Originally published at: AI Answer Engines Are Worth Trying - TidBITS
The search engine has been a key Internet tool since the early days. Before the Web was a thing, I fondly remember Archie (from “archive” without the “v”), which indexed filenames on FTP servers, followed by Jughead and Veronica (Very Easy Rodent-Oriented Net-wide Index to Computerized Archives), which extended the index-and-search concept to Gopher menu titles. Web search emerged through two distinct approaches: AltaVista pioneered automated full-text indexing that could rapidly crawl millions of Web pages, while Yahoo began as a human-curated directory that organized websites into categories. Google later took search to the next level by combining automated indexing with PageRank, an algorithm that determined a page’s importance based on how many other sites linked to it. Despite occasional challenges from competitors like Bing, Google has remained the dominant search engine for decades, to the point where the company name has become synonymous with search.
Although complaints about Google’s search quality decline are more recent, my disillusionment began a few years ago when I grew tired of the extent to which Google tracked and monetized everything I did. (I understand the tradeoff; I just hit a tipping point with everything else that was happening during the pandemic.) I tried the Ecosia search engine that Apple added to Safari’s search bar in 2020, but perhaps because it was based on Bing, I found its results lacking, just as I had with DuckDuckGo. Shortly thereafter, however, another new kid on the block, Brave Search, performed well for me and remained my go-to until early 2024 (see “Brave Search Public Beta Offers Alternative to Google,” 8 July 2021).
Although I wasn’t unhappy with Brave Search itself, over my last few years with it, I felt that the pages it was finding were of lower and lower quality. Many of my searches involve looking for confirmatory details about Apple-related topics I am at least roughly familiar with, so I can’t help but judge the results. I would often find myself reading more of the linked pages to find one whose information I trusted. Too many of them repeated what was written elsewhere or contained easily identified errors.
Looking for better results, I decided to try Perplexity. Unlike Brave Search’s limited AI summaries, Perplexity offers comprehensive AI-powered answers. It performs a search and then uses generative AI to respond to the query in narrative form, incorporating information from multiple sources. Perplexity also cites sources for each statement of fact, enabling you to confirm its claims and get more details if desired.
When OpenAI enabled ChatGPT to search the Web, I switched to using it on one of my machines, and it has performed similarly well. Then I came across You.com, which initially started as a search engine but has since evolved into a more comprehensive general AI assistant with search capabilities. Claude added Web search to its Pro accounts. And Google opened limited testing of its AI Mode, which replaces the list of links with AI-generated answers like the others. So many possibilities!
Search Engines Versus Answer Engines
To distinguish these tools from traditional search engines like Google and Bing, I’m calling them “answer engines.” Although they are performing live Web searches for you, the focus is on answering your question rather than displaying the results of the search. I see answer engines as the next step in networked knowledge acquisition because they fundamentally change how we find and absorb information online:
- From lists to synthesis: Early tools like Archie and Veronica simply showed you where files were stored. Web directories like Yahoo organized websites into categories. Google ranks Web pages by relevance. Answer engines go further by actively processing and combining information from multiple sources to provide direct answers to your questions.
- From keywords to natural language: Remember having to get your search terms exactly right? Early tools required precise text matches. Search engines have improved at handling misspellings and understanding related terms, but you must think like a search engine. Answer engines allow you to ask questions naturally, and they take the context embedded in your question into account.
- From isolated lookups to conversations: With traditional search, each query stands alone—researching a related question requires starting afresh. Answer engines maintain context throughout an entire conversation, allowing you to explore topics more deeply without needing to rephrase or restart.
- From raw results to smart summaries: Traditional search engines give you a reading assignment—a list of potentially relevant pages for you to investigate and synthesize. Answer engines do that work for you, analyzing multiple sources and presenting a coherent answer while still citing their sources so you can verify the information.
Procedural Differences Between Searches and Answers
To understand why answer engines represent such an improvement, let’s examine how their process differs from traditional search.
Since the rise of Google, we have become accustomed to searches returning a list of links that we hope will provide the information we want, but that experience is far from ideal. For every question you have, you must:
- Formulate your search: This step requires distilling your question into a set of keywords that will match the desired pages. We’ve all gotten decent at this, and it works relatively well with keywords that are relatively uncommon and unambiguous.
