Professional Documents
Culture Documents
Sign OutRating HomeRater HubManage Account
Sign OutRating HomeRater HubManage Account
arrow1483230evitanora@gmail.com
0
User Guide
Program Introduction
Using the EWOQ System
Task Instructions and Concepts
>Common Task Instructions
>Reporting a Problem
Rating Mobile Ads
General Information
>Work Requirements
>System Information
>Glossary
>FAQ
>Contact Ads Eval
This section contains the task instructions for the main rating tasks. For
each of the main tasks you will be able to see the full task instructions as
well as a screenshot of the task itself. All task instructions and task
screenshots are available in this document. You will be able to search for
terms or tasks by using CTRL+F (on Windows) or Command+F (on Mac) to
search for text in this document.
Search Ad Instructions
During your time as an Ad Quality Rater, you should expect to encounter several
different task types in the ad rating system. Instructions are always accessible at the
top of all tasks. The most common task type is the search ad evaluation and
instructions for this task type are included below. These instructions are identical to
those found on Search Ads Evaluation tasks.
Search Ads Evaluation: General Guidelines
Version: 2015-08-20
Search ad rating involves interpreting a user query. A user query is the set of
keywords that a user enters into the Google search engine. When rating a search
ad, perform the following steps:
1. Review the Google search results page,
try to understand the user query, and
form an opinion about what the user
hopes to accomplish by using a search
engine.
2. Use the evaluation criteria found in the
following instructions to analyze an
advertisement and the advertising
experience the user will have if he or she
clicks on the ad.
User Intent
An understanding of the user intent is necessary to accurately rate a search ad. The
user intent is what the user hopes to accomplish by using the Google search engine.
Note that users use the search engine to look for a variety things, and there are
many user intents.
Some queries are very easy to understand, others are more difficult, and some may
seem impossible to understand. Regardless of its meaning, you must research the
query and form an opinion about the user intent. We strongly advise you to review
the Google search results page to determine user intent. In order to objectively
determine how promising or unpromising an advertiser offering is for a particular
user query, it is important to form an opinion about the user intent before beginning
an analysis of the advertisement.
Misspelled Queries
Users often misspell queries. When evaluating a query, if it is clear what the user
means, and the misspelled version of the query has no meaning, ignore the
misspelling.
Analysis is more difficult if the query appears to be a misspelling, but the misspelled
version has a unique meaning. First consider the query as the user entered it, and
then consider if it may be misspelled. If advertisers respond to misspellings, ratings
may need to be adjusted.
Refer to the following example to better understand misspelled queries:
User query: [ goodnight moom ]
There is a famous children’s book
called Goodnight Moon. It is very
possible that the user means to
type [ goodnight moon ] but types
[ goodnight moom ] instead.
However, there is actually a novel
titled Goodnight Moom. While the
novel is quite obscure, it might be
what the user wants.
Potentially Scammy Ad
Does the ad look like a scam? An ad that seems too good to be true, sleazy, or
deceptive to users is usually bad.
Very Unpromising
There are several cases where Very Unpromising is the only appropriate rating.
Very Unpromising ads have no
reasonable chance of satisfying the
user. Try to put yourself in the user's
mindset - is it possible at all that the
creative offers something useful to
the user? If there is no reason at all
to think that the user will find the
creative useful, rate it Very
Unpromising. (Note: you might think
“It’s always possible that someone
might find anything useful, even
though it has nothing to do with the
query.” Don’t go that far!)
If the creative looks like a scam, or
leads the user to harm, rate it as
Very Unpromising.
If the creative falls into one of the
categories listed in the Machine-
Generated Ads section, rate it as
Very Unpromising.
If the creative promises to do the
impossible, such as selling a person
or city, rate it Very Unpromising.
Just because there is a strong term
overlap between the query and
creative does not mean the ad is a
good match for the query. If the
user is searching for
[ homeowners insurance ] and the
ad is for “medical insurance,” the
user will very likely not find the
creative useful and you should rate
it as Very Unpromising.
Machine-generated Ads
Some ads are partially auto-generated to take words from the query and place them
in the creative text. There is nothing wrong with this in itself. For example, if the
query is [ xbox 360 used ] and the creative says “Buy a used xBox 360 on eBay,”
that’s a good ad. Unfortunately, sometimes these machine-generated ads turn out
very badly. Very Unpromising ad creatives may have some of the following issues:
Things offered that cannot be bought
User query: [ san diego, ca ]
An ad that says “Buy San Diego
cheap on eBay” is ridiculous--you
can’t buy a city. Ads that are
unintentionally ridiculous, horrible,
or offensive, by suggesting that
you can buy concepts, human
beings, body parts, criminal acts,
or similar things are Very
Unpromising.
Part of the query removed, substantially changing the meaning
User query: [ roses lime juice ]
An ad that offers the action, “Buy
roses,” has left out so much of the
query that the entire meaning has
changed. By taking only part of
the text of the query what remains
substantially changes the
meaning.
Part of the query removed, resulting in overly general ad
User query: [ how do i remove
gum from satin ]
An ad that offers “Get information
on how to remove,” is nearly
meaningless: too much has been
removed from the query. By
taking only part of the text of the
query, the result is far too general
to be promising for the user
query.
Nonsensical, jumbled, or ungrammatical ad creative
User query: [ how do i remove
gum from satin ]
An ad that says “Search for how
do I remove gum” or “Find how do
I remove gum from satin” is
awkward and ungrammatical. Ads
that end up nonsensical, jumbled,
or ungrammatical because a
query has been crammed into a
space where it doesn’t really
belong is Very Unpromising.
Be on the lookout for these. If you’re not paying close attention to how the ads
actually look and sound, it can be easy to think these look fine—but to a user who is
actually reading the text, they can look laughable, annoying, or foolish, and in some
cases, deeply offensive or hurtful. Even those that just look sort of silly or awkward
are very bad.
Ads with Additional Features
Some ad creatives are just text and a single link to the advertiser page. Other pages
contain additional features that may or may not provide something valuable to the
user. Ad creatives may contain maps, videos, images, star ratings from customers,
multiple links to specific pages on the advertiser site, and a variety of other features.
These additional features may affect the quality of an ad creative. If the special
features add to or detract from the appeal, informativeness, or usefulness of an ad,
the Ad Creative score can be raised or lowered.
