Search Suggestions Overhaul - Just Before the Google Instant Launch

[tedster]

From what I see, a major overhaul of Search Suggestions occurred just a few days before Google Instant Search was launched. All kinds of negative words (sucks, hate, scandal, scam, ripoff, etc) were removed from Suggestions except for situations like Goldman-Sachs where there is a documented scandal.

This was a much needed change. Previously, Search Suggestions were based heavily on search frequency alone - and many people routinely add a negative word to a company name just as a way of doing research.

The side effect was, up to now, an unintentional defamation against various organizations and companies via Search Suggestions who had no online negative reports. In fact, Google recently lost a judicial decision in France based on such defamation.

But having Instant Search throw those Suggestions at people would have been over the top, especially when there were almost zero results. Anyone who wants to do "scam" search can still do it, but they usually need to type in the complete query and complete an old-style search by pressing Enter.

[tedster]

Suggestions also seem to have changed the balance of plural versus singular forms - much to the dismay of marketers whose previous research had them focusing on the plural when only the singular now appears in Suggestions.

I'd guess that many other shifts were also included in the revised Search Suggestions. Have you spotted any other patterns?

[aristotle]

Previously Search Suggestions were based heavily on search frequency alone

But now search terms containing defamatory words and plurals are filtered out of suggestions even if they are heavily searched for?

This suggests to me that Google is trying to guard people from seeing results for those search terms in the instant results.

[tedster]

Previously even a stub page like "we have no reports of scams or ripoffs for [company name] could have triggered a Suggestion if the search volume was there. So that Suggestion alone could have a "chilling effect" on the business involved.

So yes, these kinds of Instant results would have been potentially defamatory (borderline slander, I'd say, but IANAL) and IMO they should have been removed a long time ago. In fact, there was an evolving black-hat attack vector on a competitor's reputation: performing such searches in bulk, and of course, with certain stealth precautions to get around the Google safeguards.

Of even more concern to me is the shift in singular/plural balance - a lot of research on search behavior could just have been neutered.

I don't think Google is intentionally skewing the Suggestions, but the potential for bad Suggestion data to shift traffic patterns is immense. The potential for conscious and intentional manipulation is there as well - and that possibility makes me shudder.

So with Instant, there is a newly heightened power given to Suggestions. The changes in site traffic brought about by this power need to be researched in the coming weeks and understsood. I'm sure it's more than just [scam] searches and singular/plural.

[aristotle]

So yes, these kinds of Instant results would have been potentially defamatory (borderline slander, I'd say, but IANAL) and IMO they should have been removed a long time ago. In fact, there was an evolving black-hat attack vector on a competitor's reputation: performing such searches in bulk, and of course, with certain stealth precautions to get around the Google safeguards.

Tedster - I understand why you support what Google is doing here. I just wonder if it's partly a reponse to complaints from big corportaions and/or advertisers. Anyway, it's apparently a manual intervention, therefore a departure from Google's philosophy of letting everything be determined by algorithms. And it could lead to more censorship in the future.

[Robert Charlton]

Suggestions also seem to have changed the balance of plural versus singular forms - much to the dismay of marketers whose previous research had them focusing on the plural when only the singular now appears in Suggestions.

I was at first inclined to think that the singular/plural change might have been an anti-manipulation move. On sites that sell widgets, eg, it's often single widgets that are sold, whereas searcher data showed that the plural form was most commonly searched. SEOs therefore have often targeted both. I know I have.

But... Google has now apparently also revised its Keyword Tool data. Searches where the plural search apparently outnumbered the singular form by two-to-one a few months ago are now showing more or less the opposite.

On phrases I've targeted, consistent with old Google data, it's almost always been the plural that's been most productive.

I should add that on some of the very prominent widgets vs widgets searches, I'm still seeing the plural suggested instead of the singular, but that's on searches where there's a huge bias, onpage and in search, towards the plural.

How much, btw, is onpage occurrence, if any, affecting the Instant suggestions? Originally, the Suggest tool was almost entirely based on onpage data... clearly much less so over time. But some phrases have obviously not been suggested simply because they don't occur often enough on the web.

