A new per-pair rate database landed in my inbox this week, dressed up as a gift. Jobbers put one out for 2026 — English into Spanish, English into German, the usual busy corridors, each with a tidy average. TheWordPoint refreshed its "translation rates 2026" guide. Translayte has its cost-per-word benchmarks. All of it useful if you work a pair that shows up in the data. Mine doesn't. English to Turkmen isn't a row. It's a footnote, if that.
So let me say the obvious thing out loud: for a rare pair, the published rate table is not just unhelpful. It's actively dangerous.
The average is built from the pairs that commoditize fastest
Think about where those numbers come from. They're scraped and surveyed from high-volume corridors — the pairs that feed the biggest bulk workflows, which are exactly the pairs taking the hardest AI price pressure right now. The Slator year-in-review called AI the defining force of 2025, and the ELIS survey spent its energy asking whether the industry was holding steady or sliding back. That pressure lands first and hardest on the busy pairs, because that's where the engines are good and the volume justifies the tooling.
When a PM pulls up a benchmark and sees English-Spanish sitting at some low per-word figure, then slides over to quote me, the table has already done its damage. It's anchored them. Now I'm "expensive" against a number that has nothing to do with my work. There's no Turkmen MT worth post-editing — COMET doesn't even speak the language, and my QA panel knows it. The economics that produced that cheap Spanish number simply don't exist in my corridor. But the anchor doesn't care.
I've watched this happen in real quotes. The client isn't lying or lowballing on purpose. They genuinely think there's a market rate and they're near it. They're reading a map of a different country.
What to do when you're off the grid
Stop treating the absence of your pair as a weakness to apologize for. It's leverage, if you hold it right.
First, I never let the table set the frame. If a PM opens with "our benchmark for this is X," my answer is some version of: that benchmark is for pairs with a mature MT pipeline and a pool of a thousand linguists. Turkmen has neither. You're not buying a cheaper version of Spanish. You're buying scarcity, and scarcity doesn't discount to match abundance.
Second, I quote in outcomes, not just cents. For the busy pairs, the whole conversation has collapsed into per-word because the work has been reduced to filling segments. For mine, half the job is the stuff that never fit in a per-word cell — deciding whether a capitalized English term becomes a trap in Turkmen, whether a variable will even take the suffix the grammar demands, whether the "100% match" in the TM still holds once the surrounding sentence shifts the case. That's not post-editing. There's nothing to post-edit. It's building the target from the ground up, and I price it as construction, not cleanup.
Third — and this is the uncomfortable one — I'm fine being more expensive per word than the table and cheaper in total than the client expects. Because the real cost to them isn't my rate. It's finding a second Turkmen translator who can handle oil & gas terminology and a legal disclaimer in the same week. There isn't a bench. I am the bench. That's worth saying plainly, without smugness, as a fact about the supply.
The benchmark is still useful — just not as a price
I don't ignore these databases. I read them as weather reports. The Girletti and Lefer rates survey that came out through ASTTI in November is genuinely worth your time — not for the Swiss numbers, but for how it breaks down pricing methods and income satisfaction. That's the part that transfers. How people structure a quote tells you more than what they charge.
And I cross-check against editorial benchmarks, the EFA rates and similar, because more of my work now bundles translation with review. When MTPE became a mainstream line for the busy pairs this year, the smart move for a rare-pair translator wasn't to chase that work — it was to position as the human who signs off on everything downstream of it. The reviewer. The one whose name goes at the bottom. AI can draft the contract; it can't sign for it.
Here's the thing the rate tables will never print: a database is a tool for making a market legible, and a legible market is a cheaper market. Transparency is good for buyers. For sellers in a thin pair, being slightly illegible is the whole business. The moment English-Turkmen gets its own clean row with a confident average, the pressure starts, and I'll be arguing against my own line in someone's spreadsheet.
Until then, the blank cell is mine. I'd rather quote into empty space than into a number somebody else wrote for a language that isn't my problem to solve.