ZD 25.07: The New Search Engines on the Block

You can use ChatGPT, Perplexity and Gemini as your search engine. Should you?

Conversations with the youth. Questionable disorders. Looking into your dark side. The algo ignores you. AI is great but AI might also be stifling innovation. Is ChatGPT better at search than Google or Perplexity?

The Distilled Spirit

Disorderly Youth

🧒 Young Men in the Age of Trump (Esquire)
Esquire interviews a dozen young men in the aftermath of the election. They feel a lot like people coming of age at any point — they are concerned about making a living and finding their way in the world. They also seem to do a remarkable job of seeing through the BS and focusing on what matters to them. There is also a clearly Trumpy vibe among the under 25 set.

🔊 Are Fishing or Gaming Disorders (

Play Makes Us Human

)
Peter Gray has argued very strongly that applying addiction-like terminology to people using the internet or playing games is misleading and counterproductive. He compares this sort of diagnosis for his habit of fishing while he was young to point out how ridiculous the current standards are. Let the kids play.

Deep Thinking

😈 Embrace Your Dark Side (Big Think)
People have tried many methods to manage dark emotions. We are learning that these emotions are parts of being human. They are useful beyond the primitive state. Rather than using CBT and reframing emotions away one can leverage emotional calculus to steer things where you want them to be.

👎 Why The Algorithm Hates You (

Cognitive Wonderland

)
Tommy Blanchard explains the math behind why that dumb post by the influencer has thousands of like but your thought piece wastes in obscurity. Practice in writing makes you a better poster, so the more prolific and popular keep rising to the top and creating sampling bias.

AL Love / Hate

❤ Product Leader Life Improvements (

Creator Economy by Peter Yang

)
Product leadership is an interesting job where you need to process, filter combine and collate a massive amount of information from various sources. It is the kind of job where the right AI tools and strategy could help immensely. Peter walks you through a week of how you could do this in that role. Lots of takeaways for other jobs too.

😟 Is Deep Research a Knowlege Worker Job Killer? (Every)
In case you missed it, OpenAI’s Deep Research is very, very good. Good enough it has many fearing for their employment. Evan Armstrong looks at how Deep Research will effect his own economic value and if the world will need fewer researchers like himself.

😠 AI Is Stifling Tech Adoption (Vale)
The devil is in the defaults of popular tools. AI’s propensity to pick particular tech stacks, like React and Tailwind, is leading to those stacks drowning out many others. This begs the question how default answers from these tools will effect other parts of life as that becomes the basis for knowledge, like Google has been for the last decade.

The New Search Engines on the Block

Search engines have been a pretty steady thing throughout my adult life. Google won the war about the time I came of age and that was that. If I am being honest, much of my technical success boils down to being very skilled at getting results from Google. Now that I am entering middle age, it seems that things are changing rapidly. Generative AI has fundamentally changed the way search engines can serve their results.

ChatGPT, Google Gemini, and Perplexity all claim they can replace traditional search engines. How much of that claim is true? Could they replace the tried and true? I ran a few experiments across all three platforms. How well do they work for the sorts of things I have traditionally left to Google? Do we have replacements or mere complements?

Tooling, Interface and Impressions

Google goes simple — you get the traditional Google textbox with minimal customization options or advanced search filters. There is not even a choice to include or not include web data. This is the new search box for all intents and purposes. For our tests, we used Gemini Flash Advanced 2.0 with no settings changes or tweaks.

ChatGPT is the most refined. They grew up as a chatbot, and you need to explicitly click the search button to get the web search mode. Once you do, it is very seamless — the UI flashes it is searching the web. One thing that ChatGPT does better than their competitors is how they integrate sources. Having the domain names in the text is a really good touch and can help inform a reader passively in a way the footnote numbers that Perplexity uses do not. For our test, we used ChatGPT 4o with the web search enabled.

This sort of web search is Perplexity’s main use case — so they have a bit more optionality around it. First and foremost, you can choose sources beyond a general web search — academic journals and social media are what are offered today. In addition, they have a number of LLMs you can choose from including a brand-new deep research mode. For our tests, we used the web mode and the default “auto” setting for the LLM options.

Testing was done on a desktop PC using a browser, not the mobile apps so that experience could be different. Given how non-deterministic these tools are, your results could vary quite a bit. I went for breadth not depth in the protocol.

Question #1: Current Cultural Events?

‘What is going on at All-Star Weekend?’

I wrote this during NBA All-Star Weekend 2025. I used this ongoing event as a test event to check to see how well it found things about a cultural or sporting event. Here are my rankings:

1. ChatGPT won this one hands down. It dove in and explained the new format and made sure I understood the important parts. It made sure I knew when the game was on and where I could watch it. ChatGPT used a handful of sources.

2. Perplexity appeared to give the most extensive answer — but they really failed to explain the updated format for the game itself. It did do a very good job of laying out the other events of the weekend. The tool gets bonus points for laying it out in a list rather than making me find the data in a wall of text. Perplexity did do a bit of research — it had eight sources for the response.

3. Gemini was very brief compared to the others — it just listed a bulleted list, without even listing the time for the game starting mere minutes from the time of writing. It really did not do much source checking, exposing a single source for its limited statement.

