Breaking url-bar search

Gavin Sharp gavin at gavinsharp.com
Fri Aug 16 18:14:43 UTC 2013


 On Fri, Aug 16, 2013 at 9:13 AM, William Pietri <william at scissor.com> wrote:
> Hi, Gavin. Sorry if this is coming across as hostile; I'm not trying to pick
> on you.

No worries, William, I appreciate that you've been willing to discuss
this quite even-handedly. Being an open project, we often have to deal
with much less productive feedback and questioning :)

> I agree that when intuition is what you've got, that's what you should go
> with. But in this case, by "data" I mean what is traditionally meant by the
> word: reliable numbers.
>
> So in this case, it sounds like there was no data available to you, so you
> made an expert judgment call. Is that fair to say?

I don't think "no data" is fair. We had the heat map study that I
linked to previously, and telemetry data that indicates the percentage
of Telemetry users which had a modified keyword.URL preference. These
are "reliable numbers", hard data, and they informed our judgement. We
also have some hard data that hasn't been published yet, from FHR,
that suggests the vast majority of searches are performed with the
default search engine (this is in aggregate, not broken down per-user,
so we do not know e.g. how many users use one search engine in the
search bar and another in the location bar, which as I understand it
is what you mean by "the feature" in this post and others.).
Obviously, we should do a better job of publishing the data and our
conclusions derived from it, so you don't have to take my word on it.
We are working on that.

> So from that, it seems fair to say that the number of affected users is
> thousands to millions, but tens of millions is out of bounds, yes?

I don't think these back-of-the-envelope made up guesses are
particularly useful or likely to be productive. I doubt very much that
more than thousands of users are negatively impacted (and that's on
the order of 0.002% of our user base, by conservative estimates).

> Could you help me understand how the first benefit is different than the
> second? I'm not seeing any lasting user benefit to point 1 on its own;
> hijackers will presumably quickly retarget.

There are several classes of "hijackers". The blatantly malicious ones
will adapt no matter what we do; we can try to make their life harder,
but that's an arms race with no clear end. The other class of
"hijackers" (which may be too pejorative a term) are legitimate
companies with reputations to keep that will not stoop so low as to
try to forcibly prevent user control of searches, but have an
incentive to "nudge" users towards changing their search engine -
sometimes too aggressively for our taste. And sometimes these
non-malicious actors just run up against limitations of our code - for
example, prior to these changes the easiest way to change location bar
searches (set the keyword.URL pref) also ended up being the least
user-friendly, and so the "lazy" way to do this ended up being
harmful. Eliminating the lazy, user-harmful way of doing this forces
these actors to put in a bit more effort, which ensures a better user
experience.

Gavin



More information about the firefox-dev mailing list