[response set analysis] respondents unaware of negative formulated items

I’m wondering what to do when negative items does not break response set.
In other words, it remain unclear if respondent was where of the negative question. This because these questions weren’t properly market with bold lettering.
So in this scenario how to look for response set?

I came up but the following solution, but still wondering if correct.
Looking forward to our feedback

Example; 10 item measure for Brand LOVE (2 negative formulate items:4&8) via 7 point likkert.
The negative questions where recoded: 1=7, 2=6 etc.

BL1: [chosen brand] is a wonderful brand.
BL2: [chosen brand] makes me feel good.
BL3: [chosen brand] is totally awesome.
BL4_n: [chosen brand] is NOT a wonderful brand
BL5: [chosen brand] makes me very happy.
BL6: [chosen brand] is a pure delight.
BL7: I love [chosen brand]!
BL8_n: I have NEUTURAL feelings about [chosen brand]

To assess the existence of response set a new variable was computed (resp_BL) which grouped individual items scores of Brand Love into one variable, as in the example below:


After reviewing the scores, I had my doubts if respondents where aware of the negative formulated questions, so I did a paired t test to test for significant difference. Because negative items where recoded there should not be a significant difference between paired items.

H0: μ1 – μ2 = 0
H1: μ1 –μ2 ≠ 0
α: .05
BL1-BL4_n: sig: .000
BL7-BL8_n: .000

However there was a significant difference :shakehead

So BL4_n & BL_n where excluded from response set analyse.
resp_BL has now 6 scores:


Cases with SD=0 for a given group of items (i.e. BL:1-7) where indicated as response set. Because it is unlikely that a respondent will answer 8 questions in a row with the same answer (444444).
When response was found: given items block of that particular respondent was deleted. Because other item blocks had no indication of response set and therefore could contain usable data.