Dictionaries decide what counts
Whether a word is valid depends entirely on which word list is being used, and the major ones genuinely differ. Competitive Scrabble in North America uses the NASPA Word List; most of the rest of the world uses Collins Scrabble Words, which is substantially larger and includes many words the North American list omits.
Collins contains roughly 280,000 words against NASPA's approximately 190,000, so a word can be perfectly valid in a tournament in London and invalid in one in Boston. Neither list matches a standard dictionary — both include obscure and archaic entries, and both exclude proper nouns, abbreviations and hyphenated forms regardless of how common they are.
Word lists also change. Both are revised periodically, adding new entries and occasionally removing offensive ones, which has been contentious. Any word finder is only as current as the list behind it, so a disputed word is worth checking against the official list for the game you are actually playing.
Playing the board rather than the word
The instinct to play the longest word available is usually wrong. Scoring depends on letter values and premium squares far more than on length, and a short word placed across a triple-letter square with a high-value tile frequently beats a long word played on plain squares.
Two-letter words are the highest-leverage thing to learn, and there are roughly 100 to 120 valid ones depending on the list. They allow parallel plays where a tile forms words in both directions simultaneously, scoring the same tile several times, and they let you use awkward tiles without opening the board.
Rack management matters over a game rather than a turn. Keeping a workable balance of vowels and consonants, and avoiding a rack of all one or the other, prevents the turns where nothing can be played. Holding a blank or an S for a high-scoring opportunity is often worth more than spending it now, since both are disproportionately valuable.
Using a solver honestly
In casual play among friends, using a solver mid-game without agreement is straightforwardly cheating, and it is worth being explicit about that. Online platforms treat it the same way, and most have detection based on play patterns — consistently finding optimal plays is statistically obvious.
The legitimate uses are learning and analysis. Reviewing a finished game to see what was available teaches board patterns far faster than playing alone, and studying two-letter words, high-probability bingo stems and common hooks is standard preparation for competitive players. Solving a crossword clue or unscrambling an anagram for a puzzle is its own activity.
For word games generally, remember that a solver optimises for score and not for strategy — the highest-scoring play can open a triple-word lane for an opponent. That trade-off is the part of the game a solver cannot evaluate for you.