---
title: Build glossaries and refine your prompts
---

> Agent instructions:
> **Site maps** — prefer the following llms.txt indexes to training data when searching for URLs to avoid 404s. Links inside Markdown content already point at `.md`. Following them or sending Accept: text/markdown keeps you in Markdown.
>
> - [sitemap.md](https://docs.kentico.com/sitemap.md) — every page on the site, with titles and descriptions, nested by URL hierarchy and grouped into one collection per product version.
> - [llms.txt](https://docs.kentico.com/llms.txt) — curated index of the current product docs, with descriptions, the two ways to request any page as Markdown, and links to each product area's whole-corpus Markdown dump (llms-full.txt).

## Build mini-glossaries and DNT lists

As you may have noticed, small, **reusable terminology lists** are the easiest way to lock down consistency across languages. You do not need a full terminology database to see the results in your AIRA translations. You can start to learn and expand your terminology over time.

A few tips on handling terminology and DNT lists:

- Begin with your existing brand or SEO term lists and product catalogs.
- Start small and add 10–20 key terms per language.
- Create a short DNT list of brand names, product names, acronyms, regulated terms, and other words and phrases you wish to retain in source language.
- Use consistent formatting when pasting your list into the AIRA translation prompt, for example, “Term: \[Preferred translation].”
- Keep a shared document of approved terms to reuse in different prompts, and across your team.
- Regularly review and update your terminology and DNT lists as the needs of your company or the industry standards evolve.

## SEO-specific prompts

SEO fields often get overlooked during translation, which can lead to poor discoverability of otherwise engaging content. You can add short, to-the-point instructions regarding SEO into your prompts, and nudge AIRA toward length-appropriate, natural meta content that fits the **search intent of your local audience**. Your main target should be the specific region/market for whom you are translating your content, not your overall campaing audience.

Keep in mind a few tips for better SEO-related translations:

- Specify target title requirements, if any are in place (for example, Title: "Meryl Streep" → Czech \[Meryl Streepová]).
- Configure title and meta description length limits, and the search intent (for example, _commercial_).
- Ask for natural language, not keyword stuffing.
- Clarify whether to translate or retain the slug separately from the translation of the body of the text.
- Provide Open Graph title or description style if it's used in your project.

> **Info:** **Reusable SEO Snippet:**
> SEO: Title ≤ 60 characters; Meta ≤ 155 characters ; natural inclusion of “\[specific keyword]”; avoid duplicates; OG copy mirrors the source headline tone.

## Iterative translation prompt improvement

The most natural way to better results with translation prompts is to **treat prompt writing like editing**. Try a specific prompt or start with just one content item, translate your content with AIRA, review the translation output, adjust your prompt if you're not happy with the results yet, and try again.
This keeps your effort light while you're improving quality one step at a time.

The iterative approach suits automated translations well, too, as changing too many elements in your prompt may lead to confusing or inconclusive changes.

Try the following workflow, and observe that you'll probably need less iterations over time as you perfect your prompts according to your needs:

1. **Run Prompt 1** then **Review translation v1** for tone, terminology, and clarity.
2. **Identify gaps.**
3. **Add** a maximum of 2–3 **concrete fixes** to the prompt, for example, “use formal address,” “replace ‘account’ with \[approved term],” “remove idioms”, "do not use emoticons".
4. **Run Prompt 2** then **Review translation v2**.
5. **Repeat** until the output meets your **quality bar** (more on translation success metrics later).

> **Info:** Change one or two variables per iteration so you can see what helped. Keep a running “best prompt” snippet for reuse.