For an ad that contains only text and a single link to an advertiser page, use only the
previous criteria in making a decision. For ads that contain anything in addition to
text and a one link, consider the following factors, and decide whether to raise or
lower your score:
If an ad does not deserve a score of
Somewhat Promising or Very
Promising based on the previous
criteria, be cautious about giving it a
positive rating just because of
additional features.
An ad that deserves a score of
Somewhat Promising or Very
Promising based on the previous
criteria can be given a negative
rating if additional features detract
from it severely.
Use common sense when deciding
whether additional features improve
or detract from an ad enough to
move it between Somewhat
Promising and Very Promising
scores in each category.
An ad that is scammy or harmful
can never be improved by additional
features.
Additional feature should relate to
the user intent in a sensible way. If
the user is looking for information
about a current movie, a movie
trailer in the ad creative relates to
the user in a sensible way, but a
map to the movie studio where the
film was made does not. The trailer
probably improves the experience,
but the map detracts from it.
Where an additional feature is
relevant to the user intent, it should
be informative, easy to use, and
clear. If it is confusing, boring, or
hard to figure out, it may either
detract from the experience or just
fail to improve it.
An ad may have multiple additional
features. Consider all of them
together when determining your ad
creative score.
Raise or lower your rating by a small
amount if the additional feature has
little impact on the ad creative.
Raise or lower it a large amount,
according to the previous criteria, if
the additional feature has a big
impact.
Do not consider the physical size of
an ad creative rating it. If the size of
an ad creative causes it to display
incorrectly in the ads rating
interface, alert an administrator but
ignore the issue while rating it.
Only rate the landing page that opens after clicking on the Visit Landing Page
button. Do not base your score on pages that are accessible by clicking on links in
the body of the ad creative. NEVER copy and paste a link to visit the page, and
NEVER manually change the URL.
The fundamental principle of landing page evaluation is this: the user starts a search
on Google.com with a goal in mind. The user then enters a query and reviews
Google’s search results and ads. The user then clicks on the ad currently being
reviewed, and that ad takes the user to the landing page. Keep in mind that in order
for a user to have a positive experience with an advertiser landing page, he or she
should be closer to the goal expressed in the query, otherwise it is a negative
experience. The section below helps frame how distance from the user’s goal helps
determine a landing page rating score.
Distance from the User’s Goal
Carefully review the Google search results page to determine the distance from the
user’s goal. Does the Landing Page move user closer to his or her goal, further from
the goal, or neither closer nor further from the goal?
If the user is closer to the goal, the landing page deserves a positive rating. For
example, if the user is hoping to buy a specific camera, and the landing page is a
store offering that camera for sale, the user has come closer to accomplishing his or
her goal.
If the user is further from the goal, the landing page deserves a negative rating. If
the user is hoping to buy a specific camera, and the landing page is a store offering
pet food, this is a dead end. The user will need to go back to the search page or start
a new search, so he or she is actually further from the goal than before clicking on
the ad.
If the user is neither closer to nor further from the goal, the landing page deserves
a negative rating. If the user is on a Google search results page and clicks on an ad
that just takes them to a page of similar search results, which overall did not provide
any additional value, no progress has been made; the user is no closer to or further
from the goal than before clicking the ad.
Deciding this is not an exact science. Rely on good judgment. The following
guidelines more deeply explain how to generally rate landing pages, but they do not
explain how to rate a landing page in every situation.
Considering Trustworthiness
Do not give a landing page a Satisfaction Possible or Satisfaction Likely rating if you
do not trust the information found on that landing page or if you would not make a
purchase from the advertiser site. A page that offers the exact product that a user is
looking for is useless unless the user trusts it enough to actually make a purchase
there. A seemingly trustworthy merchant selling a particular camera at a particular
price might deserve a better rating than a page that clumsily aggregates a random
set of products, even if the same camera at the same price is offered on that page
too. Similarly, a page offering the exact information that the user is looking for is not
useful if there is no reason to think that the information is correct. For example, if the
user seeks some medical information, a site belonging to a medical school is a good
source of trustworthy information while a blog post by an unknown person is a much
more doubtful source. Never use a rating of Satisfaction Likely or Satisfaction
Possible if the page appears scammy or harmful.
Dissatisfaction Possible
If the page is marginally related to
the query and you think that there’s
a small chance the user would be
interested, use Dissatisfaction
Possible.
If the page can eventually lead to
what the user wants, but only
through many clicks or through
clicks that lead to an entirely
different website, use Dissatisfaction
Possible.
If the page offers something that
you think the user might be
interested in, but not what the user
was looking for and not especially
close to it, use Dissatisfaction
Possible. For example, if the user is
looking for baseball gloves, and the
landing page offers athletic socks,
there’s probably some chance that
the user might be interested.
However, it’s not what the user was
looking for, and not all that close to
it, so it deserves Dissatisfaction
Possible.
If the page can eventually give the
user what he or she is looking for,
but the process is protracted and
difficult, use Dissatisfaction
Possible.
Dissatisfaction Likely
If the page has nothing to do with
the query, use Dissatisfaction Likely.
If the query is for a product or
service, and neither the
product/service nor anything close
to it can be purchased from the
page, use Dissatisfaction Likely.
If the query or a word in the query
has two meanings, it is clear which
meaning is intended by the user,
and the advertiser responds to the
wrong meaning, use Dissatisfaction
Likely. For example, [ cars 2 ] refers
to a movie. A page for a car
dealership is clearly a bad landing
page for this query, even if it might
be a good result for [ car sales ].
If the page looks like a scam, you
think users could be harmed by it, or
it either attempts to trick the user
into downloading something by
labeling a download button in a
confusing way or tries to download a
file without action by the user, use
Dissatisfaction Likely.
If the page loads but is completely
unusable (for example, because
some content does not load, or
page doesn’t display properly) use
Dissatisfaction Likely. If enough of
the page does not load at all (for
example, you encounter a 404
error), use the Error Did Not Load
flag instead of a rating.
If the page is very bad for any other
reason, use Dissatisfaction Likely.
