If you research a buyer’s options, you need to know which businesses and publications appear under those buying conditions. LLM Top Sites lists websites named during the campaign's model research. Comparing models and personas shows which organizations appear under different buying conditions.
Read Score and Seen separately
Seen counts how often a site was named across the collected responses. Score weights each appearance by the model's configured weight and by the site's position in the response.
The difference between Seen and Score helps explain how a site appears. A site named in most responses but always near the end produces a high Seen count and a modest Score. A site named in fewer responses but always first produces the reverse.
Read the first as broadly known and rarely the recommendation. Read the second as the answer for a narrower question, which is often a more precise competitor than the ubiquitous name.
Labels and unfamiliar hosts
Sites are grouped by normalized host. Organization labels identify known owned brands, competitors, publishers and partners.
An unlabeled result is a website to classify, not automatically a competitor. Publications, review sites, associations and directories all appear here, and mislabeling a publisher as a competitor distorts every report that uses the competitor list.
Open unfamiliar websites to determine what they provide, then add the relevant ones to the campaign so other reports apply the same labels.
Use personas to examine a different decision
A company buying at volume can receive different recommendations from an individual making an occasional purchase. Select each persona and inspect which sites change position or disappear.
Three patterns are worth looking for:
- A site present for every persona is established across the market's written record.
- A site present for one persona only tells you which buyer that company is associated with in the material, which is useful whether it is a competitor or a publication.
- Your own absence for the persona you sell to most is the most actionable finding this report produces.
This reads setup research, not a live series
The report reads campaign research collected at setup. Reopening it does not collect a new round of recommendations.
Use it to understand the market when a campaign begins, and use prompt tracking for repeated observations over time. Treating this as a trend line is the one misuse to avoid, because there is no second collection behind it.
When a returned site changes your research plan, record the buyer requirement and the page you need to investigate.
See finding which companies answers name, how personas change the answer, or tracking over time.