As HABO said in comment, string_split() on ; then split on : using charindex() , left() & right() .

You can filter for the required in where clause

select t.[date], t.[user], a.[action], a.[item]
from   yourtable t
       cross apply string_split(actions, ';') s
       cross apply
       (
           select [action] = left(s.value, charindex(':', s.value) - 1),
                  [item]   = right(s.value, len(s.value) - charindex(':', s.value))
       ) a
where  t.[user]       = 'dave'
and    trim(s.value) <> 'changed'

Please try the following solution based on STRING_SPLIT() and PARSENAME() functions.

Assumption is that the actions column values don't contain dots.

SQL

-- DDL and sample data population, start
DECLARE @tbl TABLE (_date DATE PRIMARY KEY, [user] VARCHAR(30), actions VARCHAR(100));
INSERT @tbl (_date, [user], actions) VALUES
('2023-01-01', 'dave', 'changed; added:apple'),
('2023-01-02', 'gail', 'changed; removed:apple'),
('2023-01-03', 'mick', 'changed; added:apple; removed:banana; added:cherry; removed:durian'),
('2023-01-04', 'dave', 'changed; removed:banana; added:cherry');
-- DDL and sample data population, end

SELECT t._date, t.[user]
    , PARSENAME(t2.tokens, 2) AS [action]
    , PARSENAME(t2.tokens, 1) AS item
FROM @tbl AS t
CROSS APPLY STRING_SPLIT(actions, ';') AS t1
CROSS APPLY (SELECT REPLACE(TRIM(t1.value), ':', '.')) AS t2(tokens)
WHERE t1.value LIKE '%:%'
    AND [user] = 'dave';

Output

perhaps something like this

WITH SplitActions AS (
  SELECT
    [date],
    [user],
    TRIM(value) AS action_item
  FROM elg_surveyBase
  CROSS APPLY STRING_SPLIT(actions, ';')
),
ActionDetails AS (
  SELECT
    [date],
    [user],
    TRIM(SUBSTRING(action_item, 1, CHARINDEX(':', action_item) - 1)) AS action,
    TRIM(SUBSTRING(action_item, CHARINDEX(':', action_item) + 1, LEN(action_item))) AS item
  FROM SplitActions
)
SELECT
  [date],
  [user],
  action,
  item
FROM ActionDetails
WHERE [user] = 'dave'
  AND [date] BETWEEN '2023-01-01' AND '2023-01-31';