A few weeks ago, Google added a new AutoLink
feature to the Google Toolbar
3 beta which automatically recognises certain structures including street addresses,
ISBNs, package tracking numbers and automobile VINs; and converts them to useful
links for the user. This feature has been met with a great uproar from the blogging
community with many arguments for and against it.
Basically, the feature has been called by some ‘as evil as Smart Tags’, while
being recognised as ‘useful for the user’ by others. Authors have cried foul
over its modification of their documents without consent, yet it is praised
by others for giving the user the added information; ridiculed for inserting
links to a potential competitor’s site, yet applauded for giving the user some
choice in the matter; and finally criticised for the lack of an opt-in or opt-out
choice for the author or publisher, while some user’s are screaming back about
‘who’s user agent is it anyway?’
So, in light of all this discussion, I’m going to take an objective look at
the debate so far including all the major arguments I’ve found, both for and
against, giving the perspective of each affected group: content authors and
publishers, users, and Google; and examine the desire for author control by
means of an opt-in or opt-out mechanism.
Many high profile bloggers have noted their strong objections to this feature,
each relating to different aspects of the issue at hand. These problems include
document modification and the publisher’s rights to control their content; the
insertion of advertisements and the monopolistic effect this may have on competition;
and the lack of choice for the content publishers.
Counter arguments for each of these issues primarily revolve around the user’s
rights and ability to choose where and when such modifications take place, including
the fact that it’s not an automatic process. This raises questions about the
user’s ability to discern the origin of the links, how far software producers
(including Google) may go to provide such features and whether or not the proverbial
line has already been crossed.
Along with the arguments from those that don’t support AutoLink, there have
been requests for the ability to either explicitly opt-in or opt-out on a page-by-page
and/or site-wide basis, with most claiming that either would be acceptable,
yet debating about which is the most appropriate. On the other hand, those that
do support AutoLink claim that neither are acceptable and authors that should
not interfere with their user agent functions.
In summary, this can be broken down into separate issues that need to be addressed,
main objections relate primarily to the document modifications,
claiming that the integrity of the document is destroyed because the user agent
is modifying it (despite it being at the user’s request) to include links to
resources that the author did not intend to be there.
I believe that anything that changes the linking behavior of the Web is evil.
Anything that changes my content is evil. Particularly anything that messes
with the integrity of the link system.
In relation to this issue Scoble
also claims that the rights of the publisher
and of the user are in conflict with each other, and questions: in the favour
of which group should this issue be resolved.
As a user I’d love all sorts of things. But don’t you see that the rights
of the end-user are in conflict with the rights of the content producer?
Will you ALWAYS settle the argument in favor of the user?
Similarly, other’s have agreed with this and request that the rights of the
publisher be met by means of an opt-in or opt-out mechanism, which I will address
Counter claims have likened such document modifications to Ad blocker’s
that remove advertising content from pages, site scapers that extract or
alter the site’s content at the user’s request, the ability to control font
sizes, colours and the overall ability to override or disable stylesheets;
and other user agent extensions that perform document modification functions
all at the request of the user.
Cory Doctorow wrote in Why
you should love Google’s toolbar that
should be able to request a user agent to perform any document modification
at all, so long as the user agent is not, in anyway, committing fraud
(i.e. The user must know and understand that the modifications
are taking place).
I think I should be able to use a proxy that reformats my browsing sessions
for viewing on a mobile phone; I think I should be able to use a proxy
that finds every ISBN and links it to a comparison-shopping-engine’s
best price for that book across ten vendors. I think I should be able to
use a proxy that auto-links every proper noun to the corresponding Wikipedia
Dash agrees that the document is fodder for processing however
the user likes.
In the same way I can rip, mix and burn all the other media I bring into
my computer, something as straightforward as an HTML page should be fodder
for processing however I want.
In response to this issue, Dave Winer asks:
…what happens when Google isn’t satisfied to add links to our sites,
suppose they were to change the actual words? I haven’t heard Google
say they would never do that, have you?
That’s an interesting point, but it fails when you consider that such
modifications by other user agents already occur. In a follow-up
article on Boing Boing, Cory Doctorow and others proceed to list several user agents
and tools that perform document modifications at the user’s request on a
regular basis. In addition to this list, there are several other’s that
should have been included, which I would like to point out.