[tedster]

Originally, the Suggest tool was almost entirely based on onpage data

Just checking on this. You're talking about the Keywords Suggest Tool itself, not Suggestions provided in the dropdown below the query input box. Right?

[tedster]

just wonder if it's partly a response to complaints from big corporations and/or advertisers.

The lawsuit they lost was brought by a very small business.

Anyway, it's apparently a manual intervention, therefore a departure from Google's philosophy of letting everything be determined by algorithms.

Not exactly - I'm sure it's still a change to the algorithm that determines what Suggestions to use. there was a flaw or edge case in the original heuristic that wasn't well-accounted for at first.

And it could lead to more censorship in the future.

Yes, that's the potential that makes me shudder. Google's whole direction of spoon-feeding the general user (instead of asking them to mature) is dangerous. It is not healthy for the general population to INCREASE their dependency on any authority whatsoever.

[aristotle]

Not exactly - I'm sure it's still a change to the algorithm that determines what Suggestions to use. there was a flaw or edge case in the original heuristic that wasn't well-accounted for at first.

Well, maybe I should have said that they made a manual change to the algorithm to give special treatment to certain words. That's different from letting searchers determine which words are most relevant to a particular subject.

In my opinion most big corporations are not lily-white and in many cases deserve to be criticized. Google needs to find a way to filter out black-hat attacks without slso filtering out justified criticism.

[Robert Charlton]

Originally, the Suggest tool was almost entirely based on onpage data

Just checking on this. You're talking about the Keywords Suggest Tool itself, not Suggestions provided in the dropdown below the query input box. Right?

No... in talking about the "Suggest" tool, I'm referring to Google Suggest, the search suggestions, which evolved from a Labs project and eventually started offering search suggestions on the main page. It's now incorporated in Instant. (I'm assuming, btw, without having checked carefully, that the search suggestions with Instant on are the same as the search suggestions with Instant off.)

The Keyword Tool is the Google AdWords Tool external. It's available to anyone and seemed therefore to be a proper basis for discussion.

PS - I should probably clarify this...

I should add that on some of the very prominent widgets vs widgets searches, I'm still seeing the plural suggested instead of the singular, but that's on searches where there's a huge bias, onpage and in search, towards the plural.

Here, I'm really talking both about Suggest and the Keyword Tool. Sorry for that degree of confusion.

My comment, "a huge bias, onpage and in search, towards the plural" indicates that common usage is unambiguous. I'm not suggesting cause and effect.

[tedster]

My understanding was that the primary data for Suggest comes from user searches, not page spidering or potential results. Here's a section of Google's newly updated Help page.

The search queries that you see as part of autocomplete are a reflection of the search activity of all web users.

Just like the web, the search queries presented may include silly or strange or surprising terms and phrases. While we always strive to neutrally and objectively reflect the diversity of content on the web (some good, some objectionable), we also apply a narrow set of removal policies for #*$!ography, violence, and hate speech.

[google.com...]
And along the lines of "silly" try typing in (not pasting) a query for [urethra function] - the minute you get to the "f" you see the #1 suggestion become "Urethra Franklin".

[Robert Charlton]

Forgive in advance... I see this is going to get long. It's leading up to a question....

Here's a section of Google's newly updated Help page.

In a discussion from early October 2009, about Google Suggest, I cite that help page, which has changed quite a bit since then....

Google Suggest - Can We Affect the Negative Entries?
http://www.webmasterworld.com/google/4000884.htm [webmasterworld.com]

In the quote that I'd posted from the page, I'm here adding emphasis to highlight one of the main changes....

As you type, Google Suggest communicates with Google and comes back with the suggestions we show. If you're signed in to your Google Account and have Web History enabled, suggestions are drawn from searches you've done, searches done by users all over the world, sites in our search index, and ads in our advertising network. If you're not signed in to your Google Account, no history-based suggestions are displayed.

I don't know when the page changed, but the above wording was up there for the several times I subsequently checked it. I wasn't just relying on the above alone, though. I mentioned in the thread that...

