During a class, I googled things with the following rules:

1) Pick an initial term

2) Use the first full word (used as a word) from the first organic result’s snippet as the next term.

Here’s what happened:

a -> the -> Mar -> a -> [3-loop]

google -> enables -> to -> terrell -> the -> [3-loop]

query -> in -> government -> home -> world’s -> Inc. -> Mar -> [3-loop]

search -> search -> [1-loop]

random -> offers -> find -> search -> [1-loop]

ectoplasm -> ectoplasm -> [1-loop]

inclement -> etymology -> history -> the -> [3-loop]

airplane -> directed -> directed [1-loop]

aeronautical -> aeronautics -> aeronautics -> [1-loop]

happily -> in -> [4+3-loop]

several -> etymology -> [2+3-loop]

abstracted -> withdrawn -> removed -> distant -> etymology -> [2+3-loop]

apply -> oct -> optical -> find -> search -> [1-loop]

comic -> has -> words -> the -> [3-loop]

teleport -> important -> precedence -> pronunciation -> a -> [3-loop]

arranged -> over -> purchase -> purchase -> [1-loop]

In these few experiments, the longest chain achieved was 10 unique words before a loop, with ‘abstracted’. I suppose one way of trying to do better would be to look for pages whose results snippet begins with that word, find a term that gets that as the first page, and keep working backwards. Maybe later.

Observations: Lots of Wikipedia pages start with “The”. Lots of Wikipedia pages for nouns start with that noun. Lots of snippets start with a recent date, this explains why ‘Mar’ shows up so much. If the word is too obscure, the results will all be definitions, which start with common things.