Mesure de la Brand Safety

Mesure de la Brand Safety et
de la Brand Suitability

Le CESP souligne que ces grilles sont basées sur la confiance accordée par le CESP à l’égard des informations fournies par les mesureurs.

Pour naviguer dans un tableau :
– sur ordinateur : cliquez sur une de ses cellules et naviguez horizontalement avec les touches de direction du clavier ou utilisez la barre de scroll horizontal qui se situe en bas du tableau.
– sur mobile : nous vous recommandons de passer en format paysage pour un meilleur affichage.

Vous pouvez télécharger la grille que vous souhaitez au format Excel tout en filtrant les mesureurs.

Les mesureurs sont répartis dans plusieurs onglets en fonction de leur accréditation ou non selon les standards IAB 2012 (aucune solution n’est accréditée par les standards IAB 2018).

Environnement Web

Solutions accréditées - IAB 2012


Updated February 2024
Double Verify
Double Verify
Integral Ad Science
Moat Oracle
1) At what granularity level does your solution perform content analysis for Brand Safety? (calculation rules for scoring) 
a) domainYesYesYes
b) subdomainYesYesYes
c) pageYesYesYes
c-1) If yes, please explain how your page level control works? (query analysis, words on the page, meaning detection...)?DV uses proprietary technology as well as peer review to classify pages with display and video assets. DV technology is able to see through to the URL and collects data points such as incoming and outgoing links, text, context and the frequency and prominence of words and concepts. DV also analyzes a site and page's HTML code and structure, along with domain registration records and articles about the site in prominent online reference and news sites. In addition, to ensure we are accurately reflecting the dynamic nature of online content, we frequently reassess sites to ensure we have the most current classification and we maintain records of previous content to create a model of a site and page's classification not just on a single moment, but over time. DV manually reviews high-profile sites to ensure the most accurate representation of the sites.
Once a page is classified, DV javascript tag will call up advertiser brand safety settings to understand if an impression is deemed as suitable.
IAS crawls pages for page content (URL, page metada, entry links, external links and feed information) and carry out context analysis, semantic analysis, sentiment analysisMoat solution is based on the Grapeshot technology that crawls and contextually analyzes the text of pages (video speech are converting to text). Algorithms are used to identify the relative “weight” of all words within text.
If a page has been modified since the last time it was crawled, then the crawling frequency is halved, to a lower limit of every 60 minutes; if it hasn’t been modified, then the crawling frequency is doubled, to a maximum of every 30 days. It is beholden on any partners using this service to maintain a high cadence of requests/re-requests of any URLs to ensure that the crawling process is prompted and executed.
2) Which are the limitations in running the Brand Safety analysis (elements which explain that part of the page contents will not be accessible / limited access to the data)Please see below for specific instances.NoneAs browsers have gradually restricted page URL data to third-party tags, including those of verification vendors including Moat, the percentage of impressions in which we are able to verify the exact page URL of the impression has decreased. This reduces transparency in the ad tech ecosystem and results in potential brand safety risks, as verification providers cannot definitively classify the environment surrounding an ad impression without knowing the identity of the page.

Moat's metrics shed light on this development, breaking impressions into two groups: those with page-level brand safety measurability and those where only the domain was measurable. In the latter case, we calculate the average volume-weighted safety of that domain based on historical classifications and report on brand safety using the Domain Safe Rate and Domain Unsafe Rate metrics. For these impressions, we also do not increment the individual brand safety category metrics (e.g. Unsafe Rate - Crime, Unsafe Rate - Terrorism, etc.) as we lack the page-specific information to make this determination possible.

Together, this suite of metrics arms buyers and sellers with valuable new insights into which inventory is most transparent to brand safety measurement and verification versus those that may require further analysis and investigation in order to open up their inventory to independent third-party brand safety verification.
a) dynamic content in javascriptDV does not have limitations to performing brand safety analysis on pages that have dynamic content. N/AWe cannot crawl Javascript webpages and therefore not return any brand safety signal. JS pages are heavily built with multiple APIs and crawling such pages can put a load on website.
b) iframe (is your solution able to detect the URL iframe in a page?) DV does not have limitations to performing brand safety analysis on pages that utilize iframes. N/ANo, we cannot detect URL within an Iframe. The client needs to send us a Publically available URL to our API Endpoints for categorizations.
c) websites that stop bots from visiting their pagesDV will contact the publisher to request access for it's automated technology so that classification can be properly performed. In the event that the automated technology cannot be used DV may also use a manual classification process to categorize the website. N/AYes,
If Grapeshot is unable to crawl a page, information about that inability will be indicated to Grapeshot’s partner or customer.
d) Which are the other limitations in running the Brand Safety analysis?User generated content, such as comments that require user interaction to access; content that is located behind a loginN/AThere are times when Contextual Intelligence is unable to deliver a meaningful signal about a page, such as when the page has no editorial content, is behind a login, is temporarily unavailable, or when Contextual Intelligence’s crawlers are blocked from a domain or subdomain. In such an instance Contextual Intelligence will return an "Information Code" signal (via a "gx_" preface in the signal).

Sometimes the Moat Analytics tag is loaded one to several iframes deep when called to a page for measurement. In these situations, it is possible that Moat is unable to identify the true page URL; rather, Moat will identify the domain on which the page resides. In these situations, classifications will be provided at the domain level.

Content Extraction Limitations
Certain limitations exist relating to content extraction beyond Context’s control it may not be possible to access the text context on some sites and pages. A number of the known cases are listed here.

Mobile App Capabilities
In mobile apps there are limited capabilities for contextual analysis due to limitations on the ability to look into pages within many apps. Mobile In-App Categorizations are outside the scope of Oracle Contextual Intelligence’s MRC Accreditation.

Overly Complicated Layout Schemes
Some sites use complicated and large html mark-up, including heavy use of CSS to properly layout the page. Without fully rendering the site in a supported browser it can be hard to determine where the actual main article text lies.

Dynamic Content Rendering
A small number of publishers choose to render their sites dynamically. Typically, this takes one of two forms. Either the payload is part of the site html payload but hidden within some embedded content and rendered dynamically with JavaScript as the page loads, or the initial payload kicks off an extra call, or calls, to the site to obtain the actual content. Without fully rendering the page, it’s not practical to get the correct content view, especially for the latter case. The Contextual Intelligence crawler does not load or execute JavaScript. As such, content dynamically loaded using JavaScript may not be crawled.