Query Flags
Use these flags to indicate that a query is unrateable. This means that it, and the AC
and LP paired with it, are not eligible to be assigned ratings. A Search Ads query is
unrateable if it has one of the following problems:
it is in a language other than the task
language (Foreign Language)
it is unambiguously pornographic or
about sexual services (Porn)
it is complete nonsense; research
reveals no plausible meaning
(Nonsense)
it was transcribed incorrectly, using an
English rather than Cyrillic keyboard for
Russian words (Russian Transcription
Error)
If you use one of these flags, all of the later questions will turn gray and don’t need to
be answered. Note that these flags are only on the first page of a Search Ads task,
the Ad Creative rating section.
Foreign Language Query
Use this flag when the query is in a language other than the language of the task. If
the query contains words or phrases in another language, but there is enough
content in the task’s language that it is understandable, do not use this rating. If the
query appears to be in a foreign language, but research reveals that the query term
may be commonly used in your rating language or is the name of a specific group,
product, or person, do not use this rating.
Remember to check the language of the task, especially if you work in multiple
languages. If your rating language is English, you rate ads in English for English
queries. If you rate in another language, you will rate some tasks in that language
and some tasks in English. Your rating language is always designated at the top of
the task page. Even if you speak the language of the query, if the task is supposed
to be for a different language, use this rating.
Porn Query
Use this flag only for queries that are unambiguously for pornographic content or
sexual services. Queries for racy or suggestive content, medical information, or art
photos generally shouldn’t get this rating. Queries for dating services generally
shouldn’t get this rating, unless those dating services depict nudity or specifically
identify themselves as sexual rendezvous services.
Nonsense Query
Use this flag for queries that are complete nonsense, where research reveals no
plausible meaning. As you research, take into consideration that queries that may
look like nonsense might actually turn out to be meaningful. The following are
examples of queries that do have meaning and should not receive the Nonsense
Query flag:
a misspelling
a product code or model number
technical specifications
a partial web address or YouTube video
ID
a specific username or Twitter handle
an uncommon acronym or abbreviation
Don’t assume that a query is nonsensical just because you do not immediately know
what it means. Encountering a completely nonsensical query is rare. Most queries
mean something, so you should always research the query, even if at first it seems
like nonsense. Only use this rating when there is no way for you to reasonably guess
about user intent, even after researching the query.
Ad Creative Flags
If an Ad Creative meets the criteria for using one of the following flags, please use
that flag. If criteria are not met for a flag, do not use the flag.
Navigational Bullseye
Use the Navigational Bullseye flag when both these things are true:
1. The query appears to be a search for a particular website, section of a website, or
web page.
2. The creative looks like it will point to the corresponding website, section of a
website, or web page.
Not every query is a search for a particular website--in fact, the vast majority are not.
The Navigational Bullseye flag should only be used where the frame of reference is
similar or compatible between query and creative. For example, with the query,
[ ford explorer ], the Navigational Bullseye would be used for creative that appears to
take the user to the Ford Explorer section of the Ford website; however, the flag
would not be used if the creative appeared to take the user to a different page on the
Ford site (a page devoted to the Ford Focus) or a general page on the Ford site
(their homepage, for example).
Foreign Language
Use this flag when the creative is in a language other than the language of the task.
Remember to check the language of the task, especially if you work in multiple
languages. Even if you speak the language of the creative, if the task is supposed to
be for a different language, use this flag.
A creative should be legible in your rating language: if the creative contains words or
phrases in another language, but there is enough content in the task’s language that
it is understandable, do not use this flag and proceed with the normal creative rating.
If you use this flag, some of the later questions will turn gray and don’t need to be
answered.
Unexpected Porn
Use this flag when both these things are true:
1. The query is not a search for pornographic content or sexual services. If the query
has both a pornographic interpretation and a non-pornographic interpretation,
assume that the non-pornographic interpretation is the actual user intent.
2. The creative appears to offer pornographic content or sexual services.
Use this flag only for unambiguously pornographic content or sexual services. Racy
or suggestive content with no nudity, nudity in a medical context, or art photos
generally shouldn’t get this flag. Dating services generally shouldn’t get this flag
unless they depict nudity or specifically identify themselves as sexual rendezvous
services. A regular dating service may deserve a bad rating if it doesn’t match what
the query appears to be looking for, but it would not get the flag.
If you use this flag, some of the later questions will turn gray and don’t need to be
answered.
If a Landing Page meets the criteria for using one of the following flags, please use
that flag. If criteria are not met for a flag, do not use the flag.
Navigational Bullseye
Use the Navigational Bullseye flag when both these things are true:
1. The query appears to be a search for a particular website.
2. The landing page is that site.
Not every query is a search for a particular website--in fact, the vast majority are not.
Foreign Language
Use this flag when the landing page is in a language other than the language of the
task, with no obvious way of getting to a version in the language of the task.
Remember to check the language of the task, especially if you work in multiple
languages. Even if you speak the language of the page, if the task is supposed to be
for a different language, use this flag.
Don’t use this flag if there is some clear way to get to a version in the target
language. For example, if you are rating a Japanese task, a landing page in English
with a Japanese flag in the corner pointing to a Japanese version of the site should
not get this flag.
If you use this flag, some of the later questions will turn gray and don’t need to be
answered.
Unexpected Porn
Use this flag when both these things are true:
1. The query is not a search for pornographic content or sexual services. If the query
has both a pornographic interpretation and a non-pornographic interpretation,
assume that the non-pornographic interpretation is the actual user intent.
2. The landing page offers pornographic content or sexual services.
Use this flag only for unambiguously pornographic content or sexual services. Racy
content with no nudity, nudity in a medical context, or art photos generally shouldn’t
get this flag. Dating services generally don’t get this flag unless they depict nudity or
specifically identify themselves as sexual rendezvous services. A page with racy
content, nudity in an art or medical context, or dating services may deserve a
negative rating if it doesn’t match what the query appears to be looking for, but it
shouldn’t get the flag.
If you use this flag, some of the later questions will turn gray and don’t need to be
answered.
Unexpected Download
Use this flag when any of the following happens:
1. Clicking on the Visit Landing Page button initiates an attempt to download a file.
2. Some link, button, or graphic on the landing page initiates a download when
clicked, but does not clearly indicate that it will do so. For example, a big red button
that says “Enter site” or “Check the weather,” but starts a download when clicked,
deserves the flag. A similar button that says “Get It Now” or “Click here to download”
does not.
