How do you find publishers in your own referral traffic?
Part of Lookalike websites
The short answer
Open the referral rows of your web analytics, keep the domains that sent engaged visits over a comparable period, and remove payment processors, your own domains, stores, schools and nonprofits. Confirm each remaining domain is an editorial publisher, then use the strongest as lookalike seeds at the domain level, with no visitor records copied into the brief.
Key takeaways
- A referral row shows that a site sent visits, which makes it evidence of a real reading habit, not a complete map of what customers read.
- Only aggregate, domain level fields belong in a seed brief; Google Analytics requires URL paths and parameters to be free of personally identifiable information.
- Payment processors, self referrals, webmail, social, affiliate and paid traffic look like referrals in raw data and must be filtered before anyone reads the list.
- Each surviving domain still needs a publisher check, a topic label and the same quality review as any other candidate before it becomes a seed.
How can analytics referrals reveal useful publishers?
An advertiser's own website already records which outside sites send it visitors. In Google Analytics 4, the Traffic acquisition report (opens in a new tab) "shows where new and returning users come from," and its session source and medium dimension gives "the source and medium associated with a new session." Google's default channel group (opens in a new tab) files a session under Referral when its "Medium is one of ("referral", "app", or "link")". Filter the report to that channel and switch the dimension to session source, and the rows become a list of domains whose readers clicked through to the brand.
Definition
- Referral seed
- An editorial publisher found in an advertiser's own referral data, confirmed as a real publisher and used, at the domain or section level, as an example site for a lookalike search.
The value of these rows is that nobody guessed them. A hobby blog that links to a product review or a regional magazine that mentioned the brand in a gift feature is a place where the brand's subject already lives. The guide to publisher discovery finds sites the brand has never met; referral mining finds the ones already sending it readers, strong candidates for the seed list described on the lookalike websites hub.
Which referral data is safe to use in a seed brief?
A seed brief needs domains and a sense of how much engaged traffic each one sent. It needs nothing about individual visitors. Google's policy on avoiding personally identifiable information (opens in a new tab) asks that "no data be passed to Google that Google could use or recognize as personally identifiable information," and states that "Both the URL path and parameters must be free of PII." Full referring URLs and landing page paths can still carry stray parameters, so the export for planning should stop at the domain.
Canadian guidance points the same way. The Office of the Privacy Commissioner of Canada's guidelines on online behavioural advertising (opens in a new tab) say that "Information will be about an identifiable individual where there is a serious possibility that an individual could be identified," alone or combined with other information. One visit from a tiny club site on a single date sits closer to that line than a quarterly domain total, so small rows are rolled up or dropped.
| Field | Use in the seed brief? | Reason |
|---|---|---|
| Session source (referring domain) | Yes | Names the site, not the visitor |
| Sessions and engaged sessions by domain, for a full period | Yes, as totals | Aggregate counts show which sites send real readers |
| Full referring URL with query string | No; keep the domain or a clean section path | Parameters can carry identifiers a planner never needs |
| Landing page with campaign or form parameters | No | Google asks that URL paths and parameters be free of PII |
| Client IDs, user IDs, emails, single day rows for one visitor | Never | Visitor level detail adds nothing to a site search |
Engaged sessions are the more useful count. Google defines an engaged session as one that "lasted 10 seconds or longer, or had 1 or more key events or 2 or more page or screen views." Forty engaged sessions say more about a site's readers than four hundred bounces.
How do you turn referral sites into lookalike seeds?
Eight steps from a referral export to a seed list
- Pick one comparison window. Export session source rows for the Referral channel over a full quarter or a full year, so one news mention does not dominate.
- Normalize the domains. Merge www and mobile hostnames into one registrable domain and keep a section path only when a single section sent the traffic.
- Set a floor. Drop domains below a minimum number of engaged sessions that the team writes down; the number is a planning choice, not a published standard.
- Remove the noise. Take out payment processors, your own properties, webmail, social, affiliate and tagged paid traffic, using the table in the next section.
- Confirm each domain is a publisher. Look for editorial articles, an about page and advertising; stores, schools, nonprofits and brand sites leave the list.
- Read recent articles. Write one sentence on what each of several recent pieces is about, then label the site's topic in plain words.
- Map the topic to a shared label. IAB Tech Lab's implementation guidance says buyers "should also map any specific areas of interest to the Content Taxonomy categories," which lets another planner read the seed the same way.
- Choose one to three seeds per theme. Keep the clearest examples as seeds and file the rest as ordinary candidates.
Seeds chosen this way still follow the rules in the guide to seed site selection: one theme per seed, and a section rather than a whole general site. The referring URL can point to the exact section that sent the reader, before the path is trimmed for the brief.
What referral sources should be ignored?