ChatGPT provides more detailed and engaging responses than the competition, making the event discussion feel more informative and dynamic.

Question #2: Where to Go to Lunch?

‘What are a few good places to take a client for lunch near Union Station DC?’

I picked this as a test of how good their local information is — there has been a lot of changeover in restaurants. Are they effectively staying current and local? Results were not great unfortunately.

1. Gemini did the best overall. They did not share a lot of options, but they all exist and were open for lunch, which makes them better than the others by default. This is a case of the trove of Google Maps information shining through to some extent.

2. Perplexity should have won for thought, care, and formatting. Unfortunately, their first choice is only open for lunch on Friday. The rest of their suggestions were reasonably solid choices and everything did exist.

3. ChatGPT started the list with a place that has been closed for six months. That is disqualifying. Their UI is very advanced compared to the others, embedding a map that you can flip to the list rather than putting them side-by-side.

Google's wealth of local data gives it a clear edge in this category.

Question #3: Weather Inquiries

‘What is the weather in DC?’

My most common Google query is “weather in DC.” It is the quickest way to access their very effective weather widget on many platforms. I wanted to see how the chatbots could replace this function. I tested two ways — one just asking for the weather, a second asking about the snowstorm coming next week.

1. Perplexity did the best overall. They appear to pop a weather widget for every weather-related query. It is a really nicely designed one with great functionality. They did not do well with the snow next week question — no real discussion of chances of the storm coming or accumulations.

2. ChatGPT had some interesting strengths and weaknesses. They had a nice hourly forecast widget, but it started at midnight and you had to expand it to get to past 7 AM. Another interesting feature is they liked quoting snow totals in centimeters, but at least we have snow totals. I did really like how they introduced a discussion of the 60/40 chances of us getting snow at all next week.

3. Gemini just popped the same weather widget for any weather-related question, with no discussion or description. It is a nice widget, but I would have appreciated a bit more information here.

A slight edge to Perplexity, but in all cases, the weather information as presented was accurate.

Question #4: Upcoming Events

‘What are some good soccer matches on this week?’

Figuring out details about upcoming events is another common use of search engines, and a great test for how these new tools were looking at piles of very similar but factually different data and making answers based on it.

1. Gemini showed Google’s advantages in understanding information in context. They picked a pretty good slate of matches — three EPL matchups and three Champion’s League second legs and the Derby d’Italia and even put all the games on the right dates and times. Decent advice overall.

2. ChatGPT successfully picked matches from the upcoming week and presented them well. They somehow missed the Champions League 2nd leg matches and the Copa Liboratores qualifying round so they still need to work on their selection criteria. If you relied upon this you would have missed things.

3. Perplexity flat out failed this test. The matches they reported happening were happening, but they had many of them on the wrong day completely. At the same time they missed the two top line international club tournaments like ChatGPT.

Bottom line: stick with World Soccer Talk’s upcoming matches page.

Question #5: Shopping for a Specific Item

‘I need a dark brown tweed sports coat with a blue plaid pattern.’

My trusty old dark brown sports coat has been getting a bit threadbare after a decade of service. I went shopping for a replacement across the new school search engines. It was not a great experience, even if there were flashes of potential.

1. Perplexity has a better shopping experience than the others. The bar is not too high. They did list products and successfully linked you to the products on the web, in the cases where you could not buy direct through Perplexity. Beyond the first page, Perplexity also has a drill-down feature where you click on the name of something and that triggers a more specific search about that item. The main place Perplexity fell down is that they did not do a good job of finding brown and tweed.

2. ChatGPT had a minimal, elegant output with a list of items and pictures. Items were so-so — the first choice was plaid not tweed although the site called it tweed. Beneath the surface all was not well, most of the links were bunk. I rarely got to a shopping page for the right product making the experience ineffective.

3. Gemini, interestingly, was the worst in terms of utility. It did not give me a link — but it did actually give me a good list of coats to check out. Having access to Google’s product data helps a ton here.

None of the new-school options worked well here. For contrast, I put the same term into Google and it was much more effective. It found real products in the right color, and the interface let me scroll through a lot of them to make an effective selection. Sometimes newer is not better.

Time to Switch?

If I had to suggest a pecking order, this is my advice for this point in time:

• Perplexity is actually pretty good for shopping and finding items. I suspect they would be even better had I used one of the more advanced models, but that would not be fair to the competition.

• Gemini has clear advantages in terms of local and real-time information.

• ChatGPT was not quite as good as Sam Altman claims it is.

My conclusion from this exercise was that there were no clear winners and that it is premature to switch completely to a LLM-based search engine. For a lot of things — especially the little, fact-based things like when is this event and finding a specific website — Google still beats any LLM including their own. They haven't been pushed off their throne yet, though the competition is getting closer. That said, the LLM search engines are very effective in some cases. The ability to interrogate your search results is a powerful new paradigm that remains to be effectively explored.

The Look

Now AI tells you how it thinks about how to lie to you.

I need to analyze the output from the web.open_url call, which likely returns an RSS feed. Since we don’t have actual text output . . .

Did you enjoy reading this post? Hit the ♥ button above or below because it helps more people discover great Substacks like this one and it helps train your algorithm to get you more posts you like. Please share here or in your networks to help us grow!