First of all, conducting CFA after EFA on the same set of data doesn't make a lot of sense. See my Youtube video below on why the "EFA then CFA" approach isn't recommended:

https://youtu.be/Qzdr1xaoApE

Second, the small loadings can indicate multiple things, including a combination of the issues listed below:

- model misfit (a CFI/TLI of .90 or greater does not necessarily mean the model fits well)

- lack of validity (the items do not measure the intended attribute/latent variable)

- item heterogeneity (some items measure separate factors or contain a lot of specific item variance or method variance)

- item unreliability (items with small loadings may contain a lot of random measurement error variance)

Depending on these factors, you may or may not decide to "abandon" the scale (or at least certain items). See also my Youtube video below on the topic of "small loadings in CFA":

https://youtu.be/YaQCrJKe37Y

First, I agree that conducting CFA following EFA on the same data makes no sense. If you adapted an existing questionnaire with a known structure, you should proceed with CFA. Second, the goodness of fit indices yielded by CFA are within the acceptable range, the factor loading is .44 (the other loadings are pretty high), and the item is conceptually important for your measurement purpose, I wonder why you think of discarding the item.

To me, the information you provided is sufficient to conclude the model is valid (evidence of construct validity)