Raw referral rows mix editorial sites with technical and commercial traffic. Google's guide to identifying unwanted referrals (opens in a new tab) names the textbook case, "An ecommerce site that uses a third-party payment processor," where a returning buyer would otherwise count as a referral. Property owners can list such domains, up to 50 per web data stream, but an older export may predate that list.
| Source type | How it shows up | Why it is not a seed |
|---|---|---|
| Payment processors and checkout pages | Visits returning after a purchase | Part of the brand's own checkout, not a reading habit |
| The brand's own domains and subdomains | Shop, help center or blog sending visits to the main site | Self referral; it says nothing about outside publishers |
| Webmail and password recovery links | Mail domains after a reset or receipt email | Google gives a password recovery email flow as an example |
| Social networks | Filed under Organic Social when the source matches Google's list of social sites | A platform, not a publisher with its own articles |
| Affiliate and tagged paid traffic | Medium set to affiliate, cpc, display or similar values | Bought or commissioned traffic, which shows what the brand paid for, not what readers chose |
| Stores, marketplaces, schools, nonprofits and government | Retailer listings, course pages, charity event pages | Real referrers, but not ad supported publishers a plan can buy |
Paid and affiliate rows can sit in the Referral channel when tags are missing. Google's channel rules file a session under Affiliates when "Medium = affiliate" and under Display when the medium is one of "display", "banner", "expandable", "interstitial" or "cpm". An untagged sponsored post shows up as a plain referral, so check the media plan for any site that also appears as a referrer.
How often should referral data be mined?
Google Analytics date range comparisons (opens in a new tab) include "Previous period," "Previous period (match day of week)" and "Previous year," and charts can be viewed "by day, week, or month." Comparing the same window each time shows whether a referrer is new, growing or fading.
Referral review routine
- The export covers the same channel, dimension and length of period as the last review.
- The comparison with the previous year is attached, so seasonal spikes are visible.
- New referrers above the floor are opened and checked as publishers before anyone discusses them.
- Seeds whose referrals fell to near zero are re-read to see whether the site changed or simply stopped linking.
- The seed list records the export date, the floor used and who ran the review.
How does PlacementGraph use referral sites as seeds?
What does referral mining look like on a real brief?
Fictional example
A fictional string instrument rental shop in southern New England
An unnamed shop that rents violins, violas and cellos to students in Connecticut and Rhode Island wants a back to school display plan for Hartford, New Haven and Providence. Its agency exports a year of session source rows for the Referral channel and finds 140 referring domains.
Below the written floor of engaged sessions, 96 domains drop out. A payment processor, the shop's own repair blog and two webmail domains go next. A youth orchestra, a music school and a school district are real referrers but nonprofit or education sites, and a sheet music store is a store, so all four leave the publisher list.
Nine editorial sites remain: string teaching blogs, a regional arts magazine's music section and a practice guide site for parents of young players. The planner reads recent articles on each, labels two themes, string pedagogy and family music life, and keeps two seeds per theme. The other five join the candidate pool. The brief lists domains and theme labels only.
Frequently asked questions
Does a referral from a site mean its readers are the brand's customers?
No. A referral shows that some readers of that site clicked through to the brand during the period selected. It says nothing reliable about who those readers are, so do not describe the site's audience in demographic terms because of it. Treat it as evidence that the subject fits.
What if the strongest referrer has no advertising?
It can still describe the theme as an example page, because a lookalike search compares content, not ad slots. It will not appear as a buyable placement, so the plan has to find ad supported sites like it. A direct sponsorship with its owner is a separate decision outside the programmatic plan.
How do you tell an earned referral from a paid one?
Check the medium and the media plan. Google files tagged traffic with mediums such as affiliate, cpc or display outside the Referral channel, but an untagged sponsored article lands there as a plain referral. Compare each referring domain with past insertion orders and affiliate agreements.
Can referral data from one country seed a plan in another?
It can seed the search, because subject matter travels across borders. Whether a referring site counts for the new market is a separate question about its audience there. A US site that sends visits to a Canadian brand may also have readers in Canada, but that has to be checked against audience data for the target country.
Sources
- [GA4] Traffic acquisition report (opens in a new tab)Google Analytics Help
- Default channel group (opens in a new tab)Google Analytics Help
- Identify unwanted referrals (opens in a new tab)Google Analytics Help
- Best practices to avoid sending Personally Identifiable Information (PII) (opens in a new tab)Google Analytics Help
- Change and compare date ranges in reports (opens in a new tab)Google Analytics Help
- Guidelines on privacy and online behavioural advertising (opens in a new tab)Office of the Privacy Commissioner of Canada (modified 2025-08-11)
- IAB Tech Lab Taxonomies: implementation guidance (opens in a new tab)IAB Tech Lab (GitHub)
Researched and written by PlacementGraph Research. Read how we research, source and update these pages. Corrections: diana@placementgraph.com.