- Review the results list: You must review the list of returned links to determine which to investigate. Most people start at the top because search engines rank the results. However, if you prefer certain sources over others, you may need to look further down the list and read more closely.
- Evaluate the most likely source: Next, you have to open the first link that seems likely to answer the question, read or at least skim the page, and evaluate whether it’s helpful and accurate. People tend to skim very quickly, encouraging Web designers to emphasize page layouts that prioritize headings and easily parsed lists.
- Repeat as necessary: If the page lacks the information you want or you’re dubious about its accuracy, you must return to the list of links and repeat the process. Savvy searchers often open multiple result pages in new tabs, so it’s easier to switch from one to the next when comparing them.
- Recast the search: In the worst case, you may have to redo your search because it wasn’t sufficiently clear, perhaps because of choosing generic or ambiguous keywords. (For instance, finding information about Apple’s bundled apps has become more difficult now that they have generic names like Calendar and Photos instead of iCal and iPhoto.)
The process is different for answer engines:
- Ask your actual question: Instead of trying to come up with a set of keywords or worrying that multiple searches will be necessary to assemble data you can combine, go straight for your answer. For instance, I had a three-column Excel spreadsheet that I wanted to learn how to print “snaking” on one page (so the column stopped at the bottom of the page and restarted back at the top). Instead of searching on “Excel Mac print snaking columns” and sorting through videos and pages that addressed the topic in varying levels of accuracy and specificity, I asked, “In Excel for Mac, is there a way to print a thin column so it snakes from the bottom of the page back up to the top and down again?” (The answer is no, you can’t do that easily in Excel, but you can simulate it by moving the data to a Word table.)
- Verify source material: As always, AI-based summarization can make mistakes or be fooled by inaccurate sources. Before chatbots had Web access, asking them to cite their sources was sketchy at best. When their responses are informed by Web searches, however, they have no problem correctly linking to their sources. How often you verify sources depends on the stakes of your query—you might double-check sources for important technical information but not for background information queries.
- Refine or expand the search: In many cases, asking the question is all you need to do. However, because these answer engines evolved from the chatbot milieu, you can always continue the conversation. Perhaps you forgot to specify that you were referring to the Mac version of Excel, or you want to know if you can use Pages instead of Word to create snaking columns. All you have to do is continue the conversation. Perplexity goes beyond the others by suggesting further avenues of investigation so you can explore the topic space further with just a click.
The key benefit of answer engines is that they typically provide you with exactly the information you want, with no additional effort required. They’re a bit like Wikipedia in this way—you could fact-check statements with the listed sources in a Wikipedia article, and you can often learn more by following links to related topics, but most of the time, you’re happy to read the article and move on.
Of course, that’s only true if the answer engine actually answers your question. I’ve been using Perplexity or ChatGPT for about a year now, and I have been happy with the quality of the responses.
To quantify that opinion and compare against the others, I built a Keyboard Maestro macro that sends my questions to four answer engines simultaneously, opening each in a Split View pane in Arc for side-by-side comparison. I evaluate all the answers and then give each a score from 0 to 3 in a Google Sheet. (0 means the answer is completely wrong, 3 means it’s completely right, and I assign 1 or 2 for partial credit.) The ratings are still subjective and specific to my searches, not yours. However, ChatGPT and Perplexity are currently neck-and-neck with average scores of 2.63 and 2.61, respectively. Google AI Mode is third with 2.24, and You.com brings up the rear with 2.06. I don’t yet have enough data for Claude to feel confident in my results, but at 1.89, its early results are not promising.
Answer Engine Limitations
It’s important to realize that answer engines aren’t superior to traditional search engines in all situations. In particular, they struggle with navigational searches—when you know where you want to go but not the exact URL. In such cases, there’s no question for an answer engine to answer—the user simply wants a link to the destination.
