchGPT will mass launch, potentially replacing Bing as their primary search engine with the CoPilot or SearchGPT branded offering.
In terms of timing, mass adoption to SearchGPT will initially be slow but as SearchGPT comes online, or something like Perplexity becomes the default search engine on major tech platforms (for browsers like Edge, Safari, Firefox etc), we’ll see accelerated uptake through the back half of the five year window
And whilst Perplexity.ai seems like a very viable alternative, I can’t see how it will take on the might of Bing/Microsoft/ChatGPT’s offering. I suspect if Microsoft plays it right, Bing has a huge opportunity to flip search on its head if they can successfully manage to integrate the clear advantage of searchGPT for its users.
Broadly, while search revenue will decrease, the value exchange of ‘answering a question’ may switch to something else. In reality GPT’s will become more personal and more likely than not, you’ll have your own dedicated GPT which will live locally on your devices which removes the need for you as the individual to ‘search the internet’ for simple questions as it will just ask for you and give you the answer.
Simple (“which team is playing on the weekend” style) answers will stay clear cut, but a result for something more complex like booking a flight will become much more informed where your GPT / agent will potentially do all the research and then give you a few options. This is where paid search may end up playing a part given the limited number of options a user can realistically consume.
Here’s the test.
To show what this might look like from a practical perspective, I tested:
- google.com (the baseline)
- phind.com
- you.com
- perplexity.ai
- devv.ai
- bing.com / Copilot
Search test for simplicity and result (and locality): “Who are Geelong playing this weekend?” - it will need to know the context of where I am, that Geelong are an AFL team, and also have the insight to know the football calendar. (I initially tried where’s my nearest cafe but thought this might be unfair given the infrastructure advantage Google has with Maps and locality tracking via first party data).
Google (the current baseline): correct and formatted to give the answer instantly.
Phind.com: Couldn’t join the dots to answer the question, even though it has access to recent enough data. Fail.
you.com: confidently wrong. Fail.
Perplexity.ai: nailed it! Albeit a bit uglier than Google’s result
To give Perplexity.ai a little more of a challenge, and highlight the prime difference between Google and GPT powered search, I followed up with “What is the height of [Geelong’s] full forward?”.
The result, while a mix of the key forwards and the main image was different to the name, gave a relatively accurate answer and it gave Jeremy Cameron’s actual height.
Compared to Google’s answer, which has to be a separate search as there’s no ability to ask for follow ups, didn't give me an answer and made me go look at the links to find out, even though I gave context.
devv.ai - Surprisingly accurate in its response.
But failed in the follow up question around its full forward height here.
Bing.com results in a similar response to Google, but gives the chat option on the right to follow up the question:
And with the follow question, asking how tall their forward nails it:
