Most good ideas (whether they're ideas for narrative structure, a particular twist in the argument, or a broader topic) come into our minds as hunches: small fragments of a larger idea, hints and intimations. Many of these ideas sit around for months or years before they coalesce into something useful, often by colliding with another hunch. (I wrote a chapter about this phenomenon in my last book, Where Good Ideas Come From.)
The problem with hunches is that it's incredibly easy to forget them, precisely because they're not fully-baked ideas. You're reading an article, and a little spark of an idea pops into your head, but by the time you've finished the article, you're checking your email, or responding to some urgent request from your colleague, and the next thing you know, you've forgotten the hunch for good. And even the ones that you do manage to retain often don't turn out to be useful to you for months or years, which gives you countless opportunities to lose track of them.
This is why for the past eight years or so I've been maintaining a single document where I keep all my hunches: ideas for articles, speeches, software features, startups, ways of framing a chapter I know I'm going to write, even whole books. I now keep it as a Google document so I can update it from wherever I happen to be. There's no organizing principle to it, no taxonomy--just a chronological list of semi-random ideas that I've managed to capture before I forgot them. I call it the spark file.
Now, the spark file itself is not all that unusual: that's why Moleskins or Evernote are so useful to so many people. But the key habit that I've tried to cultivate is this: every three or four months, I go back and re-read the entire spark file. And it's not an inconsequential document: it's almost fifty pages of hunches at this point, the length of several book chapters. But what happens when I re-read the document that I end up seeing new connections that hadn't occurred to me the first (or fifth) time around: the idea I had in 2008 that made almost no sense in 2008, but that turns out to be incredibly useful in 2012, because something has changed in the external world, or because some other idea has supplied the missing piece that turns the hunch into something actionable. Sure, I end up reading over many hunches that never went anywhere, but there are almost always little sparks that I'd forgotten that suddenly seem more promising. And it's always encouraging to see the hunches that turned into fully-realized projects or even entire books.
I keep ideas in notebooks and a notetaking app but probably don't go and review the old stuff as much as I should. There's often good stuff buried in the weeds at the junkyard.