I just needed to do something similar, but took a different approach. I copied the relevant string into a temp table where I can add columns to track an index through each occurrence on each row.
In my example, I'm looking for substrings " - " (space-dash-space) in product descriptions, with the intent of eventually chopping those apart to show as bullet points, and I'm analyzing the data like this to see how many "bullets" products typically have.
I suspect this is more efficient than repeatedly re-writing string values, but I haven't actually benchmarked.
SELECT
ccp.ProductID, p.ProductDescription, descrlen=LEN(p.ProductDescription),
bulletcnt=0, indx=0, lastmatchat=0
INTO #DescrBullets
FROM Private.CompositeCatalogProduct AS ccp WITH(NOLOCK)
INNER JOIN Products.Product AS p WITH(NOLOCK) ON p.ProductId = ccp.ProductID
WHERE ccp.CompositeCatalogID=53
DECLARE @rows INT = 1
WHILE @rows>0
BEGIN
-- find the next occurrence on each row that's still in play
UPDATE #DescrBullets
SET lastmatchat = PATINDEX('% - %',RIGHT(ProductDescription,descrlen-indx))
WHERE indx<descrlen
-- anywhere that a match was found, increment my counter, and move my
-- index "cursor" past it
UPDATE #DescrBullets
SET bulletcnt = bulletcnt + 1,
indx = indx + lastmatchat + 2
WHERE lastmatchat>0
SET @rows = @@ROWCOUNT
-- for all the ones that didn't have a match, advance indx past the end
-- so we don't need to reprocess on next iterations
UPDATE #DescrBullets
SET indx=descrlen
WHERE lastmatchat=0
RAISERROR('processing, %d products still have bullets', 0, 1, @rows) WITH NOWAIT
END
SELECT db.bulletcnt, occurs=COUNT(*)
FROM #DescrBullets AS db
GROUP BY db.bulletcnt
ORDER BY 1