Never install downloads that a site tries to initiate in this way: it is not part of the
rating process.
If you use this flag, some of the later questions will turn gray and don’t need to be
answered.
If you have questions about this project that are not answered by the instructions
above, please review the Rater Hub, which contains additional content about Search
Ads rating. If you encounter a technical problem with this rating task, use the
“Report A Problem" link in the lower-right hand corner of the rating page.
Task 1 Screenshot:
This task has short instructions at the top of the task, and 14 questions.
Each question has an instructions section with examples. Each instruction
section is shown below its relevant question. At the end, you will see a few
screenshots of the actual task.
Version: 2017-04-24
If you can only think of a single thing that the user might be looking to do,
then the user intent should be "Very Clear". On the other hand, if you
can think of a few distinct things (2 - 5 or so) that might be valid user
intents, then the rating should be "Somewhat clear". If there can be so
many possible user intents that it is impossible to enumerate them, then
the rating should be "Very unclear".
Query: [cash]
Query: [www]
Somewh
at clear There can be a few Query: [chocolate chip cookies]
(2- 5 or so) It is a bit unclear what the user is
reasonable distinct looking for. The user might be looking
user intents. for recipes on how to make chocolate
chip cookies. Or the user might be
looking for online stores to purchase
chocolate chip cookies. It's possible
that the user is looking for information
about different kinds of chocolate chip
cookies. Since there can be a few
different valid user intents, the user's
intent is not very clear.
What is a concept?
In this question, you will have to assess whether the scope of the ad is the
same as the scope of the query and if they differ, in what way they
differ. Consider the important concepts that are associated with the query,
and those that are associated with the ad. When you compare scope, you
look at the overlap between these groups.
Partial
overla There is some Query: [new american university chemistry
p degree of building]
overlap
between the
scopes of the
ad and the
query, but
their scopes The query and the ad overlap to the extent that
cover they are both about universities, but they are
fundamentall about fundamentally different concepts. The
y different query is about the Chemistry building within New
sets of topics American University and the user is likely looking
and one is not for its location, but the ad is for degree programs
contained at the University of Phoenix.
within the
other.
Query: [bases for beds]
Query: [recipes]
Ad
Narro The range of Query: [jewellery macys ]
wer concepts
covered by
the ad is
narrower than
those covered
by the query.
Ad
Broad The range of Query: [Women's sweaters JCPenny]
er concepts cove
red by the
ad is broader
than those
covered by
the query. The query is specifically about women's sweaters
at JCPenny but the ad is about all sweaters. So,
the scope of the ad is broader than that of the
query.
Same
The range of Query: [amazon.com books]
concepts
represented
by the ad and
query are the
same.
The ad is looking for books on amazon.com and
The ad does the ad is for books on amazon.com. The ad also
not cover any describes other concepts like "Amazon Prime",
concepts not "Deals" etc., but the important concepts for both
present in the the ad and the query are the same.
query and
vice-versa
Query: [Women's sweaters JCPenny]
Sometimes the same or similar phrases can refer to two entirely different
concepts. Determine the concept that the user is likely referring to and
then assess whether the ad targets a completely different concept with
the same or similar spelling or pronunciation.
Rat Description Examples
ing
Yes
The Query: [paris]
ad misinterprets t
he concept the user
specified in the
query and targets
a completely diff
erent concept with The ad is for the purchase of plaster of paris
a similar spelling or which is very likely an incorrect interpretation
pronunciation. of the query. The user probably want to see
information about Paris, France.
Query: [transformers 2]
Think about all the concepts and requirements the user has specified in
the query. Now for each requirement, ask yourself how likely the user is to
consider an alternative. If it is very likely that the user will consider an
alternative, then the requirement is non-critical. If missing this
requirement or providing an alternative would be unacceptable to most
users, then it is critical. If an ad drops both a critical and a non-critical
requirement select "Misses critical requirement" since that leads to
worse ad experience for the user.
Misses The ad
critical misses Query: [hair laser treatment growth]
require one or
ment more
critical
requirem
ents that
the user
The user is looking for laser treatment for
specified
hair growth. But the ad is showing treatments
in the
for hair removal. The ad completely missed the
query.
key requirement of hair growth instead of hair
removal from the query and it is very unlikely
that the user can be swayed to accept this ad as
satisfying their intent.
Misses The ad
non- misses Query: [leather sofas]
critical one or
require more
ment non-
critical
requirem
ents that
the user The user is searching for leather sofas but the ad
specified is for fabric sofas. While this misses the "leather"
in the requirement for the sofa, it is likely that the user
query. might also be willing to consider fabric sofas.
Doesn't The ad
miss does not Query: [nike men's shoes]
any miss any
require requirem
ments ents that
the user
specified
in the
query.
No
The ad Query: [amazon.com books]
text is not
awkwardl
y phrased
or difficult
to make
sense of. The ad creative is not nonsensical or ungrammatical.
Same
The Query: [plumbing service venice]
product/serv
ice or
information
advertised
in the ad is
exactly for
what is
specified in The ad is exactly for what the query specified.
the query.
Question 7: How good is the ad relative to the top 3 search
results?
In this section, we want to know how good the ad is relative to the search
results. Look at the top 3-5 search results to form your judgment. Please
ignore Product Listing Ads (the sliding panel of ads at the top of
the results page) when making your decision. Below, the panel of
ads displaying different Nikon digital cameras are Product Listing Ads.
Ad The Ad is
bett likely to Query: [new egg promo]
er be more
than relevant
Sear to the
ch user
than the
search
results.
Search results:
The search results are from some other websites while
the ad is from newegg.com. So, the ad is likely to be
more relevant to the user than the search results.
In this section, we want to determine if there are any serious issues that
make it infeasible to satisfy the user's intent. Select all the options that
apply.
Languag
e Parts of the landing Query: [hotels paris]
mismatc page are in a foreign
h language. Landing page: While large parts of
the LP are in English, the hotel
descriptions within the LP are in
French. This affects the user's ability
to understand content on the page
relevant to their intent.
