OP here's a suggestion to get accurate data.

Use blind targets, and have multiple viewers get their "impressions" from the same blind targets.

1) By doing this, you don't get "impressions" based off of previous knowledge.

2) If multiple viewers are getting different "impressions" from the same blind target, more than likely their "impressions" are merely their own imaginations.

Is your method like this?

For example, if I said, what's your impression of Atlantis, more than likely since I already said "Atlantis" you already have preconceived ideas based on previous knowledge and research.

Therefore you want to use blind targets.

And of course if the viewers are all getting different impressions from the same blind target, it's more than likely their own imagination.