The post-editing advice that sounds noble, but may cost you – 4 June 2026 (AI Mastery for Translators Insights)

NAVIGATING THE PARADIGM SHIFT

The post-editing advice that sounds noble, but may cost you

I read an article last week arguing that translators should demand full rates for full MTPE, partly by pointing to an ISO standard. It sounds reasonable. If you’re liable for the quality, you deserve fair pay. No argument there.

And yes, there’s constant chatter on LinkedIn and elsewhere saying post-editing takes longer than translating from scratch and adds cognitive load.

But hear me out. I’d like to offer a different take.

MTPE clients aren’t asking for your finest work (and probably wouldn’t recognize it even if it slapped them across the face). They want the reader to understand what the translation says and they want the translation to be objectively correct.

That’s not the same brief, and treating it like one is where veteran translators get into trouble.

So the key question changes. It’s no longer, “Can I make this as good as I can?” It’s, “Is this accurate and readable?”

I know. Translator brains hate “good enough.” Mine does too. We poke, polish, revisit, rephrase, then call the anxiety “standards.” But truth be told, perfectionism doesn’t mean you actually deliver perfection. It often means you just never feel finished. That’s a cognitive load too!

I recently coached a translator who was losing time wrestling MT output into her personal style. Every sentence became a tiny moral crisis. Should she improve it? Rewrite it? Make it sound more like her?

We reset the rules: match effort to rate, watch the clock, fix meaning and readability, then move on.

Afterward, she called the process “super interesting and time-saving.” Not because the work magically became glamorous, but because she stopped fighting the actual assignment.

That’s the opportunity hiding inside MTPE.

It can be profitable if you stop selling invisible brilliance to buyers who only need usable accuracy. Clients are not paying for quality they cannot perceive, and they are definitely not paying for your inner critic to have a spa day.

So keep your standards. Just aim them at the real job: clear, correct, fit-for-purpose text delivered at the speed and price the market now expects.

And the silver lining?

If MTPE lets clients translate more content than they could ever afford at full human-translation rates, it can drive sustained growth in translation volume for translators too, just as free MT has already exploded the total amount of content out there being translated. The volumes are growing (even if they aren’t all carrying dollar signs).

The question is whether we position ourselves to capture it, or stand back and watch it flow around us because clients choose other resources.

EFFECTIVE PROMPTING

Don’t bother asking the AI its name.

When generative AI first made headlines in late 2022, I remember being genuinely curious about how these tools handled translation. Back then, I didn’t have a clue about what drove the translations behind the scenes, so I did what seemed natural: I asked the AI itself.

“What machine translation engine do you use?” I asked. Without hesitation, it confidently told me it relied on Google Translate for some tasks, and occasionally Bing Translate.

That caught my attention. To dig a little deeper, I asked what “pivot languages” it used when translating between Russian and Romanian. Again, it answered confidently, saying it sometimes used English, or sometimes Spanish, as the bridge. Even then, I thought those answers were pretty out there, but I didn’t yet realize the real issue.

Today, it’s a lot clearer: those answers weren’t just incorrect, they were made up—guesses based on what the AI had read in its training data. It wasn’t actually consulting Google or Bing Translate; it was just statistically picking an answer that previous texts seemed to provide.

The same thing happens now: if you ask GPT-5.4 what model it is inside CotranslatorAI, it will likely say it’s GPT-4.0 or something else that’s wrong.. The reason? The LLM only knows what was true when it was trained, so it’s still repeating answers from before it existed.

The lesson: as smart as these AIs can seem, they don’t actually have any knowledge about themselves or the real world outside their training. They mix together patterns, just like a really clever autocomplete, but you’re responsible to know what’s real and what isn’t.

If you want to know which model you’re using, don’t ask the AI—hover your mouse over the AI Settings drop-down menu. That’s the answer you can trust.

AI can do a lot, but it’s still clueless about its own identity. Don’t bother asking it its name.

COTRANSLATORAI TIP

Changing models in CotranslatorAI

In CotranslatorAI, you don’t switch the model directly in the Main Window. You switch the model inside an AI Settings Profile, then select that profile for the tab you’re working in. Here’s how to do it.

1. Open the settings

Go to Tools > Options. Click the OpenAI API tab and find the AI Settings Profiles area.

2. Select a profile

In the AI Settings Profiles list, click the profile you want to update. For example, you might select the Default profile, or a custom profile such as “Quick Draft” or “Legal Review.”

3. Choose the model

Use the Model drop-down list to select the model you want that profile to use. If you are not sure which one to choose, start with the recommended or predefined model.

4. Save the change

Click Save. That profile will now use the model you selected.

5. Use the profile

Return to the Main Window. In the Dashboard, choose that profile from the AI Settings Profiles drop-down list. That profile will apply to the active tab.

That’s it!

Stay sharp, leave a few commas alone.

Steven, for the team