Out of The product / service
stock the user is looking for is Query: [NIKE FREE RN FLYKNIT 2017]
out of stock / no longer
available. Landing page: The LP has other Nike
shoes available but not the model
specified in the query. The image /
listing for this specific model is
marked as Out of Stock.
Question 9: Does the LP have any issues that affect the user's
interaction with the page? Select all that apply.
Think about the user's interaction with the landing page when they visit it.
Do they have any negative experiences on any of the following
dimensions? Select all that apply.
Rating Description
Cluttere The LP is cluttered. Text and images are not well formatted
d and the LP is not aesthetically appealing. It is not clear
whether the information the user is looking for is on the page
and the user has to make an effort to locate it.
Interstiti
al / form When the user visits the LP, they first see an interstitial
(content that prevents the user from accessing the LP) or
a form that is difficult to dismiss or requires extra steps. The
screenshot below shows an example of an interstitial.
Distracti The LP has ads which attract the user's attention away from
ng ads the main content. For example, flashing ads or too many ads
or ads which are shocking and surprising.
Prods The LP prods the user to stay / makes it difficult for them to
the user leave the page through repeated messages or alerts.
to stay
Bad The LP consists of content which has bad grammar or
grammar awkward phrasing
Question 10: Does the LP have any issues that might undermine
the trust or safety of the user? Select all that apply.
Pretends
to be The LP falsely claims Query: [tickets LA Opera]
official to be the official
page when it's not, Landing page: The LP falsely claims
or has a disclaimer that they are the official site to buy
on the page but the tickets to the Los Angeles Opera, but
disclaimer is hard to it is not. The official site is
find. actually that of the LA opera.
What is a concept?
For landing pages, this means that you should focus on the main content
of the webpage and not on concepts introduced by interstitials, web
forms, product reviews on the landing page or comments by users on
content on the web page. For example, consider the landing
page https://www.amazon.com/s?field-keywords=Nike+shoes which has
Nike shoes. The main concept that should be considered is that of Nike
Shoes. The other content such as product ratings, comments and reviews,
other ads on the page, links to other products etc. aren't very important.
In this question, you will have to assess whether the scope of the landing
page is the same as the scope of the query and if they differ, in what way
they differ. Consider the important concepts that are associated with the
query, and those that are associated with the landing page. When you
compare scope, look at the overlap between these groups. You
answered this question for the ad creative above, but the answer
need not be the same for the landing page.
LP
narrower The range of concepts Query: [Nike Men's shoes]
covered by the LP is
narrower than those Landing page: The LP points to a
covered by the query. page with nike.com and contains
information about returning
purchased nike shoes.
Same
The range of concept(s) Query: [Nike Men's shoes]
covered by the LP are
exactly the same as Landing page: The LP contains only
those covered by the nike men's shoes. It doesn't
query. contain shoes by other companies
or shoes for women and children.
The LP doesn't cover any
concepts that the query
doesn't cover and vice- Query: [Plumbing services]
versa.
Landing page: LP points to the
home page for a company offering
plumbing services.
Think about all the concepts and requirements the user has specified in
the query. Now for each requirement, ask yourself how likely the user is to
consider an alternative. If it is very likely that the user will consider an
alternative, then the requirement is non-critical. If missing this
requirement or providing an alternative would be unacceptable to most
users, then it is critical. If the landing page drops both a critical and a
non-critical requirement select "Misses critical requirement" since that
leads to a worse ad experience for the user. You answered this
question for the ad creative above, but the answer need not be
the same for the landing page.
Question 14: Does the LP help the user make progress towards
their goal?
Imagine that the user has entered a query into Google, and is presented
with a page of search results (SERP, Search Results Page) and some ads,
including the ad being rated in this task. The user then clicks on this ad
and visits the landing page. Think about whether the user is substantially
closer to achieving their goal than they were before clicking on the ad.
(i.e. on the google search results page).
Please click on the query link provided above and look at the Google
search results before answering this question. When you are considering
the Google search results page, please ignore any other ads from this
same advertiser (e.g. Product Listing Ads, these are the “Shop for X”
pictures that sometimes show on top of the page).
It would take
the user the
same amount
of effort to get
to their goal
from the
Google search
results page
that it would
take to get
there from this
page.
After clicking LP:
Visiting this LP takes the user neither closer to,
nor further from, their task, since most of the
information on it was already available to the
user pre-click. It adds some new information (on
the right), but the new information added by the
page does not directly help the user in their task.
Some
progr Visiting the LP Query: [Students part time job Palo Alto]
ess takes the user
somewhat Before clicking LP:
closer to their
goal, i.e. it
offers some to
moderate
additional
information,
functionality or
variety that
helps the user
in their task,
but the user
has to take
additional
steps (say, use
a menu to
navigate to a
different part
of the website
etc.) to find
what they
were looking
for.
It would take
the user a little
more effort to
get to their
goal from the
Google search
results page
than it would
take to get
there from this
page.
Subs
tanti Visiting the LP Query: [buy cannon 70d]
al helps the user
progr make Before clicking:
ess significant
progress
towards their
goal,
compared to
the Google
search results
page i.e. it
offers
substantial
additional
information,
functionality,
or variety that
help the user
accomplish
their task,
beyond what
was already
available on
the Google
search results
page.
It would take
the user more
effort to get to
their goal from
the Google
search results
page than it After clicking (LP):
would take to
get there from
this page. (For
ex. the
product/servic
e/information
that the user is
looking for is
directly
available on
the current
LP).
Task 2 Screenshots
Version: 2016-11-08
In this task, you will research a mobile app and a query entered in Google
Search. You will then answer a question about how well the app fulfills
what the user is looking for.
Terms
You will see a query that a user entered. Just as in normal search ads
rating, it will be your responsibility to do your best to understand what the
user was looking for. You will be asked to assume that the user
entered the query in a regular search on Google.com on a mobile
device. Although this evaluation is concerning the relevance of mobile
apps, do not assume that if a user entered a search into Google that they
are searching for an app. Instead, consider the user to be an average
person entering the query into Google search. For more information on
how this is applicable to the task at hand, please see examples in the
table below Step 3.
Unrateable Query
Step 2: Familiarize yourself with the mobile app.