...when Suggest was a Lab project the Google Suggest numbers originally correlated roughly with the the number of pages returned that contained suggested target phrases, not with search popularity.

Additionally, there's the question of those Microsoft directory searches I mentioned...

While we can't discuss specific domain names or search terms here, we can mention, say "microsoft.com". Type it in the Google search box and watch how the suggestions change as you type each letter... and note what happens when you get to the dot in .com, and then the "c". Finally, you'll end up with a set of products associated with "Microsoft.com" and subdirectories on the Microsoft site, which are also of course associated with Microsoft.com.

I just tested this with Instant, and Microsoft subdirectories are still suggested. I can't believe that people remember these pathnames without help. It may be, though, that these are the residuals of what searchers did enter after they'd had the Suggest suggestions prompting them. I vaguely remember that some of the old MS directory searches suggest were really long.

In addition, I checked a variety of other prominent sites, and I noticed popular subdirectory suggestions. Checking now, on Adobe, if you type enough, you'll get this one: adobe.com/go/getair

I could be wrong, but I don't think that's a naturally popular search without some help, though it might be, as it's alliterative enough to be memorable. It also may have been a naturally popular destination at one point that was suggested by an earlier version of Suggest and now lives on as skewed data in memory. ;)

Note also this discussion, in which I discuss Google's "extrememly aggressive" spell suggestions, that I attribute this to phrase occurrance....

A Red "Did you mean: ____" Shows in Google Drop Down Suggestion
http://www.webmasterworld.com/google/4178034.htm [webmasterworld.com]

As I was seeing it, the red "Did you mean?" would pop up as queries were being typed, on searches where there was a common alternative spelling and where the misspelling occurred in less than 300 or so instances on the web. Where there was no common alternative, Suggest appeared to let it go. Search frequency also apparently entered into this.

I'll go a little further out on a limb about this and mention that from Google's phrase based indexing patents, and I don't have my fingers on the passage right now, the above feature is more or less predicted... that if there's a query without very many good results on the web, the phrase tables can be used to generate alternative suggestions.

So, the question is whether Google might still be factoring in phrase occurrence and destination popularity into the Instant/autocomplete suggestions.

If so, it might explain the perhaps skew now towards singular, perhaps intentionally to correct SEO skewing. It could be easily done algorithmically, I think, by looking a singular vs plural phrase occurrance on many sites, which, as I said, are targeting the plural even as they are mostly selling the singular.

I should add that the Negative Entry thread I cite above also discusses negative entries. I was basically just following my nose on that thread, as there wasn't much supporting discussion.

[Robert Charlton]

PS to the above - Though I say "skew now towards singular", that doesn't at all cover the change of singular/plural data in the keywords tool, which is another discussion.

[tedster]

thanks - good observations. I guess we can assume that the current Google Suggestions are a mash-up of a good number of factors, and in a junk drawer like that you might find all kinds of things!

How they created Suggestions was pretty much a side question for me until Instant threw the issue right into center stage. For now it looks like Suggestions are a critical layer in search marketing!

[RP_Joe]

I am not sure what the future of instant is. It is a increase of bandwidth. This is a problem in the wireless environment like 802.__. Many coffee shops, Airports , Schools use this method. It is a problem for wireless carriers who already do not have bandwidth.

I am not sure people like the fact that Google keeps changing. The feedback I have heard is that people don't like this change or that change. In my career in software development we learned to manage change so as not to overwhelm the user.

The only Google killer in the marketplace is Google themselves.

.

[incrediBILL]

Suggestions also seem to have changed the balance of plural versus singular forms

Not seeing that change in my niches - not yet anyway!

[tedster]

Agreed incrediBill - the first reported example I got was accurate and the site was seriously suffering because of the change. I checked a few more terms and they also had changed - the Suggestion had changed from plural to singular and the greatest traffic had been on plural.

But since those first reports, I tried to broaden the data set and I can see that this is not a uniform change at all. As you said - it may hit only certain terms or market niches.

