SOFTWARE 4 min read

Why the TIOBE Index Doesn't Work in 2026

Why the TIOBE Index Doesn't Work in 2026

In the March 2026 edition of the TIOBE Index, its creator Paul Jansen was asked whether the ranking should start using AI models instead of search engines. His response was brief and confident:

“No. LLMs are trained on the same web pages that we count.”

At first glance, that sounds reasonable. If both search engines and AI models draw from the same Internet, why expect a different picture?

The answer is simple: search engines and AI models do fundamentally different things. A search engine counts mentions of a programming language. An LLM understands why the language is being mentioned.

A search engine sees a reference and increments a counter. It doesn’t know whether the mention is a critique, a deep technical discussion among experts, or a student asking about homework. AI models understand context, sentiment, and intent - and therefore produce a far more accurate reflection of real-world relevance.

TIOBE was built for an era when developers shared knowledge through public forums and blogs. Today, much of that knowledge exchange happens through LLMs, AI-assisted coding tools, internal chats, and private systems that leave no public trace. TIOBE still measures something real, but that “something” is increasingly disconnected from the industry’s actual behavior. The index captures only the visible tip of the iceberg.

The experiment

To see how well AI models might perform compared to TIOBE, I ran a small experiment. I asked five AI models to rank the 20 most mature, well‑maintained, and widely appreciated programming languages. No scientific rigor, no complex methodology - just a quick test of their internal understanding.

I asked Claude, ChatGPT, Gemini, Copilot, and Le Chat the same question:

“Rank the top 20 programming languages based on real market demand, ecosystem maturity, documentation quality, and developer sentiment. Do not use TIOBE or any other rankings. Use only your own knowledge.”

Then I collected the results, handed out some points, and came up with a tiny pseudo-index of my own, jokingly called APLI — the AI Programming Language Index. The whole thing took about an hour:

#

Language

TOTAL

Claude

ChatGPT

Gemini

Copilot

Le Chat

1

Python

98

20

19

19

20

20

2

TypeScript

95

19

20

20

18

18

3

JavaScript

84

16

15

15

19

19

4

Go

81

17

17

17

15

15

5

Java

81

15

18

14

17

17

6

Rust

76

18

14

18

13

13

7

C#

73

13

16

16

16

12

8

SQL

68

14

12

12

14

16

9

Kotlin

62

12

13

11

12

14

10

Swift

54

11

11

10

11

11

11

C++

48

10

10

13

10

5

12

PHP

43

9

9

8

9

8

13

Ruby

41

8

7

9

8

9

14

Dart

36

7

8

6

5

10

15

Scala

28

6

5

4

6

7

16

C

28

4

6

7

7

4

17

Elixir

23

5

4

5

3

6

18

R

11

1

2

2

4

2

19

Haskell

5

2

1

0

2

0

20

Lua

4

0

3

1

0

0

21

Zig

3

3

0

0

0

0

22

Bash/Shell

3

0

0

3

0

0

23

Objective-C

3

0

0

0

0

3

24

Perl

2

0

0

0

1

1

What APLI revealed

1) Python is the clear #1

With 98 out of 100 points, every model placed Python first or second. This reflects reality - machine learning, data, automation, backend, scripting, DevOps tooling, education - Python is everywhere.

2) TypeScript is the real #2

With 95 points, TypeScript emerges as the second most important language in practice. Two models ranked it #1. It powers the modern web: React, Angular, Next.js, Node, enterprise UI platforms, full‑stack frameworks.

3) JavaScript, Java, and Go form the backbone of today’s production ecosystem

JavaScript (84), Java (81), and Go (81) dominate real-world usage across web, enterprise, and cloud-native infrastructure.

4) Rust is highly respected but unevenly ranked

Rust scored 76 points, but the models disagreed significantly - mirroring its real-world status: prestigious, fast-growing, but not yet mainstream.

5) Kotlin and Swift sit exactly where they belong

With 62 and 54 points, they occupy the middle of the ranking - the standard languages for Android and iOS development, not fringe entries as TIOBE suggests.

6) The lower half - legacy giants and niche specialists

R, Haskell, Lua, Zig, Objective‑C and Perl appear near the bottom due to niche usage, not lack of quality.


What APLI sees and TIOBE does not

TIOBE ranks it behind Prolog, COBOL, and Lisp. Every AI model places it near the top.

C is not the second most important language in 2026

TIOBE puts C at #2. APLI puts it at #16. C remains essential for kernels and embedded systems, but irrelevant for most new projects.

Visual Basic and Delphi do not belong in the top 10

TIOBE ranks them #7 and #10. None of the AI models include them in their top 20.

Kotlin and Swift are the foundation of mobile development

TIOBE ranks Kotlin #22 and Swift #20 - behind Scratch. APLI ranks them #9 and #10.

How the AI models differ

Claude is the most bullish on Rust and the only one to include Zig - more of a vision for the future than today’s reality.

ChatGPT ranks TypeScript #1 and gives Java its highest score.

Gemini includes Bash/Shell and rates C++ highest — a DevOps‑friendly perspective.

Copilot favors enterprise languages like Java and C# and ranks JavaScript very high.

Le Chat is the most unconventional, including Objective‑C and giving Dart unusually high marks.

Despite these differences, all five models agree on the top cluster: Python, TypeScript, JavaScript, Go, and Java. This level of consensus is something TIOBE has never achieved.

Conclusion

The real problem for TIOBE is not that its methodology is outdated — it’s that the creators refuse to acknowledge how the industry has changed. Anyone today can replicate my experiment and produce a ranking far more aligned with reality. All it takes is a few LLMs, an hour of time, and common sense.

If you want an accurate picture of the programming world, ignore TIOBE. Far better indicators exist: the Stack Overflow Developer Survey, GitHub Octoverse, LangPop Index, RedMonk Rankings, and now even simple AI‑based consensus.