The original IAT paper (Greenwald, McGhee & Schwartz, 1998) introduced response labels like ‘pleasant’ and ‘unpleasant’ assuming these unambiguously measured association strength and implicit attitudes.

Evidence

https://faculty.washington.edu/agg/FullText/GM&S1998/FullText.htm

XREF: connects to: To what extent does environmental visual adaptation (e.g., living in extreme high- or low-chroma landscapes) shift an individual’s implicit cognitive associations between achromatic light and moral valence?; Han, Olson, and Fazio established that the attribute response labels typically employed in the IAT contribute to the malleability of the IAT, clouding inferences about attitude change. — https://psycnet.apa.org/record/2010-03210-006; Implicit associations reflected in the IAT are influenced by societal messages, life experiences, context, and other environmental factors. — About the IAT – Project Implicit (+1 more)

Han, Olson, and Fazio established that the attribute response labels typically employed in the IAT contribute to the malleability of the IAT, clouding inferences about attitude change. (similarity 0.87)
IAT scores can be changed by the ambiguous attribute response labels of an unrelated preceding mindset task, so observed change may reflect the measure rather than the attitude. (similarity 0.87)
The IAT as conventionally implemented is open to multiple interpretations and can provide contextually malleable measurement outcomes, distinguishing the malleability of the measure from the… (similarity 0.84)
Implicit associations reflected in the IAT are influenced by societal messages, life experiences, context, and other environmental factors. (similarity 0.84)
IAT scores fluctuate with one’s current context and are better measures of biases held by places than biases held within minds. (similarity 0.83)
IATs lack psychometric evidence of construct validity; at best 36% of variance in a single IAT may reflect implicit attitudes, and the proportion could be 0. (similarity 0.80)