Assessments Context has done on these two cases suggest that the number of such URLs is a fraction of 1% percent against the whole of the internet.

There are some known sites that use these sort of rendering models specifically for their home pages, actual article pages are presented in a more normal fashion. Home pages have their own associated risks and limitations, as discussed later.

Infinite Scroll Design
Infinite scroll pages present additional content to the user as they scroll a page to completion. In these situations the last section of content from an article would cause the load of the next article from the section or site. As the user scrolled, they were presented with endless content.

As Contextual Intelligence’s crawler system does not render the page or run any JavaScript the Infinite Scrolling nature of the pages presents no challenges for the crawling system. Each article is processed individually as content in the usual method.

In almost all of the implementations Context has encountered, the URL was changed as the new article was loaded. This new URL is used in the requests for ads to be loaded for the next article. Where this happens, context is able to crawl and classify each article on its own URL and the infinite scrolling causes no issues to the crawlers or classification service.

Contextual Intelligence are aware of only a small handful of publishers globally (less than 10) that didn't update the given URL when requesting ads for the newly loaded content. In these cases, the classification will continue to be based on the original URL as we wouldn't be aware that the content of the viewed page had actually changed.

Mixed Language
Context systems can only operate using a single language. When determining the language of a site the language that is most dominant is typically used. In mixed language sites, English may be determined to be the dominant language, especially if the text is quite short, which could lead to incorrect text being extracted as this detected language is used as part of the scoring mechanism for determining the epicentre text.

Ambiguous Body/Article
Some pages can have multiple blocks of text split into multiple distinct parts in the html mark-up. In general, the text extraction system will pull all the text, but if the blocks are sufficiently distinct it is possible that only one of these sections will be selected. Selection is based on an algorithmic scoring model which looks for the most relevant text. In some rare cases the chosen block may not be the most contextually representative content on the page.

Home Pages
Publishers often put large numbers of snippets of articles on home pages which are all logically distinct text blocks. In general, where possible, Context extracts all the text on home pages, rather than trying to find one dominant text block. However, since this then results in the total extracted text covering multiple topics categorisation may not be as meaningful as for a single article.

Site Access Limitations
For reasons beyond Context’s control it may not be possible to access the text context on some sites and pages. A number of the known cases are listed here.

Paywall
Some sites implement a paywall system as part of their business model. Some have allowed Context’s systems to bypass their paywall, as they recognise the value added by this. Some sites allow a limited amount of reading before restricting access. In general, these can be processed by Context as normal, because of the implementation Context uses for accessing them. As this is the easiest model for publishers to implement this is quite a common one but poses no risk to Context at all.

Other sites will show a leading snippet of the article with a subscription link to see more. Since there is no distinct and machine-readable way to know this has happened, it is possible for only this snippet text to be considered for categorisation. This could potentially result in a mis-categorisation. Some of these sites do in fact have the entire content in the page. In these cases, Context will retrieve the entire page text.

Login Required Sites
Arguably a more extreme version of the Paywall, all content on a site requires a valid login to view. This usually results in a request to view the content causing a redirect to the login or register page. Context makes attempts to recognise this and not follow the redirect or try to extract text. Context never seeks to get login details for such sites, if the publisher declines to allow access without a login the text content will not be available for this site.

Some sites may have private content only viewable by specific logged in users. For example, email or social media. Context would be unable to view this content under any circumstances and would be unable to provide categorization for it. Context would never seek to gain access to this type of content.

IP/Geo Restricted Content
Some sites restrict which parts of the world can see their content. Some US sites, for example, decline to serve any content to EU nations as they believe GDPR regulations make this untenable. Other sites may be providing a service that for other legal or commercial reasons cannot be made available outside its primary geography.

Context’s page crawling systems are hosted in the UK, and as such the IP addresses are marked as UK addresses in geographic databases.

Such sites would typically redirect to a holding page stating the restrictions, for which Context would not extract content. However, in some rare cases publishers choose to simply replace the actual content with the restriction notice, and there is a risk that Context’s systems will use this text for categorisation, as it has no way of knowing this is not the real text. Fortunately, such cases are rare, publishers rapidly find that search engines encounter this same issue which dramatically reduces their visitor numbers.

Against the broad landscape of the internet, such sites are very rare. However, Context plans to implement, at least, US based crawling systems to minimise the impact of this in the near future.

Geographic Content
Most common for news publishers, content may be tailored to the determined geographic location of the request. As Context runs from the UK at this time it’s possible that UK content is returned as compared to US user for the same page, leading to a different categorisation. The vast majority of publishers do not actively change content in this way. Instead, they run multiple sites on different domains for each geography. Options are presented to the user to select which geography they want. Full articles are, in general, able to be processed in the correct way by Context.

The exception, in some cases, is the home page. On some sites visits to the home page always cause a redirect to the geographic home page, visitors from the “wrong” geography are simply denied access to that geography’s home page. Fortunately, such sites are very rare, as most publishers offer this is an option, rather than forcing. In some cases, Context has been able to either request the publisher allow multiple geography home pages to be seen or has been given a way to override the forced move.

Robots.txt limitations
A standard machine reading protocol, robots.txt, can be used by sites to allow or deny access to content. A well-behaved page crawling system, such as Context’s, must obey rules defined by the publisher’s robots.txt. If Context is asked for categorisation for such a denied page, it will not be possible to extract its text. It is also possible that the publisher has set the rate limit so low that getting good coverage of the site’s content is very difficult.

Rate limit restrictions
Publishers, and their systems, are often wary of being exposed to large scale requests, denial of service attacks in particular. Many have, or use, systems that try to detect potential abuse, too many requests in a short space of time can result in subsequent requests being blocked for a period of time. Context tries to avoid this with automated rate limiting strategies.

Note that as a well-behaved crawling system Context does advertise in the request to the site that it is a crawling system, no attempt is made to hide this or pretend to be a human. Some sites do actively block all crawling systems, regardless of purpose.

