Entity targeting: placing ads by the things a page names
Part of Contextual targeting
The short answer
Entity targeting places ads on pages that discuss a specific named thing, such as a product, a place, a person or an organization, identified as that thing rather than as a string of letters. It reaches coverage of a launch, a destination or a team, but a page can name an entity in a story the brand would avoid.
Key takeaways
- An entity is a specific thing in a knowledge base, while a keyword is a string; entity linking decides which thing a mention refers to from its context.
- A page that names a brand can be a review, a recall story or a lawsuit report, so an entity match is not a suitability check.
- Shared names, such as a product named after a plant or a city, are a source of wrong matches and need a test before launch.
- Entity targeting suits launches when coverage names the product, the category leaders or the event, but a brand new name may not be recognized yet.
What is entity targeting in contextual advertising?
Definition
- Entity targeting
- Selecting or excluding pages, videos or apps because their content refers to a specific named thing, identified through entity recognition and linking rather than by matching a word.
The building block comes from language technology. Google's Cloud Natural Language reference describes an entity (opens in a new tab) as a phrase in the text that is a known entity, such as a person, an organization, or location. Each entity carries a salience score in the 0 to 1.0 range, where scores closer to 1.0 are highly salient, and for most entity types the metadata is a Wikipedia URL and Knowledge Graph MID, if they are available. That field points to one identified thing.
| Entity type | Example of what a brief might name | Typical planning use |
|---|---|---|
| CONSUMER_GOOD | A product line or model | Reach reviews, unboxings and buying guides about the product or its category |
| ORGANIZATION | The brand, a retailer partner, a league or a university | Reach coverage of partners, sponsors or the brand itself |
| LOCATION | A city, a park, a stadium or a region | Reach travel, local and event coverage tied to a place |
| EVENT | A trade show, a festival or a championship | Reach previews and coverage around a dated occasion |
| PERSON | An athlete, an author or a chef | Reach interviews and profiles, with the highest suitability risk |
How is an entity different from a keyword?
A keyword is a string of characters. It matches whenever the letters appear, whatever they refer to. An entity is a referent: the specific product, place or person that a mention points to. Moving from one to the other is the task researchers call entity linking. A 2026 EACL paper, Entity-Linkings: A Unified Library for Entity Linking (opens in a new tab), defines it as the task of mapping named entities in text to canonical entries in a knowledge base, and describes two steps: detecting entity mentions, then linking each mention to a unique identifier.
| Question | Keyword matching | Entity matching |
|---|---|---|
| What is matched | Characters in the text | An identified thing in a knowledge base |
| A name with two meanings | Matches both meanings | Picks one meaning from context, and can pick wrongly |
| A brand new name | Works at once, the string is just letters | May fail until the name is in the knowledge base |
| Transparency | Easy to audit, the word is either there or not | Harder, the linking decision needs an explanation |
| How central the thing is | Not measured | Can be scored, for example with a salience value |
Keyword lists stay useful for exact strings, such as a model number; the method for building them sits in our guide to contextual keyword lists. Entity matching earns its place when a name is ambiguous or when the buyer cares whether a page is about the thing, not merely whether it contains the word.
What risks come with targeting pages that mention a brand?
A page that names a brand is not necessarily a page the brand wants. The same entity appears in reviews, recall notices, earnings stories, lawsuits and social commentary. An entity match answers whether the brand is named; suitability answers whether the treatment is acceptable, and they are separate checks. Display & Video 360's sensitive categories (opens in a new tab) help shows why the second check cannot be skipped: it lists categories such as Tragedy, content related to death, disasters, accidents, war, and notes that recent events and news may take several days to be blockable.
| Whose name is on the page | Main risk | Check before buying |
|---|---|---|
| The advertiser's own brand | Recall, complaint or lawsuit coverage carries the ad beside bad news about itself | Read recent pages naming the brand and set an exclusion for negative news treatment |
| A competitor | The ad can read as a reply to the competitor, and comparison claims carry their own rules | Keep competitor claims out of the creative |
| A person linked to the brand | Coverage of a sponsored athlete or founder can turn personal quickly | Use the strictest tolerance and review daily during news events |
Depth of mention matters too. YouTube's advertiser-friendly content guidelines (opens in a new tab) separate focus, when a segment or full video is about a given topic, from fleeting moments that are not the focus of content, and state that a passing reference to a sensitive topic is not a reason for No Ads. The same holds for brands: one mention in a roundup is weak evidence; a full review is strong evidence either way.
