> 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.md).

# Automated Language Assessment with AI

Score a candidate's spoken and written language automatically, at scale, with objective CEFR-aligned results. You send a candidate's audio or writing, and the API returns structured proficiency scores you can compare, rank, and act on. Build an assessment in five steps.

## 1. Choose the assessment type

Pick the assessment that fits your goal. Each one has its own page with prompts, scoring, and best practices.

<table><thead><tr><th width="300">Assessment</th><th width="300">Use it to</th><th>Built with</th></tr></thead><tbody><tr><td><a href="/use-cases/automated-language-assessment-with-ai/screening-interviews.md"><strong>Screening interviews</strong></a></td><td>Measure general spoken proficiency from open-ended answers.</td><td><a href="/api-reference/score-speech.md">score/speech</a></td></tr><tr><td><a href="/use-cases/automated-language-assessment-with-ai/role-specific-assessments.md"><strong>Role-specific assessments</strong></a></td><td>Screen for a specific role with domain prompts.</td><td><a href="/api-reference/score-speech.md">score/speech</a></td></tr><tr><td><a href="/use-cases/automated-language-assessment-with-ai/skills-assessments.md"><strong>Skills assessments</strong></a></td><td>Test read-aloud delivery and situational task handling.</td><td><a href="/api-reference/score-text.md">score/text</a> · <a href="/api-reference/score-task.md">score/task</a></td></tr><tr><td><a href="/use-cases/automated-language-assessment-with-ai/writing-assessments.md"><strong>Writing assessments</strong></a></td><td>Assess professional writing (email, report, essay).</td><td><a href="/api-reference/score-writing.md">score/writing</a></td></tr><tr><td><a href="/use-cases/automated-language-assessment-with-ai/assessing-existing-interview-recordings.md"><strong>Assessing existing interview recordings</strong></a></td><td>Score a candidate from an interview you already have.</td><td><a href="/api-reference/score-speech.md">score/speech</a></td></tr></tbody></table>

## 2. Choose the question prompts

The number and type of prompts depend on the assessment type:

* **Screening and role-specific:** about five open-ended questions, each inviting roughly one minute of speech. Open-ended answers reveal language range and fluency. ([score/speech](/api-reference/score-speech.md))
* **Skills:** a read-aloud script for pronunciation and fluency, plus a situational task. ([score/text](/api-reference/score-text.md), [score/task](/api-reference/score-task.md))
* **Writing:** a written prompt matched to a `task_type` (chat, short writing, essay). ([score/writing](/api-reference/score-writing.md))
* **Existing recordings:** about five one-minute segments of the candidate speaking. ([score/speech](/api-reference/score-speech.md))

Each assessment page lists concrete example prompts.

## 3. Plan which score rubric(s) to use

Each answer is scored on several scales at once. **Use `cefr_score` together with `speechace_score`.** CEFR gives a familiar level (A1–C2) for reporting, while the numerical `speechace_score` (0–100 per metric: pronunciation, fluency, grammar, vocabulary, coherence) gives finer granularity for ranking candidates and setting thresholds. The API also returns `ielts_score`, `pte_score`, and `toeic_score` if you report on those scales. See [score/speech](/api-reference/score-speech.md) and [Interpreting results](/interpreting-results/interpreting-quality-score.md).

## 4. Plan how to compute the overall asssessment score

Combine the per-question scores into one candidate result. The API scores each answer independently, so the weighting is yours to decide:

* **Equal questions:** when every question carries the same difficulty and weight, a simple average of each metric across the answers works well.
* **Custom weighting:** when questions differ, apply your own weights. For example, if Q1 is a light ice-breaker and Q5 is the core role question, give Q1 less weight so it does not pull the result up or down unfairly.

## 5. Best practices

* **`relevance_context`** — pass the question or expected topic so the API flags off-topic answers through `relevance.class`. An off-topic answer is also marked with `response_relevance_false` (see [Scoring errors & warnings](/interpreting-results/scoring-errors-and-warnings.md)).
* **`detect_dialect=1`** — confirm the candidate answered in the target language, rather than getting a misleading score from an answer spoken in another language.