Optimizing abstract text

Trying to understand some observations

[sublime1]

Can anyone point me to an existing discussions here that explain how Google selects content for its abstracts? I have read the FAQ here and understand that you should get the desired keywords higher in the page, perhaps bold or other emphasis.

But I am interested in how the abstracting algo determines where to start, and how long it can be. In some cases it just seems to take whatever text there is in little chunks, in others, it seems to take the first "sentence" that contains the terms and is short enough. Does word order or proximity matter? And is there any relationship between the abstract chosen and the SERP -- said another way, is a "good" abstract a predictor of a good SERP?

If there are existing discussions of this, any pointers would be welcomed.

Thanks in advance.

[vitaplease]

Just a trivial observation:

With a multi word search query of words occuring well spread out throughout the bodytext of a webpage:

AFAIK, Google's snippets only show the first three (non-common-words) of the search query words - highlighted in the same sequence of occurence as in the body text - in three snipped sentences of approx. 6 to 8 words.

So check your logs for multi-word search query Search Engine referrals.
Ideally could could alter the bodytext wordings to make the snippets more attractive - which would probably mean higher rankings anyway because of better proximity.

[killroy]

I would think that a concise short explanatory sentence startign with your main target keywords would be picked up as abstract.

Like

big red widgets are for sale here.

or
We sell big red widgets.

I bet those would sho in an abstract for a search for "big red widgets"

SN

[sublime1]

Thanks killroy. I think your observations are correct -- we have been able to see this behavior in our site as well as others.

I'm still wondering if people have figured out the twists and turns of the method. How many words, how do they know to find a sentence, etc.

If no one knows, I'll try to come up with a straw-man, and see if others can prove or disprove it.