Popularity in the tech world is hard to measure. I’ve talked at length about this in my discussions of programming language popularity. It really comes down to what you use to measure popularity — and how available those metrics are to those doing the analysis.
It’s difficult to generically define popularity, especially when you’re including tools that do wildly different things. For example, is a general-purpose text-to-image generator like Midjourney inherently more popular than a tool that removes backgrounds from images like Remove.bg?
Also: Microsoft is offering free AI skills training for everyone – how to sign up
Over the past few years, I’ve been carefully refining my popularity index methodology. My indexes take a page from the world of political analysis: I aggregate survey data from multiple rankings, and weight them carefully to account for the strengths and weaknesses of each dataset.
Given that the datasets have wildly different rankings and, in some cases, showcase different tools, it makes sense to apply a normalizing process across the fields of data.
The ZDNET Index of AI Tool Popularity
With this index, I decided to measure overall tool interest, mostly based on web traffic statistics available to my source data providers. Notably, the Adobe AI tools are missing from any of my datasets, possibly because traffic volume is too low, because the Adobe tools are mostly behind a paywall, or because the Adobe tools are mostly delivered in standalone desktop applications like Photoshop.
Even so, what I can provide is an overall index of interest in the various tools, which should provide insight into where users are putting their attention. Here are the top 20 AI tools, ranked by overall popularity.
Note: If you don’t want to read through my ranking methodology, scroll down to the section entitled “The #1 thing to remember when using AI.”
Aggregating multiple data sources
As source data for this analysis, I’m using data tables from four sources. The different sources add differing levels of value to the overall aggregate based on the data they contain.
Because I have four sources, each started with an assigned weight of 25% (so they all added up to 100%). Two of the sources are older, so I took 5% from them, resulting in two sources weighted at 30% and two sources weighted at 20%.
Also: The best AI for coding in 2025 (and what not to use)
But one of the sites has only rank data and no traffic data. Data with worldwide traffic measurements provide more detail into overall popularity than simple user surveys, so I reduced the strength of the survey-only source and increased the weight of the sources with traffic data.
I pulled 6% off of the weighting of the survey-only source (6 instead of 5 because it’s easier to distribute across three sources) and gave the remaining three sources an additional 2% weighting factors.
That gives us the following sources and weights. You can see from the accompanying charts just how variant the data is among the sources. I also assigned each source a three-letter ID that was used throughout data analysis.
Exploding Topics (Weight 32%, ID XPT)
Exploding Topicsanalyzes trends using web searches, conversations and mentions. Its data is derived mainly from web analytics platforms.
Source This dataset gives us an overall ranking, estimated market share, and monthly visits. Data is current through February 2025.
AI Tools (Weight 32%, ID AIT)
AI Tools contains a directory of over 10,000 AI tools. Each tool is categorized into 171 categories. Its data is derived by traffic analysis platforms.
Source This datasetgives us an overall ranking, data on changes from previous periods and estimated market share, based on monthly visitors. Data is current through February 2025.
World Bank Group (Weight 22%, ID WBG)
The World Bank Group consists of an international development organization as well as a financial institute. In March of 2024, the organization published a policy research paper entitled, “Who on Earth Is Using Generative AI?On page 12, there is a ranking of generative AI software based on traffic.
TechRadar Survey (Weight 14% ID TRS)
By This dataset gives us an overall ranking of the US and UK, but not traffic numbers. The data, even though the article was written in 2025 is from 2024.
Aggregated tool list
Next, I built an aggregated tool list. I added the top 20 tools from each source into a table. As you can see, some tools (ChatGPT, for example) are represented in all four sources’ rankings, while some are only represented in one or two source lists.
Tools that are represented with ranking data pick up weighting based on their position and traffic quantity for each source. The exception is the TechRadar survey, which only picks up position data. If a tool is not represented in a list, it does not pick up any representation data.
From this, I started to build the aggregation spreadsheet. I took the share representation of each tool and put it on a row in the spreadsheet. In total, there were 45 tools represented. Then I pulled in the raw percentage ranking from each source, leaving cells blank where there was no data.
Also: What is AI vibe coding? It’s all the rage but it’s not for everyone – here’s why
Once that was done, I built up the next section of the analysis, which was the weighted ranking for each source. Then I totaled up the values of the four weighted rankings, which gave us our aggregated ranking.
At this point, the spreadsheet was pretty scattered (technically, it was a Sparse matrix (). It was not easy to see the ZDNET final rankings. I did a sort on the result field and that gave me the data for the ranking chart at the top of this article.
If you’re interested in spreadsheet geekery, here are the 20 first rows of my aggregate analysis.
The #1 thing to remember when using AI
Before I talk about the rankings themselves, I’d like to mention one other aspect of the analysis process: I did the whole thing by hand. Oh yes, I spent hours going down a rabbit hole with ChatGPT trying to get it to take in the datasets and spit out an aggregate, but it got stubborn.
Also: How to use ChatGPT to write code – and my favorite trick to debug what it generates
Really, really stubborn. It complained it couldn’t read the data. So I converted the data into text, but it still got confused. It began conflating results from the different sources. It lost track of its progress and we had to start over, three or four times.
I have no doubt that I could have developed a series of carefully crafted prompts that would have gotten me a file I could export into Excel, but I soon realized that the negotiating and cajoling process with the AI would take longer than filling the Nespresso water tank to brew some espresso and doing it all by hand, using the technology of caffeine to aid me.
Yes, I do see the irony of an article on AI tool popularity being done entirely without the help of AI tools. And, that, perhaps is my conclusion.
As popular as these many tools are, they’re tools. They’re sometimes helpful and sometimes stubborn. If you’re going to use them, you’ll have to be able, constantly, to determine when the tool is the fastest path and when the old-school way will get you there either faster or more reliably, or both.
What does it all mean?
I was not surprised that ChatGPT leads the pack. I keep forgetting that Canva is now considered an AI tool, so that did surprise me. To so many users, when they think of AI tools, those are the two that come to mind most.
I was slightly surprised that DeepL outperformed Google Translate as an AI tool, but that may be because most people don’t think of Google Translate as an AI tool. After all, it’s been around a lot longer than the generative AI boom we’re experiencing now.
Also: 10 professional developers on vibe coding’s true promise and peril
When it comes to chatbots, its clear that Gemini and Copilot, along with Perplexity and Claude, have a long way to go to catch up with ChatGPT. That said, Apple was completely missing from the table, which can’t be good. But then again, so was Facebook/Meta.
For now, it looks like ChatGPT is lapping all the other tools, and Canva is leading the second tier. Expect the rest to be in a pitched battle for third place, where there are no standout leaders.
Stay tuned. It’ll be interesting, if nothing else.
What about you? Have you found yourself relying more on tools like ChatGPT, Canva, or Gemini lately? Which AI tools do you use regularly, and which ones do you think are overhyped? Are there any tools you were surprised to see missing from the rankings? And how do you measure popularity — by features, community buzz, or just whatever helps you get the job done? Let us know in the comments below.
You can follow my day-to-day project updates on social media. Be sure to subscribe to Follow me on Twitter/X and subscribe to my weekly update newsletter (). @DavidGewirtzon Facebook at Facebook.com/DavidGewirtzon Instagram at Instagram.com/DavidGewirtzon Bluesky at @DavidGewirtz.com (19459110), and on YouTube: YouTube.com/DavidGewirtzTV.