- The Linkification
extension for Firefox automatically converts plain
text URLs and e-mail addresses to clickable links, in a way similar
to Google’s AutoLink, by inserting <a> elements into the DOM.
- The W3C’s Semantic
Extractor scrapes content from a document and outputs
all the semantic information it can find.
- XSLT is designed to transform any HTML, XHTML or XML documents into
- AltaVista’s Babel Fish and other translators do change the actual words
of a document – that’s its purpose! As a result (because automated
translations are not perfect) the intended meaning of the author
may be unintentionally corrupted by the tool.
Graham explains, in response to some other questions from Dave Winer,
that such modifications are completely acceptable, as long as certain conditions
If a content-modifying function:
- has a definition that is completely understood by the user
- is only invocable at the user’s request and in isolation (i.e. not automatically)
- has an effect limited to the user who invoked it
then it’s entirely within the spirit of the Web, no matter what modification
it performs. No exceptions.
Indeed, I must agree in favour of allowing document modifications and note
that those conditions are met by each and every one of those tools and user
agents listed above. However the question is: are all of those conditions met
in the case of AutoLink? Basically, the answer to that question all comes down
to the user’s understanding and choice in the matter.
The User’s Choice
The user’s ability to choose the way in which they access content on the web
is central to the way the web works. A user may choose which browser they use,
which extensions they install, whether or not they want to see images, the allowed
fonts and colours in a document, and many other presentational aspects. A user
may choose whether or not they want to access a site and which links they want
to follow from a site. They choose whether or not they wish to access or download
resources, and which information or services they want from those resources.
AutoLink is just one service offered by the Google toolbar, which the user
has the complete freedom of choice to use or not. However, while some choices
are completely under the control of the user, others are not and are causing
a great deal of controversy. In relation to this issue, Dave Winer states:
The issue for authors and publishers is whether readers know they’re reading
text that’s been modified.
There are really two aspects to this question: whether or not the user knows
that changes have occurred and whether or not the user is able to determine
which changes were made. Google’s Chris
DiBona and Marissa Mayer describe how
the use of both AutoLink and the toolbar itself are entirely under the control
of the user. Chris DiBona as quoted by Robert Scoble in Ahh, evil
feature not evil if you force user to download, sign EULA, and click button? states:
We allow the user to never install the toolbar and never use the feature…
User choice is the difference.
Chris DiBona states in Oh
Since the user explicitly wants to do this, the user is choosing to augment
the content of a given website.
Danny Sullivan quoting Marissa Mayer in Google
Toolbar’s AutoLink & The
Need For Opt-Out – What’s Acceptable & What’s Not? states:
"It’s important to recognize that the toolbar is installed by people
who want Google-enhanced functionality," Mayer said. "I would argue
that the user is adding the link to the page. Google just provides the tool."
Yoz Graham also
supports this by stating that the user has both requested
the modification and has the right to do.
I say that it does nothing that the user has not specifically asked for. And
if the user has asked for it, there is no reason why they should not have it;
after all, they could save the HTML to their hard drive and edit it for exactly
the same effect. (In fact, the user could do far more wilful damage to HTML
than the AutoLink feature does.)
Since the user has full control over the feature and must request it each
time, the user is aware that such modifications have taken place, but the question
of whether or not the changes made are known to the user is much more valid.
Many have compared the appearance of AutoLink with Smart Tags, noting that
Smart Tags used a dotted purple underline. Google’s answer to this problem
is by providing a custom Google cursor and a descriptive title text, as Danny
Sullivan quoting Marissa
"The links that we add do look different. We work hard to help the user
understand that this was a link added by the Google Toolbar, that it wasn’t
a native link. We do this through a mouse rollover that is visible when you
mouse over the link."
However, Danny Sullivan questions this by asking whether that is really enough
and suggests that using a different underline, colour or other distinction
as well, may be an improvement. However, it’s also important to point out that
clicking the AutoLink button will sequentially highlight each recognised
link, however my tests using the examples
provided by Danny Sullivan seem
to indicate it will highlight any match regardless of whether the link was
added by AutoLink or already present.
