That's the problem.
You can't teach phonemic awareness. A metastudy done by Krashen not many years back looked at over 300 research articles showing the results of teaching PA. The articles were mostly supportive, if not glowing in support, for teaching PA.
And they sucked so badly that after throwing them out for unconscionably miniscule sample sizes, archaic and obsolete linguistics, lack of controls and self-willed ignoring of confounds and feedbacks, two were left standing that would be acceptable as a graduate student linguistics paper. Most would be failed as an undergrad paper. One looked at American English and the other at Israeli Hebrew, IIRC. One of the papers found a slight negative effect from teaching phonemic awareness. One found a slight positive effect. Yes, 2 out of 300+ peer-reviewed published articles in education would pass muster. Less than 1%.
Most of the metholodogical errors included an obvious confound: The goal was to teach phonemic awareness (PA), the method was to teach phonics, the data collection was testing phonics, and the conclusions supported PA. As though phonics (which involves graphemes and mapping graphemes to phonemes or phones) was the same thing as developing phonemes (which are abstract mental representations). Note that for most of humanity's existence humans had, one expects, phonemic awareness. And most people lacked the ability to read. In fact, you can read just fine using non-phoneme-based writing systems.
The archaic linguistics looked entirely at phonemes as abstract units in complementary distribution, usually including confusing phonemes and graphemes. By the late '90s the complementary distribution business was clearly a crock as psycholinguistics and token-based phonology had their influences. The phoneme/grapheme issue was cleared up rather well by the 1920s. Graphemes =/= phonemes =/= phones. Linguistics 1.
The brain has to build phonemic awareness, but does so as a result of hearing a sufficiently large number of words a sufficiently large number of times. It needs to hear those phonemes over and over to specify their range of realization and recognize alternative realizations. And it only can do that hearing a sufficiently large number of tokens of each word, and a sufficiently large number of words. It does this automatically--it was unclear as of perhaps 4, 5 years ago that this could ever be directly taught in a way that affects the automatic parallel processing of language that actually occurs. Phonematicity emerges naturally from the distinctions that the brain is forced to make between separate words and how they match different meanings.
In fact, phonemic awareness, like morphological awareness, is a moving target. Phonemic targets can evolve over a person's lifetime, for instance. Some phonemes are "acquired" earlier than others and obviously that means some are acquired later. Bybee's token-based phonology was crucial in this.
Problem is, a lot of disadvantaged kids have a linguistic deficit (regrets to Labov). Not enough verbal input at home, either in number of different lexemes or the number of times they hear them all. Instead they hear a small set of words frequently, and hear fewer hours of speech on top of that. To "teach" phonemic awareness, a good program exposes kids to a larger vocabulary and makes sure they hear these words, properly used in context and in meaningful utterances, enough times.
A lot of PA programs teach phonics. Accidentally, along the way, a large number of disadvantaged kids are exposed to a sufficiently large vocabulary, repeated in context a sufficient number of times, to have those who want to believe in PA teaching have their bias confirmed. Unfortunately, by the time that happens some kids have already learned that they're failures at reading.
= new reply since forum marked as read