That depends on a lot of factors. Finally, you must counterweight two things:
How likely is it that a respondent will click through your questionnaire, giving meaningless responses?
What will that do to your data? Will this cause type I or type II errors?
If you have a lottery for the latest iPhone, you'll definitely need attention checks. If it's a 40-minutes questionnaire that nobody will do, except when they is highly motivated, ... well.
Using inverted scale items (this is not really an attention check) has another (important) use: It will remove a general bias of people to agree oder disagree. However, the data looks like if you has a worse Cronbach's Alpha in your scales.
There actually is not soo much literature on data quality and attention checks. What you'll probably find is the IMC (works quite good, but also identifies "good" respondents who are just a bit unattentive), and maybe you'll step over my working paper on response times. The truth is, that it becomes more complicated, when you check for data quality and attention. This reveals problems in survey data, that many researchers prefer not to know about. If you'd like to make it right, then take good care -- but, honestly, you won't do yourself a favor.