Every firm profile shows two similarity views: Firms with Similar Tech and Firms with Similar Websites. Toggle ⬆ Blend to combine both signals into a single ranking. Here’s what each one measures and when to use it.
Website similarity reads the actual text of a firm’s pages — practice area descriptions, attorney bios, location copy, brand voice — and encodes it as a high-dimensional vector using the same AI embedding model used for tech profiles. The result is a single content fingerprint per firm, called a centroid, computed from all indexed pages.
Because the model reads meaning, not keywords, it surfaces firms that sound alike and serve similar clients — even when their technology stacks are completely different.
| Match % | What it means |
|---|---|
| ≥ 90% | Near-identical positioning — same geography, same practice mix, same tone |
| 80–89% | Strong overlap — same client type, similar story, competing for the same searches |
| 70–79% | Moderate overlap — shared emphasis on certain practice areas or markets |
| < 70% | Different positioning — may share a practice area but tell a different story |
A mid-size personal injury firm in Phoenix emphasizes auto accidents, Spanish-language outreach, and a “no fee unless you win” promise. Website similarity will surface other Phoenix-area PI firms that use the same language and serve the same demographic — regardless of whether they run the same CRM or ad platform. These are the firms a prospective client would compare side-by-side when searching Google.
Blend pre-computes a single merged vector for each firm by averaging two normalized signals in the same embedding space:
Both vectors live in the same 3,072-dimensional space, so averaging them is mathematically valid — the blended vector points in the direction both signals agree on. The result is searched with the same nearest-neighbor index used for tech and website views, so the ranking reflects firms that are genuinely close across both dimensions at once, not just firms that scored well on one.
Firms that only have one signal (tech scan but no content index, or vice versa) are included using the available signal as a fallback, but they will naturally score lower than firms where both signals point in the same direction.
Suppose a firm has a tech similarity score of 94% and a website similarity score of 81% against a competitor. In Blend view, those two vectors are combined before the search runs, so you’re finding firms where the combination is closest — not adding scores after the fact. A second competitor that scores 87% on tech and 86% on website may outrank the first in Blend even though neither individual score was higher, because its combined direction is tighter.
| Firm | Tech match | Website match | Blend ranking |
|---|---|---|---|
| Competitor A | 94% | 81% | Ranked #2 — strong tech signal pulls toward different content territory |
| Competitor B | 87% | 86% | Ranked #1 — balanced signals, closest overall competitor |
| Competitor C | 79% | 91% | Ranked #3 — strong content match but divergent tools pull score down |
| Question | Best view |
|---|---|
| Who sounds like us to a prospective client searching online? | Website |
| Who is positioning against the same practice areas and markets? | Website |
| Who runs the same software stack as us? | Tech |
| Who would respond to the same vendor pitch? | Tech |
| Who is our closest overall competitor — same tools and same story? | Blend |
| Which firms should we benchmark pricing against? | Blend |
| Who is most likely fighting for the same clients with the same infrastructure? | Blend |
Website similarity requires that a firm’s pages have been fetched and indexed. The content fingerprint gets richer as more inner pages are added — practice area pages, attorney bios, location pages — which makes both the Website and Blend rankings more precise automatically, without any manual update.
Blend vectors are pre-computed and indexed server-side. Toggling Blend is instant — no additional query runs when you click it.