A creates some content A, say a poem. He assigns an arbitrary number value to each letter of the alphabet. He then selects a random piece of content B somewhere on the web without analyzing it. The first letter of that content is then shifted forward by the value assigned to the first letter in the poem. Say the first word in the content A is "softly", and S is assigned the value of 73. So in the first word of content B, "The", 'T' become 96, and with mod 26 becomes "R".
After the poem, or piece of created content A ends, it repeats.
Since the subsequent letters of the randomly selected content B are unpredictable -- the next occurrence of any given letter would seem to be unpredictable, would it matter that the poem, content A, is repeating in a pattern?
I know that some letters would be linguistically more likely to be used, i.e. 'e', but since that letter in content B, although more likely to occur, would be matched with a random letter in content A, would this not produce a random string?
I'm thinking in the context of using the string that was outputted as a one-pad cipher.
While you use "true random" during the generation of your sequence, what you describe is essentially an encrpytion of $B$'s web content (so a text with certain natural langugae statistics) using $A$'s poem as a Vigenère cipher. For a sufficiently long messgae (i.e. enough repetitions of the poem) a decryption is possible.