For instance, if you’re in the market for an iPhone, a Google search for “iPhone” will present Apple’s page at the top for a quick click. When I sent that search to the answer engines, only ChatGPT gave me a link to that page. The more action-oriented search “buy iPhone 16” triggered the desired link in ChatGPT, Perplexity, and Google AI Mode, but Claude didn’t pick up on the desire for navigation.
This limitation explains the recent study conducted by Columbia University’s Tow Center for Digital Journalism, which strongly criticized the accuracy of the answer engines when searching for news. In fact, what the study tested was the answer engines’ ability, given an excerpt from an article, to identify the corresponding article’s headline, original publisher, publication date, and URL. In essence, they were asking for a specific page. That may have been necessary research methodology for reproducible results, but even the study authors admit that it doesn’t reflect typical user behavior. Additionally, although I haven’t confirmed this suspicion, I suspect that the answer engines aren’t nearly as quick as Google at indexing breaking news.
Other specialized searches that work better in traditional search engines include visual searches that return images instead of text, real-time information such as sports scores and flight status, searches for local businesses that benefit from integrated maps and business details, and other location-based information that requires an interactive map. Even if you switch to an answer engine for most searches, you may wish to keep a Google toolbar icon a click away for quick navigational searches.
Choosing an Answer Engine
Ultimately, only you can determine whether you’ll find an answer engine effective. It will depend on what you’re trying to achieve and how well you can refocus your brain to ask questions instead of performing keyword searches. Here are my recommendations, along with the details you’ll need to use each:
- Perplexity: For most people, I recommend Perplexity. Its success rate is high, and you can use it for free. I haven’t seen the need to subscribe to its $20-per-month Pro plan for more in-depth searches. Its search URL is
https://www.perplexity.ai/search?q=%s, or you can use the Perplexity Mac app. I found using the app more awkward than searching directly in Arc. - ChatGPT: ChatGPT is as good as or perhaps slightly better than Perplexity in terms of answer quality, but its search capabilities require a $20 per month Plus subscription. Only use it if you’re already subscribing. A search URL that works for me is
https://chatgpt.com?q=%s. OpenAI also offers a ChatGPT Mac app, but it requires a Mac with Apple silicon. - Google AI Mode: For now, the free Google AI Mode is only a Google Search Labs experiment, so you have to request access. I wasn’t impressed by Google AI Mode, though its connection to the Google index may make it better at breaking news. Once you have access, its URL is
https://www.google.com/search?udm=50&nord=1&q=%s. - You.com: Like Perplexity, You.com is free to use, or you can expand its capabilities with a $20-per-month Pro plan, but its success rate was sufficiently low for me that I wouldn’t recommend trying it first. Its URL is
https://you.com/search?q=%s. - Claude: I hear highly positive things about Claude as a chatbot, especially for coding, but its Web search capabilities are weak and require a $20-per-month subscription. Apart from the poor answers, Claude has usability issues. It requires the user to approve URL-filled prompts with a click and then click again to submit the search. Its URL is
https://claude.ai/new?q=%s, or you could try the Claude Mac app.
You’ll note that these answer engines, besides the experimental Google AI Mode, require or encourage a subscription. Currently, only Perplexity is experimenting with advertising using sponsored follow-up questions, which are easily ignored. (I hadn’t noticed the sponsored follow-ups in real-world usage.) The company says it will never share personal information with advertisers.
Configuring Browsers to Use Answer Engines
The leading browser makers—Google, Apple, and Mozilla—have never been enthused about adding alternative search engine options. It’s not surprising, given that Google Search accounted for $175 billion in 2024 (about 57% of Google’s revenue) and delivered roughly $20 billion to Apple and $500 million to Mozilla. Search is big business.
Nonetheless, with a little effort, you can change the default search engine in all three browsers on the Mac. Once changed, all searches you initiate from the location bar will go to your new answer engine rather than Google. Changing search engines is more difficult or impossible in Safari on the iPhone.
(Personally, I use Arc Search on the iPhone, which has its own Browse for Me answer engine and lets you select Perplexity as the default search engine. However, given The Browser Company’s lack of meaningful updates to Arc and Arc Search since late 2024, due to chasing the fever dream of a more mainstream browser, I can’t recommend Arc Search if you’re not already deeply invested in Arc.)
