To put it simply- Yes. Data is data. One can either present it either as descriptive or carry out an inferential statistics. For example, if you have two variables- Age and Income, you can catergorise them both (Ex: Age 1- 10, 11-20,etc) and present the data as a form of percentage analysis (descriptive statistics) and the same two variables can also be used to carry out a correlation test (inferential statistics) to find out if there is any significant correlation between the two.

Yes, it is possible to use inferential statistics to analyze descriptive survey data. While descriptive statistics summarize the characteristics of data that have been collected-for example, means and percentages-inferential statistics extends this because it allows the researcher to make generalizations or predictions about a population from sample data. For example, you can use techniques such as hypothesis testing, confidence intervals, or regression analysis to infer trends or relationships in the broader population from your survey data.

Inferential statistics require an hypothesis about the relationship between two or more variables, while purely descriptive statistics report one variable at a time. There is no inherent difference between whether a variable is descriptive or inferential, for example education could be reported in descriptive percentages or used as an independent or dependent variable in an inferential test.

Temesgen Tora -

I think you mean can we obtain inference from a convenience sample.  If that is what you meant, that you want to use descriptive statistics on a set of data to infer to a population when no randomized probability-of-selection scheme was used, then the traditional answer has been "No."  However, much work has been done on that in recent years.  Essentially, any chance to infer to a population relies on the availability of the right auxiliary data and that it should be of good data quality.  There are two main approaches which can also be combined.  Among those who have worked in this area, and especially on the two approaches to be mentioned, note Richard Valliant, who can be found in the bibliography of the paper linked below. Those approaches, which rely on covariates are (1) a model approach and (2) one which uses pseudo-weights which also makes use of another probability sample to help assign pseudo-weights. You may find other work of interest in this area, but covariate data are needed.

I have also made use of prediction (ratio models) with a special kind of nonprobability sampling: cutoff sampling.  The following linked paper is on that. Please see the bibliography for work by Valliant and others which you may find.

https://www.researchgate.net/publication/383976649_Cut-off_Cutoff_Quasi-Cutoff_and_Other_Nonprobability_Sampling_with_Inference

Cheers - Jim