How do you find similar websites from one example article?
Part of Lookalike websites
The short answer
One to three example articles can seed a lookalike search when the client values a specific piece or section rather than a whole site. Pick substantial articles that each stand for one clear subject, search different subjects separately, and confirm each result by reading several of its recent articles, because one matching page does not make a matching site.
Key takeaways
- An article seed works like a query document: the search looks for pages close to it in meaning, then the planner judges the sites those pages belong to.
- Article seeds beat site seeds when the client loves one column or story inside a broad site, or a subject its favorite sites cover only now and then.
- A good example article is open to read, substantial, original and about one clear subject.
- Articles on different subjects pull a combined search toward the middle, so group them by subject and search each group on its own.
- Each result is checked as a whole site, not as the one page that matched.
Can you find similar sites from one example article?
Yes, with a caution about what the result means. Searching from a document is an established method in information retrieval. In Content-Based Citation Recommendation (opens in a new tab) (NAACL 2018), Bhagavatula and coauthors write: "We embed a given query document into a vector space, then use its nearest neighbors as candidates," and then "rerank the candidates using a discriminative model." A lookalike search from an article follows the same outline: turn the article into a representation of its meaning, find pages near it, then judge the candidates.
Definition
- Article seed
- A single article, chosen by the client or planner, used as the example for a lookalike website search in place of a whole site, so the search looks for publishers whose articles mean the same thing.
The caution is that a plan buys sites, not one page. Nearest neighbors of one article are other articles, and some of them sit on sites that published one such piece and nothing else like it. The search finds the pages; the planner still has to show that the site around each page covers the subject regularly. That second step is what turns a match into a lookalike.
When are article seeds better than site seeds?
IAB Tech Lab's implementation guidance (opens in a new tab) for its taxonomies describes categories that capture a piece of content's topic context, or "aboutness," separately from attributes such as language and format. The point for planners is that aboutness belongs to a page, and a site is a collection of pages on several subjects. When the client's interest lives in a few pages, seed with the pages. The guide to seed site selection covers the case where a whole site fits.
| Situation | Better seed | Why |
|---|---|---|
| The client loves one column inside a national newspaper | Two or three articles from that column | The whole newspaper would find other newspapers |
| The client forwards a single story and says "more like this" | That article, plus one similar piece if one exists | The story is the clearest statement of what they want |
| The favorite site is a specialist that covers only the subject | The site, or its homepage and recent articles | Its pages already cover the subject |
| A new product's subject is covered only now and then by the client's sites | Articles on that subject from any good source | Site seeds would pull toward the sites' main subjects |
| The client likes a format, such as long reported features | Articles that show the format and subject together | Format lives in articles, not in domains |
How do you choose a good example article?
The article has to carry the subject clearly. Google describes its own contextual targeting (opens in a new tab) as a system that "analyzes the content of each webpage to determine its central theme." An article with a muddled theme gives any meaning based search a muddled target. Google's guidance on creating helpful content (opens in a new tab) (last updated October 5, 2026) offers questions that separate a substantive piece from filler, such as "Does the content provide original information, reporting, research, or analysis?"
What makes a good example article
- The article is open to read without a login, so its full text can be compared, not just its headline.
- One subject runs through the whole piece, and a reader could state it in a sentence.
- It answers Google's question "Does the content provide a substantial, complete, or comprehensive description of the topic?" with a yes.
- It is editorial, not a sponsored post, product listing or press release.
- It is written in a language the brief allows.
- The planner knows what the client values in it: the subject, the depth, the voice or the format.
- It comes from a site that passes the same quality checks the results will face.
Avoid articles whose subject is a news event. A piece about a product recall or a celebrity's kitchen will find other coverage of that event, which stops once the event passes, rather than publishers that cover the subject regularly.
How do mixed theme examples affect results?
Adding examples helps only while they agree. In a study of seed driven document ranking (opens in a new tab) for systematic reviews (ECIR 2022), Wang, Scells, Mourad and Zuccon report that "the use of multiple seed studies produces more stable rankings compared to single seed studies." Those seeds shared one review topic. Research on pseudo relevance feedback points to the risk when they do not: Zhu and coauthors' SIGIR 2022 paper (opens in a new tab) notes that a model "may incorrectly focus on the additional irrelevant information" in added feedback, and links existing methods to "severe query drift."
| Example set | What the search returns | Fix |
|---|---|---|
| One article, one subject | Close matches, some on sites that covered it once | Judge each site on several recent articles |
| Two or three articles, one subject | Matches anchored on the shared subject | None; this is the target setup |
| Articles on two subjects, searched together | Pages halfway between, such as general lifestyle portals | Split by subject and search each group |
| An article that mixes two subjects in one piece | Results that follow either subject unpredictably | Replace it with an article on one subject |
Splitting costs one extra search: each subject gets its own list, labeled, and the client chooses the balance. The same pull toward the middle over several rounds of expansion is described in the guide to lookalike drift.
