Wikipedia:Analyzing sample size of 1001 as 97 percent

This essay refers to the common method of the polling or sampling of data with sample size as 1,001 cases, to give a margin of error of ±3.1% (97% confidence). As often seen in political polls, when the size of a survey reaches 1,001 members, then the results for a wide variety of questions, or user preferences (etc.), is mathematically accurate to about a 97% confidence level. For example, in a sample of 1,001 random responses, if 90% of cases refer to e-mail spelled as "email" and only 10% use the standard hyphenated spelling (as "e-mail"), then there is a 97% probability that all cases would have 90% using the word "email".

Results in Google Search
When searching for phrases or words in Google Search, the last page of results often covers less than 1,000 of the total webpages which match a search. However, the closer the count, of search-result pages, to being 1,001 webpages, then the closer to having a 97% predictive sample (3% margin of error) for the trends in the total of all webpages related to that search.