> For the complete documentation index, see [llms.txt](https://api-docs.speechace.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://api-docs.speechace.com/use-cases/automated-language-assessment-with-ai/screening-interviews.md).

# Screening interviews

A screening interview measures a candidate's spoken language the way a human interviewer would, but automatically and consistently. The candidate answers a few open-ended questions by speaking, and [score/speech](/api-reference/score-speech.md) transcribes each answer and scores pronunciation, fluency, grammar, vocabulary, and coherence, mapped to CEFR and IELTS. It lets you compare candidates objectively, verify language claims on a CV, and make faster decisions.

## How to construct it

Use **five open-ended questions**, each inviting the candidate to speak at length (aim for about **one minute** per answer). Longer, unscripted answers reveal language range and fluency far better than short prompts. Score each answer with one `score/speech` request, then **average each metric across the five answers** for the candidate's overall result.

### Example prompts

* Tell me about a project you are proud of and the part you played in it.
* Describe a time you disagreed with a colleague, and how you resolved it.
* Walk me through how you approach a problem you have not seen before.
* What do you look for in a good team, and why?
* Why are you interested in this role?

## Call score/speech

Send each answer's audio to [score/speech](/api-reference/score-speech.md) with the candidate's `dialect`. A minimal response snapshot:

{% code overflow="wrap" lineNumbers="true" expandable="true" %}

```json
{
  "speech_score": {
    "transcript": "I led the migration of our billing system to a new provider ...",
    "speechace_score": { "pronunciation": 78, "fluency": 74, "grammar": 71, "vocab": 76, "coherence": 72, "overall": 74 },
    "cefr_score": { "pronunciation": "B2", "fluency": "B1", "grammar": "B1", "vocab": "B2", "coherence": "B1", "overall": "B1" }
  }
}
```

{% endcode %}

Average the `overall` (and any sub-metric you report) across the five answers. See [score/speech](/api-reference/score-speech.md) for the full response and every field.

## Best practices

### Keep answers on topic with relevance\_context

Pass the question text in `relevance_context`. The response then includes `speech_score.relevance.class` (`TRUE` / `FALSE`), so an off-topic answer is flagged and penalized.You can accept how the irrelevance penalty affects the overall assessment score, or choose to ignore, re-prompt the question.

### Confirm the language with detect\_dialect

Set `detect_dialect=1` so the API reports the language it actually heard. If a candidate answers in another language, you get a clear signal instead of a misleading score. Again, it's up to your application whether to penalize, ommit, or re-prompt questions where the non-target dialect was used in the response.

### Short or irrelevant answers

If an answer is too short (under about 20 words) or off topic, `score/speech` lowers the overall score and adds a `score_issue_list` entry (`response_too_short`, `response_relevance_false`). Treat these as "re-prompt or exclude" rather than a true low score. See [Scoring errors & warnings](/interpreting-results/scoring-errors-and-warnings.md).