Unfortunately, this issue is caused by the limitations of what a cell in Excel can handle as far as characters (32,700 chars) go. We have looked at having Xtraction do a size check on every cell being exported but the performance impact is quite troublesome and such a change would impact all Xtraction users. I would suggest putting in an enhancement request using the link below:
It's possible we could take another shot at this through different means in the future.
One suggestion would be to change your data model to only return the first 32k characters of the offending field. If you think people might want to search on the field, you might consider mapping the field twice: 1 mapping limits the output to 32k and is used for record lists, while the 2nd mapping returns the full string for use in filters. Both mappings would have the same Text (label), but different IDs in the data model. Let's say they are both called Description. When the user goes to filter on Description, the data model will use the mapping for the full field in the filter condition. When the user goes to include Description in a record list, the mapping returning the first 32k characters is used to display the Description. Then, when it is exported (to any format), only the first 32k is returned.
If you think that some folks will want to export to, say, Word or PDF and have the full contents of the field and are only trying to find a solution to exporting to Excel, you could have two mappings of the same field, both for use in Record Lists, 1 returning the full field and 1 returning the condensed version. They would need different labels, like "Description" and "Description (32k max)" and then the user would have to use whichever field they wanted.