How do you choose seed sites that lead a lookalike search to the right places?
Part of Lookalike websites
The short answer
Choose seed sites that are real ad supported publishers, whose recent articles match the brief, whose appeal the client can explain, and that pass the same quality and seller checks as any planned site. Reject seeds picked only from performance reports, seeds whose topic has drifted, and seeds that are stores, brand blogs or members of a templated network.
Key takeaways
- A seed site is a template: every flaw in it is copied into the sites a lookalike search returns.
- Placement reports are a weak source of seeds, because Google says they are not exhaustive, carry thresholds and may include invalid traffic.
- Seed failures to catch early are stores posing as publishers, drifted topics, templated networks and sites that only perform well on paper.
- A short client interview about why each seed matters is the cheapest way to make the search aim at the right quality.
Why does the choice of seed sites shape every lookalike result?
Definition
- Seed site
- A website, or one section of it, that a planner uses as the reference point for a lookalike search. Candidates are judged by how closely their content resembles the seed's.
A lookalike search does exactly what the seed tells it to. If the seed is a strong specialist publication, the search returns other strong specialists. If the seed is a thinly written site that happens to rank well, the search returns more thin sites that rank well. The search cannot know which qualities the client meant and which were accidents, so it copies all of them.
That makes seed selection a high leverage step. An hour spent checking two seeds can save hours of rejecting poor candidates later, and it avoids a costly outcome: a polished list of sites that all share the seed's hidden weakness, such as a heavy ad load or a seller path the buyer cannot use.
Where do candidate seeds come from, and how far can each source be trusted?
| Source | Strength | Weakness | Check before use |
|---|---|---|---|
| The client's favorite sites | Captures what the client values in its own words | May rest on one memorable article or a personal reading habit | Ask which article made them a favorite and read the recent ones |
| Top rows of a past placement report | Based on where ads actually served | Rewards sites that score well on paper, including made for advertising sites | Review the page layout and traffic sources, not only the metrics |
| A past approved inclusion list | Already reviewed once by the team | Reasons for approval may not be recorded, and sites change | Recheck the topic, seller path and owner today |
| Editorial reputation in the vertical | Trade awards and association directories point to respected outlets | Respected outlets may carry no advertising at all | Confirm the site sells ad space, for example through an ads.txt file |
Placement reports deserve the most caution. Google describes its Where ads showed report (opens in a new tab) as a way to learn which channels, videos or websites carried ads, but says placements appear only above minimum thresholds of impressions or interactions, that the report is not an exhaustive list, that it may include invalid traffic, and that its performance results are generally not statistically significant. A seed chosen because it topped that report may simply have been the cheapest inventory with a quick click rate.
Which failure modes make a seed site mislead the search?
Bad seeds can fall into recognizable patterns. Each one leaves a recognizable trace in the results, and each can be caught before the search runs with a few minutes of checking.
| Failure mode | What the results look like | How to catch it before searching |
|---|---|---|
| Store posing as a publisher | Online shops and brand blogs crowd the candidate list | Look for product pages marked up with schema.org Product and Offer types, carts and prices beside the articles |
| Drifted topic | Candidates match what the seed covered years ago, not what the client likes now | Compare today's articles with older captures in the Internet Archive's Wayback Machine |
| Templated network member | Many near identical sites with the same layout and thin articles | Check whether the seed's footer, layout and ad slots repeat on sister domains |
| Metrics only seed | Ad heavy, refresh driven sites that read poorly | Open several articles on a phone and count how much of the screen is advertising |
| Mixed language seed | Results split between two languages, neither strong | Note the language of each recent article and seed with the matching section only |
| The client's own brand blog | Rival brands' blogs and shops that sell no ad space | Use the publications that write about the category, not the brand's own content |
Two of these checks rely on public records. The schema.org Offer type (opens in a new tab) describes an offer to transfer rights to an item or provide a service, so pages built around it describe something for sale. The Wayback Machine shows how a site looked on past dates, though the Internet Archive's help page (opens in a new tab) tells site owners it cannot guarantee that a site has been or will be archived, so a missing capture proves nothing.
What should a planner ask the client about each seed?
A client can name a seed in seconds and take longer to explain it. The explanation is what the search really needs, because it tells the planner which of the seed's qualities to preserve. Five questions, asked in a short call or an email, can surface it.
Five questions to ask about every seed site
- Ask what they last read there. A specific recent article shows whether the seed still matches the client's memory of it.
- Ask who they picture reading it. The answer names the audience the client hopes to reach, which can differ from the brief's wording.