Requests to desist all crawling
On rare occasions publishers decline to use the robots.txt system and choose to ask Context to desist from crawling content instead. Context always abides by these requests, and a blocklist is maintained for this purpose. This is a relatively small list as the majority of sites that run ads recognise the need for content analysis.

Faked Content
Since Context advertises in its requests that it is a crawling system, there is theoretically scope for a publisher to return different content as a way of controlling the resultant categorisation. However, Context has run controlled studies on this and has yet to find any indication that this happens in practice.
3) What capabilities are available for monitoring and blocking in mobile app environments? NAN/AN/A
4) How frequently do you update the classification of different websites/pages/apps? DV updates classifications dynamically depending on multiple factors such as content risk levels, popularity/trends in volumes, news cycle, etc. For example, some pages containing multiple quickly refreshing content segments, such as homepages, have their classification refreshed close to 100 times every day. Other pages may have their classification refreshed daily or weekly. Additionally, DV mandates all websites with material traffic to have their classification refreshed every 30-90 days.Updates are done on a constant and recurring basis, depending on a lot of factors including the update frequency of each page.If a page has been modified since the last time it was crawled, then the crawling frequency is halved, to a lower limit of every 60 minutes; if it hasn’t been modified, then the crawling frequency is doubled, to a maximum of every 30 days. It is beholden on any partners using this service to maintain a high cadence of requests/re-requests of any URLs to ensure that the crawling process is prompted and executed.
5) For how many languages do you support categorisation?424131
6) Does French is one of the original languages used for categorisation?Yes

DV supports and is MRC accredited for property-level verification in French as well as 33 other languages at the page and domain levels and accredited across 173 languages for language targeting and keyword blocking.
YesYes
7) Is your solution compliant with the IAB filter categories? YesYesYes
8) Are you able to customize Brand Safety settings? YesYesYes
a) Sell Side approach: if yes, what can you customize (alerts, thresholds,..) and at what granularity level?Any dimension being reported - site/apps etc, as well as an inventory identifiers passed by sellersYes, Brand Safety customized thresholds and ad hoc context segments are availableClients can choose to block on the standard brand safety segments as well as one or more additional custom keyword segments they have defined. Also, they can choose to block on geo and domain blacklists or whitelists and on IVT detection as well.

The Moat Analytics dashboard has a built-in alert system that allows users to specify any metric for which they would like to receive alerts and the threshold that triggers one.
b) Buy Side approach: if yes, what can you customize (alerts, thresholds,..) and at what granularity level?Inclusion lists (domains, sub-domains, languages), Exclusion lists (domains, sub-domains, languages, keywords), Override lists (domains, sub-domains, URLs), content categories including categories with customizable tieringYes, Brand Safety customized thresholds and ad hoc context segments are availableClients can choose to block on the standard brand safety segments as well as one or more additional custom keyword segments they have defined. Also, they can choose to block on geo and domain blacklists or whitelists and on IVT detection as well.

The Moat Analytics dashboard has a built-in alert system that allows users to specify any metric for which they would like to receive alerts and the threshold that triggers one.
c) Buy Side approach: If yes, is your solution set up to enable clients to override your classifications?Yes,

DV's system is flexible and a Brand Safety profile is determined by the Clients. DV also allows for clients to maintain override lists in the event that even though a domain/page has a classification that the client chose as inappropriate it can be overridden and either allow the ad to serve or not count it as an incident.
Yes on client demandYes
Partners can create custom keyword blacklist segments to negatively target further pages for which a match is found.
c-1) If yes, at what granularity level?Domain, subdomain, and URLsdomain, urlKeyword level (Customizations are applied at the individual page level for any page that can be crawled)
d) What is the time needed for the customization? (to be built, and to be made available for pre-bid processes) This is available through the UI and is made available within minutes across DoubleVerify solutions and pre-bid integrationsFew hours. depends on the client's need of customizationIn a pre-bid environment, brand safety targeting is generally applied immediately as soon as the changes are made. (Precise timing may vary by individual DSP that has incorporated ODC’s brand safety technology.)

In a post-bid environment, new or updated blocking rules take effect within 24 hours.
9) Which are the processes to evaluate the accuracy of your Brand Safety segmentation? Leveraging machine learning and manual review*, DV classifies millions of websites on an ongoing basis. Our classification engine is purpose-built to analyze and create a detailed content profile of any piece of content using a comprehensive ontology, covering over 200,000 concepts using over 5.3 million rules. Each concept is associated with one or more patterns (words, phrases and regular expressions) that identifies that concept, as well as disambiguation rules to ensure that the relevant context is accurately identified. Multilingual ontologies are constructed by translating these concepts and adding language and culture-specific patterns. Classification is executed in over 40 languages.

* Manual Re-Review is conducted for high priority sites daily, with a sliding window, every 30-90 days
MOAT Oracle technology crawls hundreds of millions of webpages daily and indexes the core meaning of their text to gain an understanding of the subject matter and categorize it appropriately.
Algorithms are used to identify the relative “weight” of all words within the text (e.g., a news story on a webpage). These weighted words are generated as an atomic composition of that document. Separately creates groups of words and phrases known as “keyword segments.” These are themselves sets of words determined to reflect a particular topic. Matches the understanding of the processed text to the keyword segments and provides scores to indicate the degree of match. Contextual Intelligence will make multiple probabilistic matches between a set of such keyword segments and a document.
10) Are you able to prevent ads from being viewed post-bid?YesYesYes
11) Are you in the process of being accredited by MRC for Brand Safety based on the guidelines published in September 2018?NoNoNo
12) Have some of your features been accredited by MRC for Brand Safety? YesYesYes
a) If yes, which ones?All aspects of Desktop and Mobile Web brand safety (ad verification) are accredited across multiple languages including:

--unsuitable content incidents/blocks at domain and page level (in 35 languages at the domain and page level)

-- URL keyword and language exclusion incidents and blocking (in 173 languages)
Firewall Brand SafetyMatches the understanding of the processed text to the keyword segments and provides scores to indicate the degree of match. Grapeshot will make multiple probabilistic matches between a set of such keyword segments and a document.
13) Has your solution been accredited by MRC for alerting and blocking functions (IAB 2012 guidelines)?YesYesYes
 
