Summarising from McHugh (2013), the $\chi^2$ test has six conditions specific to the test. Two relevent assumptions here are:
- "Each subject may contribute data to one and only one cell in the test"
- "The value of the cell expecteds should be 5 or more in at least 80% of the cells, and no cell should have an expected of less than one"
The first assumption is usually used to caution about applying to panel data, but would also apply here, depending on how you tabulated your 21 items (assuming each subject could have multiple 'items'). The second assumption is self explanatory, and a practical example is given here.
A reproducible example / clearer description would help with suggestions for alternatives. However, common options are Fisher's exact test, and (according to McHugh) the Maximum likelihood ratio $\chi^2$ test. If you have matched pairs, then McNemar's test.
NB. Fisher's exact test assumes fixed totals. Read more here.