I tend to agree that while Google have certainly taken some very good steps
to make the user aware of which modifications have occurred, the feature could
certainly use some improvement in this regard. Perhaps, as Danny suggested,
different colours or underlines may be the way to go, but I think a special
Google AutoLink icon inserted beside each link may be a better option, as it
may be difficult to select distinct colours and underlines given the wide range
of such styles chosen by every author using stylesheets. Thus, if only colour
or underline was used, it would be potentially possible that the colours chosen
may match or be very close to those already present in the document, as set
by the author’s stylesheet. While the chances of that are slim, I think to be
on the safe side an icon would be the better choice.
I mentioned earlier that there are choices that are mostly out of the control
of the user. These choices relate to which content is automatically linked (ie.
The user is restricted to ISBNs, US street addresses, several package tracking
numbers for selected couriers and Automobile VINs). The user is also restricted
by the choices available for which sites in particular are linked to.
Linking to Sites and the Effect on Competition
The sites that are linked to are, at this stage, chosen by Google. While the
user has a limited choice for the map provider (Google
Maps, Yahoo! Maps or
MapQuest), ISBNs link to Amazon, Automobile VINs to Carfax.com and package
tracking numbers link to their respective courier websites.
There is much concern over this issue with many pointing out that Google is
essentially inserting advertisements for their and their partners’ services
into other sites’ content. The most common example for this issue, first described
by Gary Price, is that ISBN links to Amazon are inserted into Barnes & Noble’s
site. While, as discussed earlier, the user has the right to modify the content
of the document if they so desire, the tool is effectively leading the user
to a competitor’s site of Google’s choosing.
Zeldman noted an objection to this, as it related to Microsoft’s earlier attempt
with Smart Tags.
They extended Microsoft’s monopoly power into new markets, giving the Redmond
giant the power to decide which non-Operating System companies would live and
which would die. (Companies Microsoft’s Smart Tags division partnered with would
live; their competitors would eat worms.)
If Google only includes features that make use of only their or their partners’
services, it may indeed unfairly harm the competition, and (in the worst
case scenario) create a monopoly for those organisations. Danny Sullivan explained
the Monopoly & Monetary
this issue; however, he also expressed the intention for Google
to include more options (including Barnes & Noble) in the near future.
Don’t like that choice? When the tool emerges from beta in the near future,
it is definitely planned for people to choose some of the content providers
they want to tap into. If you want links to Barnes & Noble for ISBNs rather
than Amazon, you’ll almost certainly be able to do that or pick from others.
If true, that will certainly be a huge improvement. Personally I feel that
there should be some way for any online book store, map service, courier
or whatever to include their services, perhaps by means of some kind of additional
plugin or setting.
For example, I would like whereis.com.au as
an option for my map service and a local book store like Angus & Robertson or Dymocks,
and I’m sure others would like localised services too. However, the feasibility
for Google to include each and every online book store, map service, courier
and vehicle information service by default may not be feasible, which is why
such providers and their users should be able to further configure the toolbar
for their needs.
Opt-in and Opt-out
A user agent is a tool used by a user to access and perform certain tasks
with resources. As such, a user agent should perform these tasks in a way that
benefits the user, provide choices to the user where appropriate and provide
feedback to the user about the tasks performed. Basically, the user agent must
act on behalf of the user at all times.
However, a popular request from those against AutoLink is the ability for
the publisher to temporarily disable the feature while users are viewing their
site, with several suggested variations in the implementation details. Dave
Winer’s preference is clearly for opt-in:
…we really want AutoLink to be opt-in… I might enable it to learn how it works
to decide if I want to use it for other pages and other sites…
However, as Danny
Sullivan correctly points out, opt-in is taking it too far,
and indeed harmful to the usability of the feature.
I think that’s too far. Users do have rights. They have installed this software.
Opt-out gives any publisher seriously concerned with the tool the ability
to control it on their site. Many won’t be concerned, so requiring an opt-in
is overkill that does hurt the user experience.