Click the button labeled “research mobile app” to open the app’s Google
Play page, then review its content. You should consider all the information
available to you in order to inform your understanding of the app,
including the text that describes the app, ratings, any relevant comments
from users, and even your overall impression of how professional and
trustworthy the app looks. Some additional notes:
Unrateable Apps
If you've personally installed and used the app, or believe you have
additional knowledge about the app that would affect your ratings in this
task, select the flag labeled "I am personally familiar with this app." You
will still be required to complete the task.
Think about what the mobile app is used for and what the user was
looking for. For example, if the app’s purpose is to deliver food and the
provided query contains the word “fruit ninja”, then the user intent is
unlikely to be satisfied by the user installing and using the app. Pay
attention to the app’s quality as well. A good quality app looks
trustworthy, has a well-written description that makes the app’s purpose
clear, and gets positive reviews from users. The user intent is less likely to
be satisfied by a poor quality app.
Taking all of these factors into consideration, select a rating from [-100] to
[100] to indicate the likelihood that a specific user would be satisfied by
installing and using the app.
User query: [airline]
App: Kayak
User query: [facebook]
App: Facebook
App: Snapchat
App: Vet Clinic
If you have questions about this project that are not answered by the
instructions above, please review the Rater Hub first, which contains
additional content about rating. If you encounter a technical problem with
this rating task, use the “Report A Problem" link in the lower-right hand
corner of the rating page.
Task 3.1 Screenshot
Version: 2017-01-19
In this task, you will research a mobile app and a query entered on the
Google Play Store. You will then answer a question about how well the app
fulfills what the user is looking for.
Terms
You will see a query that a user entered. Just as in normal search ads
rating, it will be your responsibility to do your best to understand what the
user was looking for. However, unlike the normal search ads rating, it is
important to keep in mind that the query was entered on the Google Play
store. Assume that if a user entered a search into the Google Play
store, that they are searching for apps.
Click the query to open the Google Play store results page, and research
to understand what the user was looking for. You may perform additional
research outside of the Play store well. For example, if the user searches
for [watch mr robot], they are probably looking for a way to watch the USA
Network TV show Mr. Robot, even though the apps that appear in the Play
Store search results are mostly for robot games.
Unrateable Query
Click the button labeled “research mobile app” to open the app’s Google
Play page, then review its content. You should consider all the information
available to you in order to inform your understanding of the app,
including the text that describes the app, ratings, any relevant comments
from users, and even your overall impression of how professional and
trustworthy the app looks. Some additional notes:
Do not download the app.
The Play Store page may include
some additional content, such as
links to other parts of the Play
Store, or lists of “Similar” mobile
apps. Please ignore this and only
focus on the content that is related
to the main app on the page.
You may, however, follow links and
do additional research outside of
the Play Store if it helps you
understand the app.
Unrateable Apps
If you've personally installed and used the app, or believe you have
additional knowledge about the app that would affect your ratings in this
task, select the flag labeled "I am personally familiar with this app." You
will still be required to complete the task.
Think about what the mobile app is used for and what the user was
looking for. For example, if the app’s purpose is to deliver food and the
provided query contains the word “fruit ninja”, then the user intent is
unlikely to be satisfied by the user installing and using the app. Pay
attention to the app’s quality as well. A good quality app looks
trustworthy, has a well-written description that makes the app’s purpose
clear, and gets positive reviews from users. The user intent is less likely
be satisfied by a poor quality app.
Taking all of these factors into consideration, select a rating from [-100] to
[100] to indicate the likelihood that a specific user would be satisfied by
installing and using the app.
Rating Description Examples
App: Snood Free
App: DoorDash
App: Kayak
User query: [facebook]
App: Snapchat
User query: [youtube]
App: YouTube Music
User query: [taxi]
App: Taxis of Houston
App: Hangouts
App: Spotify
If you have questions about this project that are not answered by the
instructions above, please review the Rater Hub first, which contains
additional content about rating. If you encounter a technical problem with
this rating task, use the “Report A Problem" link in the lower-right hand
corner of the rating page.
Task 3.2 Screenshot
Version: 2017-04-04
In this task, you will research a mobile app and a query entered in
YouTube. You will then answer a question about how well the app fulfills
what the user is looking for.
Terms
You will see a query that a user entered. Just as in normal search ads
rating, it will be your responsibility to do your best to understand what the
user was looking for. However, unlike the normal search ads rating, it is
important to keep in mind that the query was entered on YouTube.
Click the query to open the YouTube results page, and research to
understand what the user was looking for. You may perform additional
research outside of the YouTube as well.
Unrateable Query
Unrateable Apps
If you've personally installed and used the app, or believe you have
additional knowledge about the app that would affect your ratings in this
task, select the flag labeled "I am personally familiar with this app." You
will still be required to complete the task.
Think about what the mobile app is used for and what the user was
looking for. There will inherently be a mismatch between searching on a
video site and what apps can offer. For example, if the user is searching
for a [baked alaska recipe], the user wants to see video tutorials on
making Baked Alaska. An app of dessert recipes would probably be
Satisfaction Possible/Satisfaction Likely, whether or not it includes videos.
Pay attention to the app’s quality as well. A good quality app looks
trustworthy, has a well-written description that makes the app’s purpose
clear, and gets positive reviews from users. The user intent is less likely
be satisfied by a poor quality app.
Taking all of these factors into consideration, select a rating from [-100] to
[100] to indicate the likelihood that a specific user would be satisfied by
installing and using the app.
User query: [formation
beyonce]
App: Spotify
App: Duolingo
App: Lose It!
App: MindBody
The user is looking for a yoga
class to follow along with, and
the app is a yoga class
reservation system. While the
yoga reservation system can
help the user find a yoga class,
only a small subset of users
would be satisfied with this
app.
App: Makeup Genius
If you have questions about this project that are not answered by the
instructions above, please review the Rater Hub first, which contains
additional content about rating. If you encounter a technical problem with
this rating task, use the “Report A Problem" link in the lower-right hand
corner of the rating page.
Task 3.3 Screenshot
The Task
Open the user’s current page using the Open cached page HTML button.
You should review the textual content on the cached page until you have
a sense of the main ideas and concepts on it.
Because you are reviewing the cached page HTML, the user’s page will
have most of the non-textual content stripped out. So, the layout may be
affected, and you may have to scroll to thoroughly evaluate the page.
