# How AI tender evaluation works in Tender Navigator

Source: Tender Navigator Knowledge Base, https://tendernavigator.com/knowledge-base/how-ai-tender-evaluation-works (canonical HTML). Category: AI features. Published 2026-09-08, last updated 2026-09-08.

> Tender Navigator evaluates a tender by sending the tender notice and the filter's AI prompt to a large language model, which returns a verdict, Matched or Unmatched, and a short reasoning that explains the verdict against the prompt's criteria. Evaluation runs daily for every active filter and is never triggered by a search.

## The one-sentence version

Tender Navigator evaluates a tender against the criteria configured in your filter and your business description. The resulting evaluation explains why the tender appears relevant or irrelevant.

## What is evaluated

Only tenders picked up by an **active filter** are evaluated. A filter's keywords, countries, classification codes, deadline offset and budget rule decide which new tenders it reads each day; each of those tenders is then evaluated once, by that filter. The Pre-search tool never evaluates anything, which is why its rows say **Not evaluated**.

The same tender can be evaluated by two different filters and receive two different verdicts, because each filter has its own prompt.

## What the model receives

- The tender notice as Tender Navigator holds it: title, description, buyer country, dates, classification codes and the other fields shown on the tender page.
- The filter's **AI prompt**: your description of what makes a tender a good or bad fit. See [Writing the AI prompt](/knowledge-base/writing-the-ai-prompt).

The model does not receive your account details, your other filters, or your past verdicts. It does not browse the procurement portal or open attached documents; it works from the notice text Tender Navigator ingested.

## What the model returns

- A **verdict**: Matched or Unmatched.
- A **reasoning**: a short written explanation that refers to the criteria in your prompt.

Both are stored as a result and shown on the Results page, in the Tender Info panel, in the digest email and on a shared tender page. See [Understanding Matched and Unmatched results](/knowledge-base/understanding-evaluation-results).

## Which model does the judging

Each filter chooses its **LLM model** from a list of general-purpose models from third-party providers. The default is the first model in the list. See [Choosing the AI model](/knowledge-base/ai-models). Tender Navigator's AI providers are listed in its sub-processor record, available on request from legal@synergysoft.ch, as the [AI Transparency Notice](/legal?v=ai) explains.

## When evaluation runs

Filtering and evaluation run **daily** for every active filter on every plan. There is no on-demand evaluation of a single notice: the **Evaluate** button on a search result explains this and offers Save as Filter instead. The digest email schedule is separate and only controls when results are emailed.

## What it costs

Every evaluation counts against the evaluations included in your plan for the billing cycle. When they are used up, further tenders are recorded as stuck rather than evaluated. See [Filter usage and limits](/knowledge-base/filter-usage-and-limits).

## What evaluation is not

AI evaluation is an advisory pre-screen. It does not decide whether to bid, does not check eligibility against the full tender documents, and can be wrong. The reasoning is there so that a person can check it quickly against the original notice. See [Limitations of AI evaluation](/knowledge-base/limitations-of-ai-evaluation).