Contextual signals are the parts of a web page that say what it is about: the title, headings, article body, URL path, metadata, structured data, images and captions. The article body is the most reliable of them. Navigation menus, sidebars, related story boxes and reader comments sit on the same page but describe other things, so they mislead a reading.
Key takeaways
The article body carries the most dependable contextual signal because it is the part the author wrote about this one subject.
Menus, sidebars and recommendation boxes describe the rest of the site, so they pull a page toward topics it does not cover.
Reader comments belong to the page but not to the article, so they stay out of an article's context.
Metadata and URL paths are written for search and sharing, which makes them useful clues and unreliable proof.
What counts as a contextual signal on a web page?
Definition
Contextual signal
Any part of a page, visible or in its code, that a person or a system can read to decide what the page is about and how it treats that subject.
A single article page is a bundle of signals written by different people for different jobs. The reporter wrote the body. An editor wrote the headline and maybe a separate title for search. A developer built the menu, which lists every section of the site. A recommendation widget filled the sidebar, and readers wrote the comments. Each one is on the page, but only some describe the article itself. The broader method of placing ads by content is covered in our contextual targeting guide.
Platforms read several of these signals together. Google's contextual targeting help (opens in a new tab) says it analyzes a Display Network page's text, language, link structure and page structure to determine its central theme. A planner reviewing a page by hand reads the same layers and should know which ones to discount.
Which parts of a page carry the most reliable contextual signal?
Contextual signals on an article page, how reliable each is and how each misleads
Signal
What it is
Reliability
How it misleads
Article body
The paragraphs, lists and tables the author wrote about the subject
High
A short body padded with quotes or embeds can be thin evidence
Headline and headings
The visible article name and the labels that organize its sections
High when the body agrees
Playful or clickable headlines can name nothing the article covers
Title element
The text in the page code that browsers and search engines show
Medium
Can be stale after an edit, or mostly the site's name
Images and captions
Photos, charts and the text that explains them
Medium
Stock photos and decorative images say little about the story
Structured data
Machine readable fields such as headline, section and body
Medium
Fields may be empty, generic or describe the site, not the article
Meta description and social tags
Short summaries written for search results and link previews
Low to medium
Written to win clicks, sometimes reused across pages
URL path
The words in the address after the domain
Low to medium
Slugs can survive headline changes, and some paths hold only IDs
Menus, sidebars, footers
Navigation and supporting boxes repeated across the site
Low for the article
They describe the whole site, not this page
Reader comments
Text added by readers below the article
Low for the article
Off-topic threads, arguments and spam drift far from the subject
Where to look first on a page
Read from the top tier down, and treat lower tiers as clues to check against the body.
How do navigation menus and sidebars contaminate page context?
The HTML standard draws the line clearly. The WHATWG HTML specification (opens in a new tab) says the nav element represents a section of a page that links to other pages or to parts within the page, and that the aside element represents content that is tangentially related to the content around it. The W3C's page regions tutorial (opens in a new tab) describes the main element as the way to identify the main content region of a web page.
Contamination happens when a reading treats all page text as one bag of words. A recipe page on a general news site may carry a menu with Crime, Politics and Obituaries, a trending box with the day's worst headline and a footer full of legal links. Counted together with the recipe, those words drag the page toward hard news. Repeated across every article on the site, the same menu words also make unrelated pages look alike.
Related story boxes are the hardest case because they look editorial. They sit near the article, carry real headlines and can share its section. They still describe other pages. A sidebar listing five crime stories next to a gardening column says the site covers crime, not that the column does.
Should reader comments count toward a page's context?
For an article, no, with one exception below. The WHATWG specification gives the example of user comments on a blog post marked up as article elements nested inside the post's article, related to it but separate compositions. That matches how a planner should treat them: comments respond to the article, they are not part of what the author published.
Comments drift. A thread under a pasta recipe can turn into an argument about politics within a dozen replies.
Comments change after review. A page judged on Monday can carry very different reader text on Friday.
Comments attract spam. Link drops and promotional posts add vocabulary the publisher never chose.
Comments sometimes are the content. On a forum or discussion thread the posts are the placement, so they must be read, and the unit being judged is the thread, not an article.
Judge the article without its comments and record separately whether comments are present and moderated. Comment sections raise suitability questions of their own, which belong to the suitability review rather than the topic match.
What does page metadata add that the article body does not?
Metadata describes the page in fields a machine can read without guessing. Schema.org's Article type (opens in a new tab) defines articleBody as the actual body of the article and articleSection as the sections, such as Sports or Lifestyle, an article belongs to. The Open Graph protocol (opens in a new tab) defines article:section as a high-level section name and article:tag as tag words associated with the article. These fields add things the body may not state outright: the publisher's own section label, publication and modification dates, the language and the author.