 
1) Is your solution compliant with the GARM categories and tresholds?YesYesYes
2) Does your solution offer support for the GARM categories and Brand Safety floor?YesYesYes
2) bis If yes, could you please provide some information on your methodological approach?DV has developed a proprietary category taxonomy to provide coverage for the range of content that may pose a safety or suitability concern, in the spirit of the GARM standards.
DV provides a Brand Safety Floor setting, inclusive of five content categories. The content within DV’s Brand Safety Floor aligns with the APB/GARM Brand Safety Floor.
Please look at the "Description of Methodology"

Moat was part of the GARM working group, that helped develop the expanded guidance for each of the GARM categories.
3) Does your solution offer support for the GARM category risk levels?YesYesYes
 
 
Types of signal used for Brand Safety evaluation 
1) Do you use textual analysis of the page?YesYesYes
a) Do you need a prerequisite for implementing textual analysis?NoNoClients need to send us the URL for categorization and as long as we are able to access the URL, we should be able to categorize the page.
b) Do you have any limitations? If yes, which ones?NoNoNo
2) Do you use image analysis? YesYesNo but MOAT Oracle are in beta
a) Do you need a prerequisite for implementing image analysis?NoNoOur beta clients are publishers, who need to integrate with us directly
b) Do you have any limitations? If yes, which ones?NoNoOn launch we will support JPEG
3) Do you use video analysis? YesYesYes, MOAT Oracle categorise videos
a) Do you need a prerequisite for implementing video analysis?NoNoClients integrate with us directly
b) Do you have any limitations? If yes, which ones?NoNoClients integrate with us directly
4) Does your solution allow a semantic analysis?YesYesYes, MOAT Oracle use probabalistic analysis of all words on the page, which allows us to quickly and accurately determin the meaning and safety of the page.
5) Do you use the metadata analysis?Yes, DV looks at the Metadata when classifiying content.No
6) Do you use other signals for Brand Safety evaluation?YesYesNo
a) If yes, could you please specify which ones?DV Brand Safety consists not only of site, page, and app classifications, but also keyword targeting and inclusions/exclusion lists. For Keyword Targeting DV utilises the URL associated with the page to identify whether any of the keyword(s) designated as inappropriate by the client are present. Inclusion/exclusion domain and app Id lists are also provided by the client to dictate the sites, pages, and apps their ads should or should not be appearing on.IAS has partnered with the Global Disinformation Index (GDI) to transform how we help brands avoid misinformation, ensuring journalistic integrity and reaffirming support for quality news sites. The new partnership builds on IAS’s expertise in brand safety and suitability, further protecting brands from running ads on sites that GDI has identified for misinformation. N/A
7) Do you take into account user generated content for Brand Safety? YesYesYes
a) If yes, which ones (photos, videos, comments)? Image, video, and textual user-generated contentCommentaires, photos, videosTextual user-generated content, video
8) What capabilities are available for monitoring and blocking in mobile app environments? NAN/AN/A
Types of signal used for Brand Suitability evaluation 
1) Do you use textual analysis of the page?YesYesYes
a) Do you need a prerequisite for implementing textual analysis?NoNoNo
b) Do you have any limitations? If yes, which ones?NoNoAs browsers have gradually restricted page URL data to third-party tags, including those of verification vendors including Moat, the percentage of impressions in which we are able to verify the exact page URL of the impression has decreased. This reduces transparency in the ad tech ecosystem and results in potential brand safety risks, as verification providers cannot definitively classify the environment surrounding an ad impression without knowing the identity of the page.

Moat's metrics shed light on this development, breaking impressions into two groups: those with page-level brand safety measurability and those where only the domain was measurable. In the latter case, we calculate the average volume-weighted safety of that domain based on historical classifications and report on brand safety using the Domain Safe Rate and Domain Unsafe Rate metrics. For these impressions, we also do not increment the individual brand safety category metrics (e.g. Unsafe Rate - Crime, Unsafe Rate - Terrorism, etc.) as we lack the page-specific information to make this determination possible.

Together, this suite of metrics arms buyers and sellers with valuable new insights into which inventory is most transparent to brand safety measurement and verification versus those that may require further analysis and investigation in order to open up their inventory to independent third-party brand safety verification.
2) Do you use image analysis? YesNo but soon with Context (IAS has acquired Context, a digital content classification company
Context’s technology will be integrated into IAS’s Context Control suite of suitability and contextual targeting solutions. This will enable IAS’s marketing partners to identify brand suitable content beyond standard frameworks and contextually target with granularity.)
No but MOAT Oracle are in beta
a) Do you need a prerequisite for implementing image analysis?NoN/AOur beta clients are publishers, who need to integrate with us directly
b) Do you have any limitations? If yes, which ones?NoN/AOn launch we will support JPEG
3) Do you use video analysis? YesNo but soon with Context (IAS has acquired Context, a digital content classification company
Context’s technology will be integrated into IAS’s Context Control suite of suitability and contextual targeting solutions. This will enable IAS’s marketing partners to identify brand suitable content beyond standard frameworks and contextually target with granularity.)
Yes, MOAT Oracle categorise videos
a) Do you need a prerequisite for implementing video analysis?NoN/AClients integrate with us directly
b) Do you have any limitations? If yes, which ones?NoN/A 
4) Does your solution allow a semantic analysis?YesYesYes, MOAT Oracle use probabalistic analysis of all worsd on the page, which allows us to quickly and accurately determin the meaning and safety of the page.
5) Do you use the metadata analysis?Yes, DV looks at the Metadata when classifiying content.NoClients need to send us the URL for categorization and as long as we are able to access the URL, we should be able to categorize the page.
6) Do you use other signals for Brand Suitability evaluation?YesYesNo
a) If yes, could you please specify which ones?DV Brand Safety consists not only of site, page, and app classifications, but also keyword targeting and white/Exclusion lists.
For Keyword Targeting DV utilizes the URL associated with the page to identify whether any of the keyword(s) designated as inappropriate by the client are present.
White and black domain and app Id lists are also provided by the client to dictate the sites, pages, and apps their ads should or should not be appearing on.
The IAS CTV content-level brand safe targeting solution will be based on curated PMPs, created through the SSP integrations in the Publica's systemN/A
7) Do you take into account user generated content for Brand Suitability? YesNoYes
a) If yes, which ones (photos, videos, comments)? Image, video, and textual user-generated contentN/ATextual user-generated content, video
8) What capabilities are available for monitoring and blocking in mobile app environments? NAN/A

