Serialization of XML Data
When casting the xml data type explicitly or implicitly to a SQL string or binary type, the content of the xml data type will be serialized according to the rules outlined in this topic.
Serialization Encoding
If the SQL target type is VARBINARY, the result is serialized in UTF-16 with a UTF-16-byte order mark in front, but without an XML declaration. If the target type is too small, an error is raised.
For example:
select CAST(CAST(N'<Δ/>' as XML) as VARBINARY(MAX))
This is the result:
0xFFFE3C0094032F003E00
If the SQL target type is NVARCHAR or NCHAR, the result is serialized in UTF-16 without the byte order mark in front and without an XML declaration. If the target type is too small, an error is raised.
For example:
select CAST(CAST(N'<Δ/>' as XML) as NVARCHAR(MAX))
This is the result:
<Δ/>
If the SQL target type is VARCHAR or NCHAR, the result is serialized in the encoding that corresponds to the database's collation code page without a byte order mark or XML declaration. If the target type is too small or the value cannot be mapped to the target collation code page, an error is raised.
For example:
select CAST(CAST(N'<Δ/>' as XML) as VARCHAR(MAX))
This may result in an error, if the current collation's code page cannot represent the Unicode character Δ, or it will represent it in the specific encoding.
When returning XML results to the client side, the data will be sent in UTF-16 encoding. The client-side provider will then expose the data according to its API rules.
Serialization of the XML Structures
The content of an xml data type is serialized in the usual way. Specifically, element nodes are mapped to element markup, and text nodes are mapped to text content. However, the circumstances under which characters are entitized and how typed atomic values are serialized are described in the following sections.
Entitization of XML Characters During Serialization
Every serialized XML structure should be capable of being reparsed. Therefore, some characters have to be serialized in an entitized way to preserve the round-trip capability of the characters through the XML parser's normalization phase . However, some characters have to be entitized so that the document is well-formed and, therefore, able to be parsed. Following are the entitization rules that apply during serialization:
The characters &, <, and > are always entitized to &, <, and > respectively, if they occur inside an attribute value or element content.
Because SQL Server uses a quotation mark (U+0022) for enclosing attribute values, the quotation mark in attribute values is entitized as ".
A surrogate pair is entitized as a single numeric character reference, when casting on the server only. For example the surrogate pair U+D800 U+DF00 is entitized to the numeric character reference 𐌀.
To protect a TAB (U+0009) and a linefeed (LF, U+000A) from being normalized during parsing, they are entitized to their numeric character references 	 and 
 respectively, inside attribute values.
To prevent a carriage return (CR, U+000D) from being normalized during parsing, it is entitized to its numeric character reference, 
 inside both attribute values and element content.
To protect text nodes that only contain white space, one of the white-space characters, generally the last one, is entitized as its numeric character reference. In this way, reparsing preserves the white-space text node, regardless of the setting of the white-space handling during parsing.
For example:
declare @u NVARCHAR(50)
set @u = N'<a a="
	'+NCHAR(0xD800)+NCHAR(0xDF00)+N'>"> '+NCHAR(0xA)+N'</a>'
select CAST(CONVERT(XML,@u,1) as NVARCHAR(50))
This is the result:
<a a="
	𐌀>"> 
</a>
If you do not want to apply the last white-space protection rule, you can use the explicit CONVERT option 1 when casting from xml to a string or binary type. For example, to avoid entitization, you can do the following:
select CONVERT(NVARCHAR(50), CONVERT(XML, '<a> </a>', 1), 1)
Note that, the query() Method (xml Data Type) results in an xml data type instance. Therefore, any result of the query() method that is cast to a string or binary type is entitized according to the previously described rules. If you want to obtain the string values that are not entitized, you should use the value() Method (xml Data Type) instead. Following is an example of using the query() method:
declare @x xml
set @x = N'<a>This example contains an entitized char: <.</a>'
select @x.query('/a/text()')
This is the result:
This example contains an entitized char: <.
Following is an example of using the value() method:
select @x.value('(/a/text())[1]', 'nvarchar(100)')
This is the result:
This example contains an entitized char: <.
Serializing a Typed xml Data Type
A typed xml data type instance contains values that are typed according to their XML schema types. These values are serialized according to their XML schema type in the same format as the XQuery cast to xs:string produces. For more information, see Type Casting Rules in XQuery.
For example, the xs:double value 1.34e1 is serialized to 13.4 as shown in the following example:
declare @x xml
set @x =''
select CAST(@x.query('1.34e1') as nvarchar(50))
This returns the string value 13.4.