Google Chrome and Chromium Browsers
You can configure most of the answer engines directly within the search settings for Chrome or another Chromium-based browser like Arc, Brave, Edge, or Vivaldi. Navigate to Settings > Search Engine, click the Add button, and enter the details for your answer engine, replacing the query with %s—the necessary URLs are above. Click Add and, once back in the list, click the three-vertical-dot menu and choose Make Default.
This approach isn’t supposed to work for ChatGPT, although using https://chatgpt.com?q=%s worked fine for me. If necessary, you can instead make ChatGPT the default search engine by installing the ChatGPT search Chrome extension.
Safari
Apple doesn’t allow custom search engines in Safari. However, the Customize Search Engine extension lets you configure additional search engines using the same approach as Chrome. After you’ve installed and enabled the extension, click its toolbar icon to access its settings.
Firefox
It doesn’t appear that you can configure search engines directly in Firefox. Instead, you need to add a search extension. I’ve found unofficial extensions for Perplexity and ChatGPT, as well as an official one for You.com. Once installed, choose Firefox > Preferences > Search and choose the desired one from the Default Search Engine pop-up menu.
Will Answer Engines Hurt the Internet?
I’ve left the most controversial aspect of answer engines for last. Publishers have long been willing to let search engines index their websites in return for traffic, which they can monetize through ad revenue and subscriptions. Many publishers even allow Google and Microsoft crawlers to index paywalled information so it serves as a teaser for those who follow such links from search results.
(We should also remember that a vast amount of content comes from sources other than for-profit publishers. Consider government databases, university resources, personal blogs, open-source documentation, non-profit research publications, and public forums where people freely share their expertise.)
One of the criticisms of answer engines is that they send significantly less traffic back to the sites from which they obtain their source material. That’s indisputable—the entire point of an answer engine is that it answers your question rather than making you read all the sources independently.
At the heart of this issue is the conflict between what is best for the user versus what is best for businesses that generate revenue on the Internet. It’s the same argument publishers make against ad blockers and that Facebook makes against Apple’s Ad Tracking Transparency. Users hate ads, and most people understand that the surveillance advertising industry is morally and ethically bankrupt. Nonetheless, advertising remains the dominant model for many publishers.
Technological advances often disrupt long-standing business models. The rise of the Internet radically reshaped commerce, fundamentally altering how businesses interact with customers and disrupting brick-and-mortar stores. The music industry underwent a seismic shift due to the advent of digital music and streaming services, which hurt physical CD sales. Smartphones created an ecosystem of mobile apps that impacted numerous industries. Online advertising captured a significant portion of ad spending from traditional media outlets. Craigslist and similar platforms demolished conventional classified advertising, tanking newspaper revenue. We live in a constantly changing world, and everything must evolve or fade away.
Although not everyone has realized it yet, generative AI is a sea change on par with these previous advances. As the cognitive scientist Alison Gopnik has noted, large language models are cultural and social technologies, much like writing, print, and markets, as well as library card catalogs, Google, and Wikipedia. All of these technologies enable people to access, synthesize, and leverage information that others have created or accumulated. Given this trajectory, it’s inevitable that AI will become increasingly woven into our daily digital interactions.
So yes, I think business models predicated on eyeballs and attention will suffer, and companies that rely on such models will have to adjust their approaches to survive. That’s undoubtedly stressful, but given the multitude of ills surrounding advertising, perhaps this will encourage companies to refocus on serving customers, rather than exploiting them. It’s also likely that content licensing or revenue sharing of some sort will play a role, at least for larger publishers.
However, what I would prefer to see is a system that pays micro-royalties based on the materials used to generate responses. The technical, legal, and social hurdles to implementing such a system are significant, but some form of business collaboration between for-profit content creators and AI companies will be necessary in the long run.
For now, though, answer engines represent a real improvement in how we find and absorb information online. While traditional search engines remain best at taking you to specific pages, services like Perplexity and ChatGPT excel when you need a direct answer or want to explore a complex topic. The more you use them, the better you’ll become at formulating questions that get at the heart of what you want to know, and the more time you’ll save compared to traditional searching.