Solutions non accréditées - IAB 2012


Updated February 2024
Adloox
adloox
1) At what granularity level does your solution perform content analysis for Brand Safety? (calculation rules for scoring)
a) domainYes
b) subdomainYes
c) pageYes
c-1) If yes, please explain how your page level control works? (query analysis, words on the page, meaning detection...)?Query analysis and words on page
(detection of the meaning and/or list exclusion)
2) Which are the limitations in running the Brand Safety analysis (elements which explain that part of the page contents will not be accessible / limited access to the data)When Adllox tracking either JS or image pixel cannot be executed or when insufficient data points are being collected.
a) dynamic content in javascriptNo
b) iframe (is your solution able to detect the URL iframe in a page?) No, Adloox solution can detect the URL iframe in a page and the real delivery content URL
c) websites that stop bots from visiting their pagesNo
d) Which are the other limitations in running the Brand Safety analysis?Content Verification is not possible when the invocation tag is in multiple nested iframes, unless an ad server/ DSP / SSP is able to supply data through ‘Source URL’ macro.
3) What capabilities are available for monitoring and blocking in mobile app environments? N/A
4) How frequently do you update the classification of different websites/pages/apps? Daily
5) For how many languages do you support categorisation?Adloox does not intend to share this information
6) Does French is one of the original languages used for categorisation?Yes
7) Is your solution compliant with the IAB filter categories?
8) Are you able to customize Brand Safety settings? Yes
a) Sell Side approach: if yes, what can you customize (alerts, thresholds,..) and at what granularity level?Custom blacklist / whitelist / keyword exclusion – URL/Domain level granularity
b) Buy Side approach: if yes, what can you customize (alerts, thresholds,..) and at what granularity level?Custom blacklist / whitelist / keyword exclusion – URL/Domain level granularity
c) Buy Side approach: If yes, is your solution set up to enable clients to override your classifications?Yes
enabling campaign level keyword blocking to be uploaded and adapted in real-time
c-1) If yes, at what granularity level?URL/Domain
d) What is the time needed for the customization? (to be built, and to be made available for pre-bid processes) Maximum 24 hours
9) Which are the processes to evaluate the accuracy of your Brand Safety segmentation? Manual review & AI/ML
10) Are you able to prevent ads from being viewed post-bid?Yes
11) Are you in the process of being accredited by MRC for Brand Safety based on the guidelines published in September 2018?No
12) Have some of your features been accredited by MRC for Brand Safety? No
a) If yes, which ones?N/A
13) Has your solution been accredited by MRC for alerting and blocking functions (IAB 2012 guidelines)?No
1) Is your solution compliant with the GARM categories and tresholds?Yes
2) Does your solution offer support for the GARM categories and Brand Safety floor?Yes
2) bis If yes, could you please provide some information on your methodological approach?We use ML to categorize content using the GARM floor and risk levels framework allowing us to classify content ranging from domains, sub domains, urls and apps.
Then we proceed to manual verification of the content classification to ensure there is a human revision component in our workflow
3) Does your solution offer support for the GARM category risk levels?Yes
Types of signal used for Brand Safety evaluation
1) Do you use textual analysis of the page?Yes
a) Do you need a prerequisite for implementing textual analysis?No
b) Do you have any limitations? If yes, which ones?Content Verification is not possible when the invocation tag is in multiple nested iframes, unless an ad server/ DSP / SSP is able to supply data through ‘Source URL’ macro.

When Content Verification is not possible through a Javascript tag, Adloox utilizes a crawler to parse content and determine if a URL or domain meets clients brand suitability requirements.
2) Do you use image analysis? No
a) Do you need a prerequisite for implementing image analysis?N/A
b) Do you have any limitations? If yes, which ones?N/A
3) Do you use video analysis? No
a) Do you need a prerequisite for implementing video analysis?N/A
b) Do you have any limitations? If yes, which ones?N/A
4) Does your solution allow a semantic analysis?Yes
5) Do you use the metadata analysis?No
6) Do you use other signals for Brand Safety evaluation?No
a) If yes, could you please specify which ones?N/A
7) Do you take into account user generated content for Brand Safety? Yes
a) If yes, which ones (photos, videos, comments)? All levels (manual review)
8) What capabilities are available for monitoring and blocking in mobile app environments? N/A
Types of signal used for Brand Suitability evaluation
1) Do you use textual analysis of the page?Yes
a) Do you need a prerequisite for implementing textual analysis?No
b) Do you have any limitations? If yes, which ones?No
2) Do you use image analysis? No
a) Do you need a prerequisite for implementing image analysis?N/A
b) Do you have any limitations? If yes, which ones?N/A
3) Do you use video analysis? No
a) Do you need a prerequisite for implementing video analysis?N/A
b) Do you have any limitations? If yes, which ones?N/A
4) Does your solution allow a semantic analysis?Yes
5) Do you use the metadata analysis?
6) Do you use other signals for Brand Suitability evaluation?
a) If yes, could you please specify which ones?
7) Do you take into account user generated content for Brand Suitability? Yes
a) If yes, which ones (photos, videos, comments)? Comments
8) What capabilities are available for monitoring and blocking in mobile app environments? N/A