When is entity targeting useful for a launch?
Launch coverage names things: the new product, its category, the announcement event and the places it goes on sale. Entity targeting can reach that coverage while it is fresh. Its weakness is the new name itself. A 2022 NAACL industry paper, ReFinED (opens in a new tab), explains that early linking techniques could only link mentions to entities that appeared in the training data, a problem because knowledge bases are continuously growing, and that newer models link to entities added to the knowledge base after training, termed zero-shot. The same paper notes that Wikipedia covers just 3 percent of the entities in Wikidata.
| Entity | Before launch | Launch week | After launch |
|---|---|---|---|
| The new product name | May not be recognized yet; use keywords for the exact string | Check whether pages are linked to the right product | Strongest signal once reviews appear |
| The product category | Reach buying guides and roundups | Keep running beside category coverage | Keep, with suitability checks |
| The launch event or venue | Reach previews and schedules | Reach live coverage, with the strictest review | Stop once coverage turns to other news |
| The brand itself | Exclude negative treatment | Review daily | Return to standard checks |
Entity plan checklist before a launch
- Every target entity is described precisely, with its knowledge base entry or a one line definition.
- Every shared meaning of each name has been listed and either excluded or tested.
- The brand's own name has a negative news exclusion and a named reviewer.
- A test line item has run and a sample of served pages has been read.
- Creative makes no claim of endorsement by the publisher or by any person named on the page.
What does an entity plan look like on a real brief?
Fictional example
A laptop launch in Arizona
A fictional laptop maker is launching a lightweight model in Arizona, with English display ads around Phoenix and Tucson timed to back to school season. The model's name is borrowed from a desert cactus, which is also the subject of Arizona travel and nature pages.
The planner lists three meanings of the name: the laptop, the plant and a park named after the plant. The target is described as the laptop model announced by the maker, beside reviews, student buying guides and launch coverage. Plant and park pages are excluded, and so are recall and complaint stories that name the maker.
A one week test shows a cluster of nature blogs among served pages, so the planner tightens the description and adds the plant's topic as an exclusion. Competitor names stay out of targeting. The plan keeps a keyword line for the exact model number, since the new name is not yet reliably recognized.
How does PlacementGraph judge a page by meaning rather than a matched name?
Frequently asked questions
Does entity targeting need personal data?
No. A person entity is someone a page writes about, not the person reading it. Entity targeting reads the content of the page and identifies what it refers to, without knowing who is viewing. Measurement tags added to the campaign are separate data flows.
Can entity targeting reach video and podcasts?
Only through the text available about them. On YouTube that means titles, descriptions and similar metadata, and Google Ads says keyword and topic targeting act as a hint on YouTube while placement targeting is fully supported. For named entities on YouTube, a reviewed channel list is the more predictable route.
Is targeting pages about a celebrity a good idea?
Only with the strictest review. Coverage of people can turn from positive to negative quickly, and an ad beside a profile can look like an association the person never agreed to. If a person is targeted, review served pages daily and pause at the first news story.
How should an entity plan be reported to the client?
List each target entity with its description, the meanings excluded and the date of the last test. Add the domains that took the most delivery and a few served pages that show what the match looked like. Show one right match and one caught wrong one.
Sources
- Entity (Cloud Natural Language API reference) (opens in a new tab)Google Cloud Documentation
- Entity-Linkings: A Unified Library for Entity Linking (Sawada, Fujita, Sakai and Watanabe) (opens in a new tab)Association for Computational Linguistics, EACL 2026 System Demonstrations
- ReFinED: An Efficient Zero-shot-capable Approach to End-to-End Entity Linking (Ayoola et al.) (opens in a new tab)Association for Computational Linguistics, NAACL 2022 Industry Track
- Sensitive categories in Display & Video 360 brand safety targeting (opens in a new tab)Display & Video 360 Help
- Advertiser-friendly content guidelines (opens in a new tab)YouTube Help
- About the Where ads showed report (opens in a new tab)Google Ads Help
- About contextual targeting (opens in a new tab)Google Ads Help
- Native Advertising: A Guide for Businesses (opens in a new tab)US Federal Trade Commission, December 2015
- The Canadian Code of Advertising Standards (opens in a new tab)Ad Standards (Canada), revised July 2019
Researched and written by PlacementGraph Research. Read how we research, source and update these pages. Corrections: diana@placementgraph.com.