How do you check results from an article seed?
Results from an article seed arrive as pages. Each page has to be traced back to its site and its section before anything goes on a list.
Six checks for each result of an article seed search
- Open the matching page. Confirm it covers the seed's subject, not just its vocabulary.
- Read several recent articles from the same site. Count how many cover the subject; a single match is a one off, not a lookalike.
- Find the section. If the matches cluster in one path, note it; Google lets advertisers exclude "by domain, subdomain, path name, or individual pages," so section level decisions are possible in Google Ads.
- Run the standard checks. Review page quality, made-for-advertising risk, suitability, authorized sellers and audience in the target country.
- Mark good and bad results. Use them to refine the next round; research on relevance feedback (opens in a new tab) (EMNLP 2022) explores how such judgments can be integrated into reranking.
- Confirm after launch. Google's Where ads showed report (opens in a new tab) shows where ads ran, but Google says not to use it "as an exhaustive list of placements for your ads."
Section level buying has its own example in Google's help on excluding webpages and videos (opens in a new tab): a seller of organic cleaning products "targets a popular product review website, and adds a site exclusion for the website paths about food gadgets." An article seed search that matches one section of a broad review site leads to the same kind of decision.
How does PlacementGraph search from example pages?
What does an article seed search look like on a real brief?
Fictional example
A fictional multicooker maker in Yukon
An unnamed Whitehorse maker of electric pressure cookers and multicookers wants a winter display plan for Yukon, in English. The client cannot name a favorite site but forwards three articles it loved: a food magazine feature on batch cooking soups and stews for long winters, a home cooking blog's guide to pressure cooking dried beans, and a newspaper column on renovating a small kitchen.
The planner sees two subjects. The first two articles are about cooking methods; the third is about kitchen renovation. The renovation column goes to a separate search, labeled as a test theme, and the two cooking articles seed the main search together.
The cooking search returns pages from home cooking blogs, a Canadian food magazine and a general lifestyle portal. Reading recent articles on each, the planner keeps the blogs and the magazine's cooking section, where the subject recurs, and drops the portal, which had one stew recipe among travel and celebrity pieces. The list labels every site with the seed article it matched.
Frequently asked questions
Can a paywalled article be used as an example?
Only if its full text can be read for the comparison. A search that sees only a headline and a teaser compares a few sentences, not the article. If the client's favorite piece sits behind a paywall, ask for an open article from the same section, or describe its subject in a sentence and use that as the brief.
Will the seed article's own site appear in the results?
It may, because its other articles can be close to the seed. Treat it like any other result: it goes on the list only if several of its recent articles cover the subject and it passes the standard checks. A broad site whose one great article was the seed is better bought by section, if at all.
Should the example article be recent?
Recency matters less than whether the subject is still covered. A three year old guide to a cooking method can be a fine seed if sites still publish on that method. An old article tied to a past event is a poor seed, because the coverage it matches has stopped.
Can a video be used as an example for a website search?
A video's title and description can suggest the subject, but a website search compares article text, so the planner should find an article that covers the same ground. For finding channels from a video, the guide to finding similar YouTube channels covers that separate search.
Sources
- Content-Based Citation Recommendation (opens in a new tab)Bhagavatula, Feldman, Power and Ammar, NAACL 2018 (arXiv)
- Seed-driven Document Ranking for Systematic Reviews: A Reproducibility Study (opens in a new tab)Wang, Scells, Mourad and Zuccon, ECIR 2022 (arXiv)
- LoL: A Comparative Regularization Loss over Query Reformulation Losses for Pseudo-Relevance Feedback (opens in a new tab)Zhu, Pang, Lan, Shen and Cheng, SIGIR 2022 (arXiv)
- Incorporating Relevance Feedback for Information-Seeking Retrieval using Few-Shot Document Re-Ranking (opens in a new tab)Baumgärtner, Ribeiro, Reimers and Gurevych, EMNLP 2022 (arXiv)
- IAB Tech Lab Taxonomies: implementation guidance (opens in a new tab)IAB Tech Lab (GitHub)
- Contextual targeting (opens in a new tab)Google Ads Help
- Creating helpful, reliable, people-first content (last updated October 5, 2026) (opens in a new tab)Google Search Central
- Exclude specific webpages and videos (opens in a new tab)Google Ads Help
- About the "Where ads showed" report (opens in a new tab)Google Ads Help
Researched and written by PlacementGraph Research. Read how we research, source and update these pages. Corrections: diana@placementgraph.com.