Metadata fields a contextual review can use, in their specifications' words
Field
Defined as
Use in a review
articleSection (Schema.org)
Sections in a magazine or newspaper, such as Sports, Lifestyle
The publisher's own filing of the story
dateModified (Schema.org)
The date the work was most recently modified
Spots old pages that were refreshed
og:description (Open Graph)
A one to two sentence description of the object
A summary to test against the body
article:tag (Open Graph)
Tag words associated with this article
Publisher keywords, sometimes broad
keywords (OpenRTB 2.6 Content)
Comma separated list of keywords describing the content
What the seller says about the page in the bid request
Metadata is written for search results, link previews and ad requests, so it serves those audiences first. Google's snippet documentation (opens in a new tab) says Google primarily uses the content on the page to determine a snippet and uses the meta description only sometimes. Its title link guide (opens in a new tab) lists obsolete and inaccurate title elements and micro-boilerplate text among common issues. A reviewer should do the same: let metadata point to a subject, then confirm it in the body.
How do URL paths help or mislead contextual classification?
A readable path is a useful first clue. Google's URL structure guidance (opens in a new tab) recommends readable words rather than long ID numbers and words in the audience's language. A path such as /food/recipes/weeknight-chili tells a reviewer where the publisher filed the page before a word of the body is read, and OpenRTB 2.6 passes the page address in the site's page field, so the URL can reach the buyer even when nothing else does.
Four ways a URL path misleads
Read the slug as history. Slugs are set at first publish and can survive later headline changes, so a path can describe a story the page no longer tells.
Discount section folders. A /news/ folder can hold a recipe, and a /lifestyle/ folder can hold hard reporting.
Expect no words at all. Some paths are dates and numeric IDs that say nothing about the subject.
Watch for redirects. An address saved in a plan can now lead somewhere else, so record the final page that loaded.
What does a signal review look like on a real page?
Fictional example
An espresso machine brief in North Carolina
An unnamed Raleigh appliance retailer wants display ads beside coffee brewing content for readers in Raleigh, Charlotte and Wilmington. The planner reviews a candidate page on a regional news site whose path is /news/local/2024/03/coffee-shop-story and whose headline reads How to descale your espresso machine at home.
The body is a step by step descaling guide with a photo of a portafilter and a caption naming the part, so the top tiers agree: this is home coffee care. The path is stale history from an older story about a coffee shop closing, and the planner notes it but does not rely on it. The sidebar carries a trending box about a downtown robbery and the menu lists Crime and Courts, both discarded as site furniture. Thirty comments below the guide argue about a local zoning vote, which the planner records as present and off topic, to raise in the suitability review.
The match reason reads: home descaling guide in the body, headline and caption agree; path stale; sidebar and comments excluded.
Which page signals does PlacementGraph read?
What should a contextual signal record contain?
Signal record for one reviewed page
The final page address that loaded, with the date it was read.
The subject as stated by the article body, with the passage that shows it.
Whether the headline, title element and captions agree with the body.
Any metadata section label or tags, quoted with their field names.
Whether the URL path matches the current story or is stale.
Menu, sidebar and related story text marked as excluded from the topic judgment.
Whether comments are present, and a note for the suitability review if they are off topic.
Frequently asked questions
Do image alt text and file names count as signals?
They can help, with care. The W3C's images tutorial (opens in a new tab) says decorative images should carry an empty text alternative, so many images rightly say nothing. Informative images with captions are better evidence. File names can be camera defaults or stock library codes and should not decide a topic.
What if only a teaser is visible before a paywall?
Judge the page on what loads and say so in the record. A two paragraph teaser plus a headline can confirm a subject but not its treatment. If the publisher exposes the full text in structured data, note that the review used it, since a reader without a subscription sees less.
Does a missing structured data block make a page weaker?
No. Good publishers can use little or no structured data. Its absence removes one clue and says nothing about editorial quality or suitability. Record the field as absent rather than empty, and rely on the body and headings for the topic.
How should a page that embeds a video be read?
Read the article text around the embed as the page's context, and treat the video as a separate placement with its own signals. An embedded clip can be on a different subject from the article that hosts it, so its title and channel should be checked rather than assumed to match.
Share a brief in plain language. Get contextual websites and YouTube channels that fit it, with the evidence behind every pick and a deal ready for your usual buying platform.
01
Share your brief
The campaign in plain language.
02
Get your shortlist
The strongest contextual sites and channels, with the evidence behind each.
03
We build the deal
Your shortlist becomes a curated deal through our exchange partners.
04
Activate and grow
Buy in your usual platform. Agencies earn benefits as their investment grows.