I'm guessing there will now be regular Suggestion updates to go along with ranking algo updates. Maybe they were always there and just escaped notice, I don't know. I do know that a Suggestion update definitely lit up my radar last weekend, just a few days before the launch of Google Instant.

[anand84]

From what I see, not all negative words are filtered. Those negatives that would seemingly contain a balanced article of the company's wrong-doing still exists. For instance, "Scam" is a word that a New York Times or other major media might use to write about wrong-doings. But "Sucks" or "Rip-off" they might not..

I still see "scam" as part of the suggestions, but not the other two..I think the point here is to provide balanced negative reports.

[Nuttakorn]

I have seen some article talking about three main factors to drive Google suggest : Content, Anchor Text and Search Volume. What do you think about this?

[indyank]

thanks - good observations. I guess we can assume that the current Google Suggestions are a mash-up of a good number of factors, and in a junk drawer like that you might find all kinds of things!

Tedster and Charlton,

interesting discussion...

Well I know of atleast one example where suggest has been manipulated by a spammer...to me... that is incredible as the guys seem to have deciphered how the google suggest works so quickly... and if that secret is known, it is going to be a huge weapon for spammers in particular...

I could message you the example..if you like to see it...

[tedster]

Nuttakorn - yes, those three factors make sense as the main "buckets" - and then there would be detailed factors within those buckets.

indyrank - it's Instant that's new. Suggestions have been with us for a while. Yes, I do know that there are people working to manipulate the Suggestions. I wouldn't say there's any one secret being used to spamming Suggestions - it's more like a combination effort aimed at all three of the buckets.

It's not my way of doing things but, it was an inevitability I guess, given the potential Suggestions have to drive traffic. And that potential has just taken a major jump.

[J_RaD]

well this explains why i didn't get any instant results on the day it rolled out when i typed "google sucks"

it won't even auto suggest a negative term which I think is bad because its using the power of the searchers. Example, product A really does suck and its a total waste.

when you type Product A it will auto suggest Product A sucks

why? well thats easy cause 10000000 of people were searching/saying/posting/blogging etc that exact thing.

now they get off free? not good.

if product A it awesome i want it to auto suggest that to me, if its bad i want it to auto suggest that as well, if its auto suggesting such things i can see why and make my own mind up.

[tedster]

if its auto suggesting such things i can see why and make my own mind up

A very good point. I do note that Google has not purged all negatives from their Suggestions - not at all. The first example I thought of trying was [domain registration scam] and both the Suggestion and the Instant results were there.

There are legal implications both ways. Google is slipping from merely organizing the world's content into offering an editorial element. They could be "evil" both because of Suggestions they do show and Suggestions they don't show.

Just as the regular Press doesn't just report the news but also shapes it, Google is not a neutral passageway for the world's information. As others here have remarked, Google is not exactly a "search engine" anymore.

This influence is subtle, but that subtlety has the potential to be more insidious. If people were already asking for transparency in the ranking algorithm, it seems to me that transparency in the Suggestions algorithm is even more critical, especially with the Google Instant interface.

[jonathanleger]

The "scam" queries aren't completely gone. Start typing "google sca" and "scam" does get autocompleted.

This is also true of other well known names. While the "scam" suggestion doesn't appear on the initial list after the name is entered, if you type "[person's name] s" "scam" will often appear in the suggestions.

I also tried typing "[software title] s" for a piece of software I use and "scam" appeared in the list. When I checked the AdWords Keyword Tool for the phrase it showed "not enough data" -- so apparently the suggestions are not just based on people actually searching for the phrase, since Google's showing zero data for that query.

[J_RaD]

so apparently the suggestions are not just based on people actually searching for the phrase, since Google's showing zero data for that query.

even more interesting! makes you really wonder how much googs SERPS are manipulated

[Robert Charlton]

Suggestions also seem to have changed the balance of plural versus singular forms...

I've revisited this, because it frankly didn't sound right... and, at least on the big money, high-volume search terms, I'm not seeing it. That said, I am seeing some Suggestions changing over time, whether or not I'm signed in, or have flushed cookies, or I've used different browsers, so some kind of refinement of the suggestions could account for some changes. Localization also enters into it.