You may also see an Open cached page image button. You can use the
cached page image for context or in order to help you understand how the
page is intended to be displayed, which will help you find the main
content. However, you should base your evaluation of the page on the
cached page HTML.
Boilerplate text like the navigation menu and any text not related to the
main content should be considered as contributing to minor topics of the
page.
If the main content of the content page is an image or video, please look
for textual information associated with the image or video that explains
what it is about. This text could be a title, caption, or short description. Do
not consider comments when viewing YouTube pages.
NOTE: If a cached page is not available for this task, please use the live
page for rating. There will be a note below the live page button if this is
the case.
First, select one of the following classifiers that best describes the page’s
content:
A news article
Topical The main content of the page is A sports forum
Content primarily focused on a single topic or a A blog mostly
small number of topics. Additional
about cooking
secondary, minor, or sidebar topics are
okay, as long as the page is mostly
about one or a few topics.
A portal page
Diverse The main content of the page is The homepage
Content divided among several or many of a newspaper
different topics. A list of links or
articles that are
not focused
around a single
subject area
A list of
products for sale
that are not
focused
Song lyrics
Fictional The main content of the page consists Excerpts from a
Content of fictional (non-real) material. novel
Fan fiction
Video game
walk-through
Videos
Low- The main content of the page contains Images with
Text very little or no text. minimal captions
Content Comics
Now that you have thoroughly reviewed the user’s current page, you can
develop a profile of the user’s current interests. Think about the types of
products, services, ideas, and activities a user who is visiting this page is
seeking or might be interested in (i.e. current content of the page, as well
as anything related to the content on the page).
Now that you have a good understanding of the keyword’s scope and
meanings, recall your profile of the user’s current interests. If the keyword
contains a specific location, you can assume the user belongs to that
location:
Very Likely
Somewhat Likely
Unlikely
Very Unlikely
Task 4 Screenshot
The Task
Business Assessment
(2) Who purchases these products or services (i.e. who are the customers
of this business)?
Think carefully about the customer base. Consider whether the business
appeals to a broad set of customers, or a very narrow set (such as a
business-to-business seller). Consider also whether customers need to
spend a lot of money to purchase these products or services, or whether
the prices are more broadly accessible.
You may or may not see a list of the top search queries for the business
page, depending on whether this data is available. These are typical
search queries by which users have chosen to navigate to the business
page. You can use these to supplement your understanding of the
business, if doing so is helpful.
User Assessment
Next, visit a user’s current webpage. You should develop a profile of the
user’s current interests and state of mind by reviewing the page the user
is visiting. Try to think about the types of products, services, and activities
a user who is currently visiting this page is seeking or might be interested
in. Think about the user’s state of mind and the various things they are
interested in, and even the kinds of things they might become interested
in impulsively while in this current state of mind.
Now that you have a good understanding of the user’s current interests
and state of mind, recall your profile of the business’ customer base and
its products or services. If the business is targeting customers in a specific
location, you can assume the user belongs to that location:
Likely
There are good reasons to believe the user is researching the type of
products or services offered by the business. The user appears to match
the type of customer targeted by the business.
Possible
Unlikely
Task 5 Screenshot
You should research the keyword and query to form an opinion about
what the advertiser and user meant by them, and what each may have
been looking for.
Task Description
In each task, you will be presented with a keyword and a query. You will
be asked to decide what these terms mean, and how their user and
advertiser intents are related to each other. We want to know whether
advertisements that are relevant to the keyword will also be relevant to
the query.
When you are ready, you will be asked to define how the intent of the
keyword and the intent of the query are related.
If you are not able to answer this question because the keyword or query
is unclear, ungrammatical, missing or in a foreign language, select the
'unrateable' checkbox next to the keyword and query. You will then be
asked to identify whether the problem lies with the keyword, the query, or
both.
Your task is to determine how the keyword and query are related. There
are five options:
Below you will find additional guidance and specific examples that will
help you correctly categorize tricky and unclear cases.
Select this option if the keyword and the query have the same intent. For
this task, this does not mean that the terms need to contain all the same
words or be phrased in the same way: as long at the keyword and
query refer to the same product, service, or concept they should
be rated as having the same intent.
You should pick this option if you think the keyword and query pick out
the same things as depicted in the picture below. The query, fruit, and
the keyword, fruit, pick out the same set of items.
Not every pair you rate will be as easy as the pair above. Below we review
several common types of Query-Keyword pairs that have the same intent.
Sometimes you will encounter a mix of the cases covered below. In this
case, you should still select this option.
1) Synonyms: the keyword and query express the same intent
using different words, for example, synonyms or paraphrases.
In the table below are some examples of keywords and queries that have
very similar search intents. These keywords and queries target the same
information even though they are not exact paraphrases of each other.
Keyword Query
male infertility low sperm how to get pregnant with low sperm
count count
romantic music online love songs online playlist
global map turkey turkey's location on the world map
Note: You can usually tell that the keyword and query express the
same search intent by checking the actual search results for
each. However, you should be careful since sometimes queries
and keywords that have different search intents show similar
search results. For example, 'flights la ny' shows similar results to
'flights ny la' even though the search intents are different.
In the table below you can see keywords or queries that should be rated
as having the same intent even though the keyword or query has
additional words that are not included in the other.
Keyword Query
buy pro tools 11
pro tools 11 download
download
where to
rx bars
purchase rx bars
hyundai accent hyundai accent price
adidas shop adidas
Note: if the keyword or query expresses a specific price point, for
example "cheap" or "discount", that is not present in the other
you should not select this option. For example the keyword or query
"handbags" includes expensive handbags like Louis Vuitton handbags, but
the keyword or query "discount handbags" would exclude expensive
handbags. Therefore, you should treat "discount handbags" as more
specific than "handbags".
Below you will find some additional tricky examples that seem like they
might belong in this category, but don't.
This option applies if the query adds a more specific aspect (for example,
a product sub-type, location, or brand) to the keyword, but preserves the
main intent of the keyword.
You should pick this option if you think the query picks out a narrower set
of things than the keyword as depicted in the picture below. The
query, apples, picks out a narrow set of items than the keyword, fruit.
Again, the case depicted above is simpler than many cases you will see.
Below we review several common types of queries and keyword that
should fall in this category.
