How do you completely de-personalize Google results? - Gabriel Weinberg's Blog | Hacker News Reader

How do you completely de-personalize Google results? - Gabriel Weinberg's Blog

gabrielweinberg.com

78 pointsricg18 comments

Is it an oddly academic question?

If every search is personalised then what does a personalisation-free search tell you? It's possible that it will no be closer to what most people see as very few people jump through the hoops you're jumping through.

It might even be less like the majority of search results.

I think the crux is objectivity vs subjectivity. You'd like to be able to get an objective set of results to a query.

andybak's point is a good one though, if you get an objective SERP then that's not going to tell you anything very useful as all your potential customers are getting a personalised version.

Looks like an opening: have a group of users sorted across demographics that you can automatically send a query too. The system returns the SERPs from each user and compiles a master SERP showing position based on demographics of the user? So you can, say, run a search for "search engine" across 1000 user machines and then look at average placement (with variance) for Windows 7 users aged 30+ in USA, or whatever.

You could probably do a similar thing with Tor (or a botnet) but the relationship between the SERP data and the demographic of the user is probably the most pertinent information and that would be lost then.

The snag is that running searches on other people's profile will alter that profile. The results will be less accurate over time.

Not to mention they would get some very strange things popping in Google Now! Oh, it's a tangled web we weave...

This is a very good point. Ideally, one should be able to explicitly override personalization-related input as much as possible, so that one can freely explore what kind of results other groups of people get. I suspect that Google are reluctant to provide such an interface as it might make it easier to game the system.

Ideally, there should be a feature that lets you see the personalized (that is, real) search results of others. Maybe divided into target groups ("programmers", "male users under 20", etc.). Of course, everyone's real search results will be slightly different, incorporating "+1"s of Google+ friends for example.

But, I assume, the bulk of the search ranks will be the same for people in the same (advertising) target group. (I have nothing to back up that assumption - it just would be the most sensible thing to do, I think.)

I'm too skeptical.

This looks like the sort of post you'd write if you wanted lots of people to search for "search engine", normally and with Tor, and click on your result; wide-scale social engineering. Like a pick-pocket putting up a "beware of pick-pocketers" sign.

Would be interested to see if there is a surge in the stats for visits from Google SERPs (over Tor) following this post.

Yes, you are.

And if I were skeptical, i'd say its the sort of post you write if you were an advisor/investor of a competing SE that features user privacy.

I'm the OP. The traffic from that query is negligible for us. My proximate motivation was to answer the question posed and the specific questions at the bottom of the post. My root motivation is to better understand the implications of the Filter Bubble.

Oh verily. It's more a mental condition on my part I think ... I was just about to check the link and then #bam# ...

Could you explain how the hypothetical pickpocketing scheme works?

Supposedly a pick pocket places a sign (or locates near one). People seeing a warning check their pockets for their wallets/valuables thus showing the pick-pockets where to find the valuables.

ctrl+shift+n (incognito window).

At least gets around the Google login / plus / social info. Still going to have something to do with your location, but I guess there are no totally objective results if you consider that part of the search personalising.

I think it also has to do with where your IP is geolocated. I was testing search results for my site, Scribophile, which is a writing community. When searched from a fresh US IP, it ranked 3rd in the results. When searched from a fresh EU IP, it ranks 4th, with a UK-only writing website taking the 3rd spot.

Issue automated searches for the term you're interested in from VPS/cloud servers in multiple geographic locations at random times during the day over the course of a few weeks, shuffling your user agents between the current popular browsers. Take some average of the SERP rankings.

An average ranking would be no different than a personalized one in terms of accurracy

reminds me of this artwork, personal depersonalization system, http://bengrosser.com/projects/personal-depersonalization-sy... . he wrote code to clutter his google history with junk by continuously searching for dictionary words, changing location, etc.

Not only does personalization change things around, but time of day and IP location seem to affect things as well. I think.

So I have a site that was ranking around #18. On weekends, it drops down to around #30. Then at some times of day it might rise as high as #16.

It's like Google is also creating a time/location model of content.

All this stuff with Google is all so much reading the tea leaves, though. Very frustrating. If your marketing consisted of having a few signs around town you could easily go out and count the traffic and observe patterns. With Google, it's much more random and complex.

Reply on news.ycombinator.com