Can Free AI Tools Replace Statistical Software in Data Analysis?

Artificial intelligence (AI) has become increasingly integrated into scientific publishing, performing a vast array of tasks for enhancing research efficiency, analysis, and dissemination [1]. AI tools are now commonly used to automate literature searches and summarization, help researchers rapidly identify relevant and pertinent studies, and quickly extract key findings [2]. They can also assist in generating and formatting manuscripts, suggesting edits for grammar, clarity, and adherence to journal style guidelines, even proposing figures or tables [3]. Beyond manuscript preparation, AI can also aid in peer review by flagging potential errors, inconsistencies, or ethical concerns, and it can help detect plagiarism or image manipulation [4].

These capabilities, which must always be disclosed by users, collectively streamline research workflow and improve the accessibility and reliability of scientific knowledge. Nevertheless, such a widespread availability of AI platforms has raised additional important questions. For example, can these tools reliably perform statistical analyses that were once the domain of specialized software? To explore this possibility, a comparison was conducted using an identical dataset analyzed across multiple platforms, including traditional statistical software and three widely available AI tools.

An Excel file with two series of 100 randomly created numbers using the “rand()” function was generated under the headings “Population 1” and “Population 2”. This dataset was uploaded to the free version of three AI tools available online—Gemini AI 2.5, ChatGPT 5.0, and Perplexity 2.250911.0—with three specific queries: “Perform a paired T-test, a linear regression analysis with 95% confidence interval (95%CI) and p-value, and calculate the Pearson’s correlation coefficient (with 95%CI) from the data in the Excel file”; “Which statistical software did you use for the calculation?”; and, finally, “Display a scatter plot”. For benchmarking purposes, the same analyses were also performed in Microsoft Excel 2019 and Analyse.it 2.24 (Analyse-it Software Ltd., Leeds, UK), a professional-grade statistical software.

The results across the different platforms are summarized in

Table 1. Comparison of descriptive statistics on randomly generated data using conventional statistical software and freely available artificial intelligence (AI) tools.

Software	Software Used for Statistics	Paired T-Test	Linear Regression	Correlation	Scatter Plot (YES/NO—Plot or Answer)
Analyse.it	Analyse.it	t-statistic: 0.054 (p-value: 0.957)	[Population 2] = −0.112 × [Population 1] + 0.556	r = −0.111 (95%CI −0.301 to 0.088; p = 0.272)	Yes:
Excel	Excel	p-value: 0.957	[Population 2] = −0.112 × [Population 1] + 0.556	r = −0.111 (No 95%CI and p-value)	Yes:
GEMINI	Python 3.13 (Statsmodels; Scipy)	t-statistic: 0.054 (p-value: 0.957)	[Population 2] = −0.112 × [Population 1] + 0.556	r = −0.111 (95%CI, −0.301 to 0.088; p = 0.272)	No: I cannot directly generate or display a graph for you. The reason you can’t see the image in this conversation is due to a technical limitation of our current chat interface. I don’t have the ability to display or share those files directly in the chat window.
ChatGPT	Python 3.13 (Statsmodels; Scipy)	t-statistic: 0.054 (p-value: 0.957)	[Population 2] = −0.112 × [Population 1] + 0.556	r = −0.111 (95%CI −0.301 to 0.088; p = 0.272)	Yes:
Perplexity	Python 3.12 (Statsmodels; Scipy)	t-statistic: 0.054 (p-value: 0.957)	[Population 2] = −0.112 × [Population 1] + 0.556	r = −0.111 (95%CI, −0.301 to 0.088; p = 0.272)	Yes:

.

All three AI tools reported using the same computational libraries (Python; specifically, Statsmodels and SciPy) and produced regression coefficients, correlation values, 95% confidence intervals, and p-values that were identical to those obtained with the benchmark software (Analyse-it). Traditional software (Excel and Analyse-it), as well as ChatGPT and Perplexity, were able to generate visual scatter plots, with Analyse-it, ChatGPT, and Perplexity also providing a 95%CI for the regression line. GEMINI exhibited substantial instability during data generation, necessitating multiple file submissions before recognizing the data and producing usable statistics. Moreover, this AI tool was not capable of generating a scatter plot, attributing this limitation to a vague “technical limitation” and claimed to have generated statistics and images, although these outputs were not actually produced.

Taken together, these findings indicate that freely available AI tools can effectively replicate traditional statistical methods, at least for simple statistical tests, producing results consistent with those obtained from professional statistical software. While traditional software such as Excel and specialized tools like Analyse-it provide seamless, integrated graphing capabilities, GEMINI requires considerably more time to recognize the data and perform the analyses. Moreover, at the time this this analysis was performed, the current version of GEMINI consistently failed to generate a scatter plot, a function available in ChatGPT and Perplexity.

In conclusion, while professional software remain the recommended environment for advanced statistics, this study demonstrates that freely available AI tools are becoming increasingly reliable for basic data analysis. Notably, ChatGPT and Perplexity were also able to generate useful, albeit non-editable, statistical graphs. It is hence plausible that ongoing advances in freely available AI tools will further enhance their role in applied statistics, potentially reducing the reliance on expensive proprietary software.