- Ask what they would miss if it disappeared. Depth, a local focus, a trusted reviewer or a tone of voice are each a different search.
- Ask whether they would run on it today without checks. Hesitation can reveal a doubt about ad load, comments or a recent change of owner.
- Ask what would make them drop it. The answer becomes a rule for rejecting lookalikes later.
Write each answer beside the seed in one line. When candidates come back, the planner can test them against those lines instead of against a domain name. The answers also explain to the client why some candidates were rejected even though they cover the same subject.
How do you score candidate seeds before the search starts?
A simple scorecard keeps seed decisions consistent between planners and clients. Each criterion is a yes or no, and the total decides what happens to the seed. The point is not precision. It is making the reason for each seed visible before anyone builds on it.
| Test | Passes when |
|---|---|
| Publisher | The site earns from editorial content and advertising, not from selling products |
| Current fit | Most of its recent articles match the brief's theme |
| Stated reason | The client can say in one line why it matters |
| Sound pages | Its articles read well on a phone and are not dominated by ads |
| Buyable | An ads.txt file at the domain root lists at least one direct seller |
The buyable test reads the file described in the IAB Tech Lab ads.txt specification (opens in a new tab). A seed that sells only through resellers can still be a fine content reference, so a planner may keep it as a seed while noting that it may not join the plan itself.
How does PlacementGraph use seed sites?
What does seed site selection look like on a real brief?
Fictional example
A quilting supply shop in Texas
Tarrowind Quilting, a fictional fabric and quilting supply retailer with stores in Austin and San Antonio, sends its agency four favorite sites. The planner runs each through the scorecard before searching.
The first, a long running quilt pattern magazine, passes all five tests and is used as is. The second turns out to be a large craft retailer's blog, with product cards and prices beside every post; it fails the publisher test and is replaced. The third, a sewing blog the client loved years ago, now posts mostly about home organization, so only its archived quilting section is used. The fourth topped last spring's placement report, but its articles split into slideshow pages with ads covering most of a phone screen, so it is dropped.
The client call explains the gap the dropped seeds leave: the owner values pattern makers who teach techniques step by step. The planner adds a guild run tutorial site that sells ad space as a replacement seed, and the search starts from two clean seeds and one section.
When should a seed list be replaced instead of reused?
Seeds age. A seed that was right for one flight can quietly stop being right for the next, and reusing it carries old assumptions into a new list. Replacing seeds is cheap compared with reviewing a list built on a stale one.
Signs that a seed should be replaced
- The brief's product, audience or message changed since the seed was chosen.
- The seed's recent articles cover a different subject from the ones the client named.
- The seed changed owner, layout or ad load, or joined a group of sister sites.
- The seed's ads.txt file no longer lists a direct seller the buyer can reach.
- The client can no longer say in one line why the seed matters.
For the step by step search that follows once the seeds are settled, see how to find similar websites. For the review that keeps ad heavy sites out of both the seed list and the results, the MFA guide sets out the signals to check.
Frequently asked questions
Can a rival brand's website be used as a seed?
Only with care. A brand's own site sells its products and may carry no third party advertising, so a search seeded with it returns more brand sites and shops. Instead, look at the publications that review or cover the rival's products, and use those as seeds, since they publish about the category.
What if the client's favorite seed fails the checks?
Explain the specific failure, then separate the seed's role from its place on the plan. A site with weak seller paths may still be a useful content reference, while a site with a heavy ad load should not guide the search at all. Offer a replacement seed that keeps what the client valued.
Can a seed site be in a different country from the campaign?
Yes. A seed only defines the kind of content wanted. A respected specialist outlet abroad can describe the target content well even if it has few readers in the campaign's market. The lookalikes it finds still need their own audience check for the target country before they are planned.
Should seed choices be shown to the client in the final plan?
Yes. Listing each seed, the reason the client gave for it and what was changed or dropped makes the plan easier to defend. When a client later asks why a familiar site is missing or why an unfamiliar one is present, the answer traces back to the seeds.
Sources
- About the Where ads showed report (opens in a new tab)Google Ads Help
- Leading Trade Groups Define Made For Advertising Websites (opens in a new tab)ANA (Association of National Advertisers), September 26, 2023
- Offer (opens in a new tab)schema.org
- Using the Wayback Machine (opens in a new tab)Internet Archive Help Center
- ads.txt (Authorized Digital Sellers) (opens in a new tab)IAB Tech Lab
Researched and written by PlacementGraph Research. Read how we research, source and update these pages. Corrections: diana@placementgraph.com.