Keyword Query
aftermarket auto cheap after market auto
parts parts
wineries in long best long island
island vineyards
lunch recipes easy lunch recipes
buy printer color printer
micro sd 64gb micro sd card
buying coffee
coffee beans
beans online
play games play games for free
Keyword Query
child do i have enough to gain full child
custody custody
what do you do for fleas on a 2 week old
puppy fleas
puppy
Examples that don't fall into this option (query more specific).
Below you will find some additional tricky examples that seem like they
might belong in this category, but don't.
Keyword Query Correct Rating Explanation
Although one could complete the
keyword with "chromecast" to get
set the query, there is no way to know
The keyword is
set up up chromec that the advertiser intended this.
unrateable.
ast We can't know what the advertiser
meant because the keyword is
incomplete.
Although, the query only adds
The keyword words to the keyword, the query
cast
cast iron and query are shifts the intent from a product
iron repair
related. (cast iron) to a service (repair)
related to the product.
Although the query adds words to
The keyword
2015 the keyword, the additional words
2015 and query have
truck for only specify a commercial intent,
truck the same
sale which can be assumed in the
intent.
keyword.
The query "target" picks out a
The keyword
merchant that sells many products,
and query have
laundry including laundry baskets.
target related but
basket However, since Target does not
different
primarily sell laundry baskets, you
intents.
should chose Related.
You should select this option if the query picks out a broader set of things
than the keyword as depicted in the picture below. The keyword, apples,
picks out fewer items than the query, fruit.
When rating, you should consider the same aspects of the keyword and
query that you would take into account when deciding whether the query
is more specific than the keyword (option 2) above.
If the keyword does not preserve the main intent of the query, do not
choose this option: instead mark the pair as having related but different
intents (Option 4) or unrelated intents (Option 5).
This is the case for example if the keyword and query are about related
but different products, or if the query is at the same time more and less
specific than the keyword.
One way for a keyword and query to be related is if they pick out
overlapping sets of items. For example, the query, cats, and
keyword, happy animals, pick out different but overlapping items in
the picture below.
Below we review several types of queries and keywords that should fall
into this category.
1) Different Products: the keyword and query specify related but
different products, services, problems, or topics.
Note: If the query shows that the user's main intent is focused on
the product in the keyword itself, then the query is more specific
than the keyword (option 2). For example, if the user is looking
for reviews of a product or service, wants to compare the product
or service to other alternatives, or is looking for coupons,
discounts, or delivery of the product, the query is more specific
than the keyword.
4) More and Less Specific: the keyword is at the same time more
and less specific than the query.
The instructions for Option 1 (same intent) discussed cases in which the
keyword and query can use different words or different orders of words to
express the same intent, but sometimes you will encounter keywords and
queries that use the same or similar words, but express different intents.
This is often the case when the order of the words in the keyword or query
imply a direction.
Below are some examples of keywords and queries that are unrelated.
Keyword Query
lord of the rings wedding rings
las vegas entertainment stock option
options quotes
international year of the
the international
potato
Note: Porn-related terms are expected for this task, and should
be rated normally.
Keep in mind that not all keywords or queries that seem nonsensical
actually are. It is important to research the keyword and query to make
sure that you are not misunderstanding its intent.
If you encounter a technical problem with this rating task, use the "Report
A Problem" link in the lower-right hand corner of the rating page.
Task 6 Screenshot
1. Project Overview
In this project you will review webpages and decide if they are related to
the topic of Religion. The purpose of this work is to collect data for
building classifiers which can automatically detect such content,
ultimately allowing advertisers to prevent their ads from appearing on
such pages.
2. Process
Please scroll through the page, if necessary, until you have a strong sense
for what the page is about.
The default assumption is that pages are rateable. However, you may
come across unrateable content, which you can flag by ticking the
following checkbox:
Cached pages often show imperfect formatting and may even be missing
images, but please still do your best to complete the task, and only tick
the checkbox if the page:
Please note: when considering the language of the page, don't limit
yourself to the perspective of users in any one country. E.g., if you are
working on Portuguese tasks and you imagine the typical viewers of the
page are from Portugal, or from Brazil, try to consider the task from their
perspective (as best you can).
If you mark the page as unrateable, please leave a brief explanatory
note in the comment box.
If the page is rateable, please assess whether the page is related to the
topic of Religion (see Section "Detailed Guidelines" below for guidance),
and select the most suitable label from the following four options:
3. Detailed Guidelines
religious places
o places of worship (e.g.
churches, temples)
[link]
o holy sites [link]
religious hobbies, special
occasions and events
o religious festivals and
holidays (e.g. passover,
hanukkah) [link]
o religious gifts and
merchandise
[link] [link]
Other religious content
o news stories related to
religious institutions or
clergy [link]
o religious dating and
social networking sites
[link]
non-religious spirituality
o new age, inspirational
and self-help [link]
o astrology and divination
[link] [link]
o occult and paranormal
practices [link]
o yoga practice [link]
o neutral meditation (e.g.
acme meditation) [link]
o feng shui [link]
religious holidays
mentioned or discussed in
a non religious context
o top Christmas gifts
[link]
o Easter egg arts & crafts
[link]
o St. Patrick's day pub
crawls [link]
o Hannukah gifts for
everyone [link]
o Pastry recipe for Easter
& Eid [link]
Other non-religious
content
o mythology [link]
o psychological tests
[link]
Borderline cases
It is likely that you will encounter borderline cases where you are unsure
of the page’s membership in the category, and where you could easily
imagine other reasonable people disagreeing with your decision. To allow
you to indicate that certain pages are not obviously either “Definitely
Religion” or “Definitely NOT Religion”, we provide the “borderline”
labels “Probably Religion” and “Probably NOT Religion”. You will
need to use your gut instinct when deciding which of these to use, but
still try to follow the spirit of the guidelines by carefully considering which
side of the spectrum would be more suitable for the given page in the
target language/countries under consideration. It may also help to
consider a hypothetical advertiser who has opted out of having their ads
appear on pages with content about Religion. Would that advertiser be
likely to consider the page you’re rating as too related to Religion to show
their ads on it? If so, you should label the page as “Probably Religion”
rather than “Probably NOT Religion”.
Task 7 Screenshot