Last updated Oct. 11, 2003.
by Greg R.
Google does not always behave as advertised nor deliver
the results expected. This page aims to document both ongoing and
short-lived inconsistent search behavior on Google. Contact with
reports of any inconsistencies you observe. For more details about how Google works, see the full Google Review.
INCOMPLETE RESULTS: Google, like other search engines, has always
given inaccurate numbers (see Counting below). But at
least as early as Oct. 1, 2003, some search results would just stop and not give any more
listings. For example, a search on
game candle gave "Results 1 - 1 of about 321,000." Gary Stock has posted
detailed descriptions of the strange phenomena at Google. He has dubbed these GoogleNACK (as in Negative ACKnowledgements).
Seth Finkelstein postulates that the malfunction is related to Google's spam defenses.
While some of the early examples, like
game candle, are fixed, others continue to be problems. As of
Oct. 11, 2003, both
keyboard bracelet and
motorcycle candle fail with "Results 1 - 19 of about 48,600"
and "Results 1 - 69 of about 64,000" respectively.
SITE LIMIT FAILURE: A search such as
site:www.google.com google should only find pages at Google.
Yet with the number of hits set to 100, some records come up from
adobe.com, digits.com, osdn.com, and even washington.edu. The search of
site:google.com google seems more accurate at least in the top
100 hits, but it pulls in results from directory.google.com,
PHRASE SEARCH INACCURACIES: Phrase searches do not always work right on Google.
Sometimes this is due to Google finding the words within the phrase query
in anchor text on pages that link from that anchor text to the page shown
as a result. But the words do not necessarily appear in exactly the
correct order within the anchor text. See the search on
"american medical association" for example, a problem that dates back to at least 2000.
Another example from Dec. 22, 2002: A search on
"montana mountain ranges"
found a page that contained "Montana's Mountain Ranges" but did not
contain the exact phrase given.
HIDDEN QUERY TERMS: This can happen with any search engine, but it
surprising none the less. Sometimes query words do not show up in some
results or in the cache. This can be due to the query word matching a link
on another page that points to the URL that is listed on the search
results page. At other times it has nothing to due with that. Take for
example a search on
cameras run on Aug. 16, 2003. The 3rd ranked result was for
"Jersey Swimwear, USA" at www.milkduds.com. Neither that page nor the
cache copy at Google contain the query word. The cache copy states that "These
terms only appear in links pointing to this page: cameras."
I then tried
link search on that URL, but none of the pages that were returned used
the query term in their link to that URL (and I checked in the cached
version rather than the live page). After more digging, the best
explanation I could find is that another URL,
www.intertain.net/~cameras, redirects to www.milkduds.com,
even though Google does not show this. So apparently, Google is matching
on words within redirected URLs.
Status: Varies, irregular occurrences
FIELD SEARCHES IN COMBINATION:
Field searches on Google have some strange rules. Since at least April 2000, the
related: field searches
cannot be combined with any other query words. Initially, the search
related:searchengineshowdown.com found exactly the same hits as
As of 2003, adding another term makes Google search for the field name as a search
term, so the above search is the same as
bookmarklets "related searchengineshowdown com" This is
especially problematic with link searching since it means that you cannot
exclude pages on the same host that link to the site. The field labels that
start with "all," (such as
allinanchor:) also cannot be combined with other search terms, or else
the additional terms will be searched within that field as well. To combine
these field searches with other terms, use
The one exception is the
site: limit which has the opposite rule in that it cannot be used alone and must have another search term.
results in an error message while
INACCURATE OR: Google introduced the OR operator in Oct. 2000, a welcome
addition when searching terms with synonyms. Unfortunately, the OR does not
always work properly. There have been problems since at least Nov. 2002. One
example of the problem reported to me was a search for
kabul OR kaboul OR kaboel found less results than just
kabul OR kaboul
even though it should have found more.
COUNTING: Google can't count accurately. The total number of
search results presented is almost always stated as "about" some number that
when large is typically a multiple of 1,000 or 10,000. But is even this "about"
number anywhere close to reality? We can't see more than 1,000 or so results, so
it is hard to be sure, but on small searches, the numbers are often inaccurate.
For example, a search (Oct. 2003) on
grapevine durian claimed
to find "about
236" records, but when changing the number of results displayed to 100,
Google changed the estimate to "about
234." Then, go to the 2nd page of results and turn of the site clustering so
that more than two pages per site are displayed (or just add &filter=0 to the
URL, and Google jumped back to "about
236." But on the second page of results Google only gives "Results
101 - 152 of about 229." Sometimes Google drops the "about" modifier
on very small searches, but the numbers are still wrong.
FIELD SEARCH COMMANDS FAIL: Since at least May 27, 2003, certain field
searches on Google have stopped working. The
intitle: field searches are not working properly, even though the
allintitle: do work. See the Field
Searching section of my review for how they are supposed to work. What this
means is that you can no longer use Google to search for some words in the title
with others in the body. See for example the search
intitle:tourism "market research" which finds pages that do not
include 'tourism' in the title.
Status: Fixed as of Aug. 16, 2003
MISSING INDEX TERMS: Google may report zero results for
some terms even when it has indexed pages that contain those terms. One
example reported to me on Feb. 1, 2000 is the term
AW870 (a type of speakers). The search resulted in zero hits. However, a search on another speaker, the
found 71 hits, including some that had AW870 in both the displayed
summary and the title. Checking the cached version (which is what
Google indexed) shows that the term did indeed appear on the page when
Google indexed it.
Status: Fixed as of Sept. 2000