Never, ever make a free entry question. You'll wind up with "balls", "suck it", etc. all the time. Predefine 3-4 set ranges, and let them perform a single click to answer the question. Much more accurate results.

more accurate results, or more plausible-looking random results?

Two things:

1. As the researcher you can set a range of validity for open-ended numeric if you know it ahead of time. We recommend doing this.

2. The system automatically detects outliers in all cases and puts them in a separate bucket to reduce noise.

An example of open-ended numeric for pricing is here:

https://www.google.com/insights/consumersurveys/view?survey=...

From what I can tell #2 doesn't seem to be working in my case.

You can see my survey at https://www.google.com/insights/consumersurveys/view?survey=...