Environnement In-app

Solutions accréditées - IAB 2012


Updated February 2024
Double Verify
Double Verify
1) At what granularity level does your solution perform content analysis for Brand Safety? (calculation rules for scoring)
a) domainDV evaluates the app, but also the layers of content within the app if a URL exists within the request.
b) subdomainDV evaluates the app, but also the layers of content within the app if a URL exists within the request.
c) pageDV evaluates the app, but also the layers of content within the app if a URL exists within the request.
c-1) If yes, please explain how your page level control works? (query analysis, words on the page, meaning detection...)?DV uses proprietary technology as well as peer review to classify apps and layers of contents within the app with display and video assets. Our technology is able to see through to the true Bundle ID and use both own classification and store categories/age/user ratings to determine if the App is appropriate based on the advertisers selections.. Where URLs are available InApp we can provide further classification based on our ability collects data points such as incoming and outgoing links, text, context and the frequency and prominence of words and concepts and provide page level classification.
DV also created a first-to-market white/Exclusion list based on App ID (Bundle ID for Apple/Application ID for Android). This solution works across any / all partners so long as our JavaScript blocking wrapper is used.
2) Which are the limitations in running the Brand Safety analysis (elements which explain that part of the page contents will not be accessible / limited access to the data)Please see below for specific instances.
a) dynamic content in javascriptDV does not have limitations to performing brand safety analysis on pages that have dynamic content.
b) iframe (is your solution able to detect the URL iframe in a page?) DV does not have limitations to performing brand safety analysis on pages that are contained within embedded browsers used within apps to display content.
c) websites that stop bots from visiting their pagesDV will contact the app developer to request access for it's automated technology so that classification can be properly performed. In the event that the automated technology cannot be used DV may also use a manual classification process to categorize the app.
d) Which are the other limitations in running the Brand Safety analysis?User generated content, such as comments that require user interaction to access; content that is located behind a login
3) What capabilities are available for monitoring and blocking in mobile app environments? DV content categories, white/Exclusion lists, app rating, age rating, app store category
4) How frequently do you update the classification of different websites/pages/apps? DV reclassifies apps every 30-90 days.
5) For how many languages do you support categorisation?42
6) Does French is one of the original languages used for categorisation?Yes

DV supports and is MRC accredited for property-level verification in French as well as 33 other languages at the page and domain levels and accredited across 173 languages for language targeting and keyword blocking.
7) Is your solution compliant with the IAB filter categories? Yes
8) Are you able to customize Brand Safety settings? Yes
a) Sell Side approach: if yes, what can you customize (alerts, thresholds,..) and at what granularity level?Any dimension being reported - site/apps etc, as well as an inventory identifiers passed by sellers
b) Buy Side approach: if yes, what can you customize (alerts, thresholds,..) and at what granularity level?Inclusion lists for apps, Exclusion lists for apps, App star reviews threshold, App age rating setting, app override controls, content categories including categories with customizable tiering
c) Buy Side approach: If yes, is your solution set up to enable clients to override your classifications?Yes,

DV's system is flexible and a Brand Safety profile is determined by the Clients. DV also allows for clients to maintain override lists in the event that even though a bundle ID/App ID has a classification that the client chose as inappropriate it can be overridden and either allow the ad to serve or not count it as an incident.
c-1) If yes, at what granularity level?Bundle ID/ App ID
d) What is the time needed for the customization? (to be built, and to be made available for pre-bid processes) This is available through the UI and is made available within minutes across DoubleVerify solutions and pre-bid integrations
9) Which are the processes to evaluate the accuracy of your Brand Safety segmentation? Leveraging machine learning and manual review*, DV classifies millions of apps on an ongoing basis. Our classification engine is purpose-built to analyze and create a detailed content profile of any piece of content using a comprehensive ontology, covering over 200,000 concepts using over 5.3 million rules. Each concept is associated with one or more patterns (words, phrases and regular expressions) that identifies that concept, as well as disambiguation rules to ensure that the relevant context is accurately identified. Multilingual ontologies are constructed by translating these concepts and adding language and culture-specific patterns. Classification is executed in over 40 languages.

* Manual Re-Review is conducted for high priority sites daily, with a sliding window, every 30-90 days
10) Are you able to prevent ads from being viewed post-bid?Yes
11) Are you in the process of being accredited by MRC for Brand Safety based on the guidelines published in September 2018?No
12) Have some of your features been accredited by MRC for Brand Safety? Yes
a) If yes, which ones?All mobile app features are accredited by the MRC at the property level. This includes
A- pp contextual classification in 35 languages

- Exclusion list or Inclusion list apps based on App ID

- App star reviews

- App age rating

- App store categories
13) Has your solution been accredited by MRC for alerting and blocking functions (IAB 2012 guidelines)?Yes
1) Is your solution compliant with the GARM categories and tresholds?Yes
2) Does your solution offer support for the GARM categories and Brand Safety floor?Yes
2) bis If yes, could you please provide some information on your methodological approach?DV has developed a proprietary category taxonomy to provide coverage for the range of content that may pose a safety or suitability concern, in the spirit of the GARM standards.
DV provides a Brand Safety Floor setting, inclusive of five content categories. The content within DV’s Brand Safety Floor aligns with the APB/GARM Brand Safety Floor.
3) Does your solution offer support for the GARM category risk levels?Yes
Types of signal used for Brand Safety evaluation
1) Do you use textual analysis of the page?Yes
a) Do you need a prerequisite for implementing textual analysis?No
b) Do you have any limitations? If yes, which ones?No
2) Do you use image analysis? Yes
a) Do you need a prerequisite for implementing image analysis?No
b) Do you have any limitations? If yes, which ones?No
3) Do you use video analysis? Yes
a) Do you need a prerequisite for implementing video analysis?No
b) Do you have any limitations? If yes, which ones?No
4) Does your solution allow a semantic analysis?Yes
5) Do you use the metadata analysis?Yes, DV looks at the Metadata when classifiying content.
6) Do you use other signals for Brand Safety evaluation?Yes
a) If yes, could you please specify which ones?DV Brand Safety consists not only of site, page, and app classifications, but also keyword targeting and inclusion/exclusion lists. For Keyword Targeting DV utilises the URL associated with the page to identify whether any of the keyword(s) designated as inappropriate by the client are present. Inclusion/exclusion domain and app Id lists are also provided by the client to dictate the sites, pages, and apps their ads should or should not be appearing on.
7) Do you take into account user generated content for Brand Safety? Yes
a) If yes, which ones (photos, videos, comments)? Image, video, and textual user-generated content
8) What capabilities are available for monitoring and blocking in mobile app environments? DV content categories, white/Exclusion lists, app rating, age rating, app store category
Types of signal used for Brand Suitability evaluation
1) Do you use textual analysis of the page?Yes
a) Do you need a prerequisite for implementing textual analysis?No
b) Do you have any limitations? If yes, which ones?No
2) Do you use image analysis? Yes
a) Do you need a prerequisite for implementing image analysis?No
b) Do you have any limitations? If yes, which ones?No
3) Do you use video analysis? Yes
a) Do you need a prerequisite for implementing video analysis?No
b) Do you have any limitations? If yes, which ones?No
4) Does your solution allow a semantic analysis?Yes
5) Do you use the metadata analysis?Yes, DV looks at the Metadata when classifiying content.
6) Do you use other signals for Brand Suitability evaluation?Yes
a) If yes, could you please specify which ones?DV Brand Safety consists not only of site, page, and app classifications, but also keyword targeting and white/Exclusion lists.
For Keyword Targeting DV utilizes the URL associated with the page to identify whether any of the keyword(s) designated as inappropriate by the client are present.
White / App domain and app Id lists are also provided by the client to dictate the sites, pages, and apps their ads should or should not be appearing on.
7) Do you take into account user generated content for Brand Suitability? Yes
a) If yes, which ones (photos, videos, comments)? Image, video, and textual user-generated content
8) What capabilities are available for monitoring and blocking in mobile app environments? DV content categories, white/Exclusion lists, app rating, age rating, app store category

