> ## Documentation Index
> Fetch the complete documentation index at: https://docs.istari.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Scoring and relevance
> How GOI ranks your search results and decides what counts as a relevant match.
Every GOI search returns organizations in order of **relevance**: the best matches first. This page explains, how that ordering is decided across the [three search methods](/goi/dashboard#three-ways-to-search), and what the relevance score next to each result actually means.
## The relevance score
Most searches attach a **relevance score** to each result. It runs from **0 to 1** (often shown as a percentage), where a higher number means a closer match to what you asked for. The score value of each organization can be seen in the respective profiles of the organizations. We use it to sort the table, so the organization at the top is the one we're most confident about.
How that score is calculated depends on which search method you used.
## How each method scores
### Semantic search
When you [describe what you're looking for](/goi/semantic-search), we don't match your words literally. Instead, we turn both your description **and** every organization's profile into a numerical "meaning fingerprint" (an embedding), then measure how close each organization's fingerprint sits to yours.
* The closer the meaning, the higher the score.
* This is why a search for *"recycled aluminium producers"* can surface an organization that never uses those exact words but clearly does that work.
* Results are ranked from most to least semantically similar.
### Similarity search
[Similarity search](/goi/similarity-search) works the same way, but instead of starting from a description, we start from a **reference organization**. We build a rich profile of that organization from its public footprint, turn it into a meaning fingerprint, and find the organizations whose fingerprints sit closest to it.
If you add **prioritize** or **avoid** keywords to steer the search, we nudge that fingerprint toward the terms you want and away from the ones you don't *before* scoring, so the ranking reflects your emphasis, not just the raw reference organization.
### Keyword search
[Keyword search](/goi/keyword-search) is the one method that matches your terms **literally**. It scores each organization on how well your keywords match its profile text, using a well-established text-ranking approach (BM25). In plain terms, an organization ranks higher when:
* Your keywords appear **more often** in its profile, and
* Those keywords are **rarer** across the database overall (a distinctive term counts for more than a common one).
Because this is exact-term matching, there's no "meaning" interpretation, a profile either contains your words or it doesn't.
## When you combine methods
If you mix a description with keywords (a **hybrid** search), we run both the meaning-based search and the keyword search, then blend the two rankings so an organization that does well on both rises to the top. A **balance** control lets you lean more toward meaning or more toward exact keywords depending on what matters for your query.
## How we decide what's "relevant enough"
Ranking puts the best matches first, but we also draw a line so the results stay useful.
* For meaning-based searches (semantic and similarity), GOI applies a **relevance floor**. Organizations scoring below roughly **0.35** are left out, so you don't have to scroll past weak, loosely-related matches. Raising this floor gives you fewer but tighter results; lowering it casts a wider net.
* Keyword searches return every profile that genuinely contains your terms, ranked by match quality.
* Any [filters](/goi/advanced-search-filters) you apply: location, organization size, sector, must-include / must-exclude keywords, are applied **alongside** scoring. Filters decide *whether* an organization qualifies at all; the relevance score then decides the *order* of those that do.
As a rough guide for meaning-based scores:
| Score | What it usually means |
| --------- | ------------------------------------------------ |
| 0.35+ | Relevant: the default floor for general searches |
| 0.55–0.65 | Closely related: good for cutting noise |
| 0.8+ | Near-identical matches only |
## Why two organizations can swap places
Scores are about *closeness*, not absolute truth, so small differences near the top are normal. The further down the list you go, the looser the connection to your query. If the top results look off, it's usually faster to **refine the query or adjust your filters** than to scroll, a sharper description or an extra must-include keyword reshapes the ranking immediately.
## See also
* [Semantic search](/goi/semantic-search): meaning-based search from a description.
* [Similarity search](/goi/similarity-search): meaning-based search from a reference organization.
* [Keyword search](/goi/keyword-search): exact-term matching with BM25 ranking.
* [Search results](/goi/search-results): reading and working with the results table.