In some specialized niche searches, particularly technical searches where the searchers might tend to be precise, I am seeing singular searches outnumbering plural searches... and this is consistent in both Suggest and the "exact" match search volume information on the Keywords Tool. This is a surprise, as I'm familiar with some of the areas I've checked and I wouldn't have expected it.

But the mainstream singular/plural Instant suggestions are almost always in line with expected and experienced search volume, again using "exact" phrase match information. On major terms, plural searches continue to outnumber singular searches by as much as ten to one... even though onpage occurrance was more or less a toss-up.

At the same time, I also see suggested examples where onpage occurrance is clearly the driving force.

Localization... which has consistently been the big personalization factor that I've seen on Google serps... throws some interesting variations into the Instant suggestions. I'm seeing that a cityname search for an area that's local to the searcher, eg, might return "cityname weather" as the first suggestion... generally not true if you're searching from far away, where that's not likely to be of such immediate interest.

And, as an example of the kind of thing that might be skewing some singular and plural searches, I'm going to use a specific example (but one whose mention here isn't likely to affect search results)...

When I start searching "paris" on google.co.uk, I get several suggestions, one of which is "paris hotels".

But when I type in "paris" on google.com from the west coast US, I have usually gotten "paris las vegas" as the first result, since Las Vegas is a relatively short trip from the west coast.

Continuing on google.com... if I ignore "paris las vegas" and continue typing as far as "paris ho", I see...

paris hotel
paris hotels
paris hotel las vegas.

The Google AdWords keywords tool shows this order of search volume for these four phrases, with the plural hotels roughly 9 times the search volume of the singular hotel either in Local or Global exact match searches....

paris hotels
paris hotel
paris las vegas
paris hotel las vegas

I haven't looked hard to see if there are any other such ambiguities that might skew singular/plural searches like this, but it's obvious that name ambiguity and localization are involved in these

The Suggestions also are changing or evolving quite a bit in some areas I've spot checked. I noticed, btw, that the first time I tried the "paris" search after Instant came online, "paris france" was the first suggestion. I haven't seen it since. The other Paris, btw, even with "paris h" typed, has never been suggested, even though her search volume outnumbers "paris hotels" globally by roughly 50%.

[tedster]

The Suggestions also are changing or evolving quite a bit in some areas I've spot checked.

Yes, I've been seeing this too over the past four days. I now think Suggestions may be one of the continually updated areas for Google - making good use of the Caffeine Infrastructure.

[Robert Charlton]

I am seeing in some other countries, where you have to be signed in to use Instant (and thus Instant is not yet the default), that the singular form in some searches is shown before the plural. These are searches where I know the AdWords Tool reports exact search volumes for the plural are higher, though not by as wide a margin as in the US. It will be interesting to see, as the suggestions evolve, whether these will shift. Currently, the plural isn't even shown in proximity to the singular, so it may be hard for some searchers... those who are 'obedient' to the list... to make a choice.

I should mention, incidentally, that the Suggest and the Instant suggestions are not the same... they can't be since there are only 5 Instant suggestions vs 10 Suggest suggestions... but they seem to be picked in the same order from the same list. I haven't checked the details carefully, but it seems that the first 5, minus near redundancies, are used.

[Robert Charlton]

PS to the above... a lot also depends on how the searcher uses the interface.

Continuing to type would produce more choices in which the plural would be seen... but scrolling down to the singular and clicking would never show the plural as an option.

[Robert Charlton]

I should mention, incidentally, that the Suggest and the Instant suggestions are not the same... they can't be since there are only 5 Instant suggestions vs 10 Suggest suggestions... but they seem to be picked in the same order from the same list.

PS - Since I posted the above, Instant has reverted back to 10 Suggestions. They seem to be trying it both ways. Anyone else seeing this?

PPS - Very soon after posting, I left google.com and came back, and Instant, at least in the west coast US, is now back to 5 Suggestions. I'm also seeing European engines now offering Instant by default whether you're signed in or not... and on these it's 10 suggestions. I assume it will keep on changing for a while.