Solutions non accréditées - IAB 2012


Updated February 2024
Adloox
adloox
Integral Ad Science
Moat Oracle
1) At what granularity level does your solution perform content analysis for Brand Safety? (calculation rules for scoring) 
a) domainYes (we exclude apps as well)Yes, appN/A
b) subdomainN/AYes, appN/A
c) pageN/AYes, appN/A
c-1) If yes, please explain how your page level control works? (query analysis, words on the page, meaning detection...)?N/A
2) Which are the limitations in running the Brand Safety analysis (elements which explain that part of the page contents will not be accessible / limited access to the data)When Adllox tracking either JS or image pixel cannot be executed or when insufficient data points are being collected.Yes supported by IAS but need publisher adoption of OM SDK 1.3 and above (content_url field communicated through OM SDK 1.3 and above).Moat handles brand safety in two ways for app inventory. They are either supplied an app URL which they then crawl as per the standard page-level crawling methodology; or, they are suppled an app bundle ID whereby they then crawl the app description page and factor in other indicators such as the app store age rating.
a) dynamic content in javascriptNoN/AN/A
b) iframe (is your solution able to detect the URL iframe in a page?) No, Adloox solution can detect the URL iframe in a page and the real delivery content appN/AN/A
c) websites that stop bots from visiting their pagesNoN/AN/A
d) Which are the other limitations in running the Brand Safety analysis?N/AN/AIn mobile apps, Moat has limited capabilities for contextual analysis due to limitations on the ability to look into pages within
many apps.
3) What capabilities are available for monitoring and blocking in mobile app environments? Blocking will be effective as usual but Adloox will filter out non brand safe apps using Bundle IDs.
For delivery in a WebView context, Blocking capabilities will be similar to what we have on regular web environments.
Visibilite, fraude, brand safetyMoat handles brand safety in two ways for app inventory. They are either supplied an app URL which they then crawl as per the standard page-level crawling methodology; or, they are supplied an app bundle ID whereby they then crawl the app description page and factor in other indicators such as the app store age rating.
Then, besides contextual brand safety, Moat can block on geo, IVT and domain.
4) How frequently do you update the classification of different websites/pages/apps? DailyUpdates are done on a constant and recurring basis, depending on a lot of factors including the update frequency of each app.If a page URL (for in-app webviews) or app store description (for fully native in-app environments) has been modified since the last time it was crawled, then the crawling frequency is halved, to a lower limit of every 60 minutes; if it hasn’t been modified, then the crawling frequency is doubled, to a maximum of every 30 days. It is beholden on any partners using this service to maintain a high cadence of requests/re-requests of any URLs to ensure that the crawling process is prompted and executed.
5) For how many languages do you support categorisation?Adloox does not intend to share this information4131
6) Does French is one of the original languages used for categorisation?YesYesYes
7) Is your solution compliant with the IAB filter categories? YesYes
8) Are you able to customize Brand Safety settings? YesYesYes
a) Sell Side approach: if yes, what can you customize (alerts, thresholds,..) and at what granularity level?Custom blacklist / whitelist / keyword exclusion – App level granularityYes, Brand Safety customized thresholds and ad hoc context segments are availableClients can choose to block on the standard brand safety segments as well as one or more additional custom keyword segments they have defined. Also, they can choose to block on geo and domain blacklists or whitelists and on IVT detection as well.

The Moat Analytics dashboard has a built-in alert system that allows users to specify any metric for which they would like to receive alerts and the threshold that triggers one.
b) Buy Side approach: if yes, what can you customize (alerts, thresholds,..) and at what granularity level?Custom blacklist / whitelist / keyword exclusion – App level granularityYes, Brand Safety customized thresholds and ad hoc context segments are availableClients can choose to block on the standard brand safety segments as well as one or more additional custom keyword segments they have defined. Also, they can choose to block on geo and domain blacklists or whitelists and on IVT detection as well.

The Moat Analytics dashboard has a built-in alert system that allows users to specify any metric for which they would like to receive alerts and the threshold that triggers one.
c) Buy Side approach: If yes, is your solution set up to enable clients to override your classifications?Yes
enabling campaign level keyword blocking to be uploaded and adapted in real-time
Yes on client demandYes
Partners can create custom keyword blacklist segments to negatively target further pages for which a match is found.
c-1) If yes, at what granularity level?AppsappKeyword level (Customizations are applied at the individual page level for any page that can be crawled)
d) What is the time needed for the customization? (to be built, and to be made available for pre-bid processes) Maximum 24 hoursFew hours. depends on the client's need of customizationIn a pre-bid environment, brand safety targeting is generally applied immediately as soon as the changes are made. (Precise timing may vary by individual DSP that has incorporated ODC’s brand safety technology.)

