Googlits are not the best measure of correctness but...
"Likeliest" gets 900K googlits, while "most likely" gets 173 million. Seems to me that a ratio of almost 200 to one in favor of most likely provides a pretty good indication that the latter is the preferred method. And while I don't mind if other people use them, I can and do refrain from breaking a useful rule. And I teach my kids the same thing. Whether in writing or parol, the use of words and phrases that break with the rules tends to mark one as perhaps being a bit less than educated, whether we like it or not.
Somewhat along those lines, did you see the story in the Charlotte Observer the other day about admissions to major NC universities? UNC-Chapel Hill reported that they were admitting about 1/3 of all the applicants for freshman status. The article stated that they were turning away applicants who had a 4.0 average in high school and who had a combined score of 1300 on their SATs.
It is not outside the realm of possibility that admission decisions will end up being based heavily upon the new essay portion of the SATs, and I am determined that neither of my children will be graded down for saying likeliest instead of most likely.
The downside of teaching my kids to think and speak this way is that Theo, my 10-year-old, comes home and complains that his fourth-grade teacher does not use correct English and we have to keep reminding him that it is considered bad form to correct her English. And her spelling for that matter, come to think of it. One of the other teachers in the school wrote a note home last year about Sasha's missing library book, saying she could not make an acception (sic) in her case. ARGH!!!!
