Hi Daewon Shin , 

Regarding Trados Studio 2022 with AI Professional: To help us look further into why terminology from your termbase isn’t being reflected, could you please send us the log files generated from the OpenAI plugin? These logs can help us understand what might be preventing the plugin from accessing or applying your termbase terms when generating translations in Trados 2022.

On the Terminology Handling and Custom Prompts (Trados Studio 2024): You are absolutely right, the default prompt is designed to cover general scenarios, but for more nuanced terminology handling (such as capitalization based on context), customizing the prompt is a great approach. You can update the prompt to explicitly specify your requirements, for example:

“When you use words from the connected termbase, please apply the correct capitalization according to the sentence context, even if the termbase entry is in uppercase. Adjust for singular/plural and other grammatical rules as appropriate.”

The effectiveness of this will depend somewhat on the language model you are using, but specifying clear instructions in your prompt can definitely improve the contextual handling of terminology.

Hello Patrick Andrew Hartnett ,

I sent an email with collected log files. 

Please check them out and advise me what to do.

Thank you!

Hi Daewon Shin Thanks for sharing your log files.

Thank you also for confirming that this problem is less relevant when using the plugin 'OpenAI provider for Trados Studio 2024'.
The main difference I see between AI Professional (Trados Studio 2022) and AI Assistant (Trados Studio 2024) is probably related to how each plugin passes instructions to the engine. In Studio 2024, the prompt is split into structured system and user sections, which gives the model clearer guidance and allows it to apply terminology and context more intelligently.

Regarding capitalization and the use of terms: rather than embedding long and detailed style or formatting instructions directly into the translation prompt, you’ll get better (and cheaper) results by keeping the system instructions concise and letting the Terminology Provider supply the necessary terms. The model can then use the termbase more naturally,  including adjusting capitalization based on the context.

You can still guide capitalization behavior with a short instruction such as:

“When applying terminology, adapt capitalization according to sentence context.”

This way, the prompt remains compact (around 300 tokens, as opposed to 1200 tokens that I calculated from the your data), the termbase does the heavy lifting, and cost and performance remain balanced.

Hi Patrick Andrew Hartnett ,

Thank you for the response. 

The key issue here is that the Termbase is completely ignored in Trados 2022. It seems the AI engine does not refer to the Termbase at all even it does in 2024 with the same settings. This specific issue does not seem to be related to the prompt to me, since, again, the AI engine NEVER uses the term on the Termbase. Thus, I believe there is a connection issue between the AI engine and Termbase in 2022. Here are follow-up questions:

1. Have you and your team had any chance to review this Termbase not reflected issue through the provided logs? Or is it really a prompt instruction-related issue according to your review?

2. What is Terminology Provider you mentioned above?

3. Can you please give me detailed explanations about the token consumption? I did not know that prompt itself also spends tokens. I'd like to understand general token consumption, for example, how many tokes are consumed for each letter/word, how are the "word count" is calculated in oriental languages, on what basis tokens are used in prompt, etc?

Thank you very much for your consistent support!

Hi Daewon Shin , 

1. On the logs and root cause From what I can see in your logs, terminology entries are indeed being added to the prompts in Trados Studio 2022 — but not consistently. The AI Professional plugin relies on an older integration path to access termbase data, and that can affect how reliably the terms are included.

In Trados Studio 2024, this process has been redesigned. The new AI Assistant is built on the updated AI Framework, which connects to the termbase through a modern .NET API. This provides a more stable, context-aware way of including terminology.

So, what you’re observing isn’t related to the prompt itself, but rather to differences in how deeply the terminology features are integrated between the 2022 and 2024 versions.

To confirm this more precisely, I’d need to review your actual project setup — ideally with access to the termbase and some examples of the missing terminology. But based on your logs and description, the assumptions above align with what we’d expect.

2. About the “Terminology Provider” That just refers to the default termbase in Studio that provides terms to any consumer (Editor, TM, plugins, etc.). In 2022 this is still handled via legacy COM interfaces for third-party integrations, while in 2024 it’s .NET-based — part of our move toward 64‑bit and better extensibility for developers.

3. Token usage Yes, prompts themselves also consume tokens. Think of tokens as pieces of text (about 3–4 characters in English).

• 100 tokens ≈ 75 English words
• In languages like Korean or Chinese, each word can be 1–4 tokens depending on how it’s formed. So longer prompts = higher cost and slower responses, which is why we recommend keeping them focused and letting the Terminology Provider handle the heavy lifting.

Hi Patrick Andrew Hartnett ,

I am a little bit confused if we've been on the same page or not and I'd like to summarize what we've discussed to confirm if I correctly understood.

1)The issue I've been struggling with is that the AI engine did not apply the terms from the connected termbase at all in Trados 2022 unlike 2024. The termbase here refers to the built-in Multiterm Termbase in Trados Studio, not the list of terms I put into the prompt.

2) The issue happened regardless of the API engines, and rather related to the Trados version, as I've tried different API engines (ChatGPT, DeepL, and Google Translate) and they all reflected the default termbase in 2024, but none of them did in 2022. 

3) What I am a bit confused is if you are talkin about the default termbase or the list of terms input to the prompt, and the reason of this confusion is that you said it'd be better to let the Terminology Provider instead of longer prompt, when that's exactly what I'm trying to do but have been failed with. (and the other reason is that I think I might have confused you with the logs which contain list of terms entered in the prompt..)

Based on the summary above, can you please clarify if we are talking about the same Termbase and your diagnosis was based on this understanding? Or if there was a confusion between us, it changes your answer and the way of issue handling?

By the way, I tried another test to deliver you a log file in Trados 2022 in a more controlled environment (but with the same configuration), and very strangely, the default termbase is reflected just like it is in 2024. When I tried this many times before and so did my colleagues, it never worked properly, but now this time it works. Still, I'd like to understand what happened before and what's the countermeasure to prevent it from occurring again. 

Thank you very much for your support. 

Hi Daewon Shin , would you be available for meeting via teams?  It might be better this way to clear up any confusion + allow you to demonstrate the issues you have with the plugins and configuration used. 

I will contact you via your e-mail address to confirm your  availability to schedule the teams call and discuss further

Hi Daewon Shin , would you be available for meeting via teams?  It might be better this way to clear up any confusion + allow you to demonstrate the issues you have with the plugins and configuration used. 

I will contact you via your e-mail address to confirm your  availability to schedule the teams call and discuss further