In a post-bid environment, new or updated blocking rules take effect within 24 hours.
9) Which are the processes to evaluate the accuracy of your Brand Safety segmentation? Manual review & AI/MLMOAT Oracle technology crawls hundreds of millions of webpages daily and indexes the core meaning of their text to gain an understanding of the subject matter and categorize it appropriately.
Algorithms are used to identify the relative “weight” of all words within the text (e.g., a news story on a webpage). These weighted words are generated as an atomic composition of that document. Separately creates groups of words and phrases known as “keyword segments.” These are themselves sets of words determined to reflect a particular topic. Matches the understanding of the processed text to the keyword segments and provides scores to indicate the degree of match. Contextual Intelligence will make multiple probabilistic matches between a set of such keyword segments and a document.
10) Are you able to prevent ads from being viewed post-bid?YesYesYes
11) Are you in the process of being accredited by MRC for Brand Safety based on the guidelines published in September 2018?NoNoNo
12) Have some of your features been accredited by MRC for Brand Safety? NoNoYes
a) If yes, which ones?N/AN/AN/A
13) Has your solution been accredited by MRC for alerting and blocking functions (IAB 2012 guidelines)?NoN/ANo
 
 
1) Is your solution compliant with the GARM categories and tresholds?YesYes
2) Does your solution offer support for the GARM categories and Brand Safety floor?YesYes
2) bis If yes, could you please provide some information on your methodological approach?We use ML to categorize content using the GARM floor and risk levels framework allowing us to classify content ranging from domains, sub domains, urls and apps.
Then we proceed to manual verification of the content classification to ensure there is a human revision component in our workflow
Please look at the "Description of Methodology"

Moat was part of the GARM working group, that helped develop the expanded guidance for each of the GARM categories.
3) Does your solution offer support for the GARM category risk levels?YesYes
 
 
Types of signal used for Brand Safety evaluation 
1) Do you use textual analysis of the page?YesYesYes
a) Do you need a prerequisite for implementing textual analysis?N/ANoN/A
b) Do you have any limitations? If yes, which ones?NoNo
2) Do you use image analysis? NoYesNo but MOAT Oracle are in beta
a) Do you need a prerequisite for implementing image analysis?N/ANoOur beta clients are publishers, who need to integrate with us directly
b) Do you have any limitations? If yes, which ones?N/ANoOn launch we will support JPEG
3) Do you use video analysis? NoYesYes, MOAT Oracle categorise videos
a) Do you need a prerequisite for implementing video analysis?N/ANoClients integrate with us directly
b) Do you have any limitations? If yes, which ones?N/ANoClients integrate with us directly
4) Does your solution allow a semantic analysis?YesYesYes, MOAT Oracle use probabalistic analysis of all words on the page, which allows us to quickly and accurately determin the meaning and safety of the page.
5) Do you use the metadata analysis?NoWe use metadata for In-App Brand Safety analysis.
6) Do you use other signals for Brand Safety evaluation?NoYesNo
a) If yes, could you please specify which ones?N/AIAS has partnered with the Global Disinformation Index (GDI) to transform how we help brands avoid misinformation, ensuring journalistic integrity and reaffirming support for quality news sites. The new partnership builds on IAS’s expertise in brand safety and suitability, further protecting brands from running ads on sites that GDI has identified for misinformation. N/A
7) Do you take into account user generated content for Brand Safety? YesYesYes
a) If yes, which ones (photos, videos, comments)? All levels (manual review)Commentaires, photos, videosTextual user-generated content, video
8) What capabilities are available for monitoring and blocking in mobile app environments? Blocking will be effective as usual, but only in a WebView contextBrand Safety, FraudeMoat handles brand safety in two ways for app inventory. They are either supplied an app URL which they then crawl as per the standard page-level crawling methodology; or, they are supplied an app bundle ID whereby they then crawl the app description page and factor in other indicators such as the app store age rating.
Then, besides contextual brand safety, Moat can block on geo, IVT and domain.
Types of signal used for Brand Suitability evaluation 
1) Do you use textual analysis of the page?Yes,
application-by-application only
YesYes
a) Do you need a prerequisite for implementing textual analysis?Yes if OMD SDK 1.3 and aboveNo
b) Do you have any limitations? If yes, which ones?Moat handles brand safety in two ways for app inventory. They are either supplied an app URL which they then crawl as per the standard page-level crawling methodology; or, they are supplied an app bundle ID whereby they then crawl the app description page and factor in other indicators such as the app store age rating.
2) Do you use image analysis? NoNo but soon with Context (IAS has acquired Context, a digital content classification company
Context’s technology will be integrated into IAS’s Context Control suite of suitability and contextual targeting solutions. This will enable IAS’s marketing partners to identify brand suitable content beyond standard frameworks and contextually target with granularity.)
No but MOAT Oracle are in beta
a) Do you need a prerequisite for implementing image analysis?N/AN/AOur beta clients are publishers, who need to integrate with us directly
b) Do you have any limitations? If yes, which ones?N/AN/AOn launch we will support JPEG
3) Do you use video analysis? NoNo but soon with Context (IAS has acquired Context, a digital content classification company
Context’s technology will be integrated into IAS’s Context Control suite of suitability and contextual targeting solutions. This will enable IAS’s marketing partners to identify brand suitable content beyond standard frameworks and contextually target with granularity.)
Yes, MOAT Oracle categorise videos
a) Do you need a prerequisite for implementing video analysis?N/AN/AClients integrate with us directly
b) Do you have any limitations? If yes, which ones?N/AN/A 
4) Does your solution allow a semantic analysis?YesYesYes, MOAT Oracle use probabalistic analysis of all words on the page, which allows us to quickly and accurately determin the meaning and safety of the page.
5) Do you use the metadata analysis?No 
6) Do you use other signals for Brand Suitability evaluation?YesNo
a) If yes, could you please specify which ones?The IAS CTV content-level brand safe targeting solution will be based on curated PMPs, created through the SSP integrations in the Publica's systemN/A
7) Do you take into account user generated content for Brand Suitability? YesNoYes
a) If yes, which ones (photos, videos, comments)? CommentsN/ATextual user-generated content, video
8) What capabilities are available for monitoring and blocking in mobile app environments? Adloox can filter out non brand safe apps using data retrived directly by its Javascript tag and relying on telemetry available through OM SDK.Moat handles brand safety in two ways for app inventory. They are either supplied an app URL which they then crawl as per the standard page-level crawling methodology; or, they are supplied an app bundle ID whereby they then crawl the app description page and factor in other indicators such as the app store age rating.
Then, besides contextual brand safety, Moat can block on geo, IVT and domain.