Yoz Graham agrees stating such features would never work if they required
Content creators should not have to provide specific opt-in permission; if
they had to do this for every such feature out there, most of them would never
Others tend to agree also, yet still request the slightly less harmful opt-out
feature through various methods such as: robots.txt
by Dave Winer; or a meta tag similar to that provided by Microsoft for Smart
Tags, as mentioned by Jeffrey
Zeldman in response to the performance limitations
of the various script solutions.
Client-side wear and tear could go away like a bad dream if Google would
do what Microsoft did with Smart Tags: namely, provide a meta tag that
Drew McLellan (via Zeldman) explains how the script
solutions, initially created
by Chris Ridings and modified by others, work by cycling through the links in
the page and removing those that are found to have been inserted by the Google
Both the opt-in and opt-out requests have been met with opposition
from Google’s Marissa Mayer, as quoted by Danny Sullivan:
"If you had opt-in or opt-out, that’s overall a lot less useful," Mayer
said. "If the links sometimes won’t show because there’s a publisher opting-out,
that’s bad for the user experience."
It is also noted that the toolbar will not modify any existing links, thus
being a form of opt out – an option which has been taken up by Barnes & Noble
by making all ISBNs on their site links to their own resources. However Danny
replies by stating that doing so for every link just to block AutoLink is just
not an effective nor feasible solution for everyone. However, in such cases,
I question the publisher’s right to prevent the user obtaining more information
where the publisher had failed to do so.
In response to the script solutions, Phil
Ringnalda has reminded us of exactly
user agent it is and provided a work around to bypass Chris Riding’s
original checklinks() function.
However, this solution will fail (without modification) on the site of
any publisher that alters the name of the function; and as Danny Sullivan points
is getting absurd.
…it highlights how quickly things have become absurd. You have third-parties
working to prevent AutoLink and potentially others working to prevent preventing
AutoLink. It’s a mess.
In my opinion, the valid issues against AutoLink seem to relate to the monopolistic
effect of the links; yet the requested opt-in or -out solution does not seem
to address this problem directly. Rather it gives authors control over how a
user agent processes their document in order to completely prevent document
modifications; particularly by implementation specific user agent features.
Publishers that take such action to interfere with user agent behaviour can
be seen very user-hostile; and user agents that obey such directives are acting
on behalf of the publisher rather than the user.
Also, in order to opt-in or -out of such user agent features, it requires
the use of proprietary extensions; regardless of whether they validate or not.
This was, and is, true of the
nofollow relationship, and will always be true
for AutoLink and other similar features.
For example, Microsoft have introduced various proprietary extensions in order
to disable user agent features including the Smart
meta tag, the image
meta tag and attribute and other (though more widely implemented) features
autocomplete. Some of those validate and others don’t, but that doesn’t
change the fact that they are proprietary extensions intended for use by a single
user agent to perform a very user hostile act – namely, to disable a user agent
feature against the user’s request. Thus any opt-in or opt-out functionality
introduced by Google would be quite harmful, just as the script solutions have
already proven to be.
As discussed, the document modifications performed by AutoLink are done so
at the user’s request, and the publisher has no right to interfere. While
some publishers may feel that the user is violating their rights, the fact
is that once a user has received your document, they have the right to perform
whatever functions they like with it, so long as the results are
limited to the user that performed them.
While Google, in an effort to provide a tool for performing useful functions
for the user, have certainly shown some desire to put the feature under the
user’s control by allowing them to decide where and when such functions are
performed and attempted to provide some (although limited) indication of
the results, Google’s role in the process is not impartial. It is Google
that chooses the possible link destinations with little ability for user
alterations and as such, has received much criticism over the potential effects
It is for these reasons that publishers and authors are requesting the ability
for their sites to be exempt from this functionality; however, any attempt to
override an implementation specific user agent feature is a user hostile act
which only serves to interfere with the user’s choices provided by the tool.
In my opinion, AutoLink is nothing more than a very advanced favelet (or bookmarklet)
that may prove very useful to many end users in the long run; however much more
control needs to be given to the user, particularly in regard to the service
providers chosen for the link destinations. Beyond that, I only hope that someone
will port this functionality as an extension to Firefox and content publishers
and authors cease their attempts to control a user’s system.