Half of Freelance Linguists Have Considered Leaving the Field: What the Ones Staying Are Doing Differently

Every few months someone writes to us about a freelance translator career change and asks for a second opinion before they commit to it. The messages have a pattern. It is rarely one bad year that pushes a person out, and it is almost never enthusiasm for AI that keeps a person in. Survey work circulated through Slator and run by professional associations over the past two years has put the share of freelance linguists who have at least considered leaving the profession somewhere around half, and that figure now shapes how translators talk to each other about their own prospects. We have spent enough time with people on both sides of that decision to notice what separates them, and it is less dramatic than the discourse suggests.

What the survey numbers actually say about a freelance translator career change

The headline number is real, and it is also softer than it sounds. "Considered leaving" is a low bar. It covers the translator who spent one evening reading about teaching jobs and it covers the translator who has already registered a company in another sector. Those two people are not in the same situation, and survey instruments almost never separate them.

Two methodological problems matter here. The first is self-selection. Surveys of freelance linguists are usually distributed through professional bodies, mailing lists, and industry media, and people who are angry or worried answer at higher rates than people who are quietly busy. That skews the result in a predictable direction. The second is that the question is asked during a period of loud narrative about AI, which primes the answer. Ask about career doubt in any freelance profession during a technology shift and you will get elevated numbers.

None of that makes the finding meaningless. Rate pressure on generic content is documented, and CSA Research and Nimdzi have both tracked the shift in what buyers are willing to pay for undifferentiated volume. What the survey does not establish is that half the profession is leaving. It establishes that half the profession has thought about it, which is a statement about sentiment.

The useful move is to stop treating the number as a forecast and start treating it as a description of a market sorting itself. Work that was priced as a commodity is being repriced. Work that was never a commodity is mostly holding. Where an individual translator sits relative to that line predicts their next three years better than any survey aggregate does. That is the question worth answering about your own practice, and it is answerable with your own invoices.

The people who stayed stopped selling words

The single most common thing we hear from translators whose income held is that they changed the unit they sell. Per-word pricing made sense when the work was linear in word count. For a growing share of jobs it no longer is.

One example from a conversation last spring: a German to English translator working mostly on pharmaceutical regulatory documents had a long-standing client cut its per-word rate on the grounds that machine output was now the starting point. She did not accept and did not walk away. She quoted the same work as a per-document fee based on her own estimate of hours, with the raw output explicitly out of scope as the client's input rather than her deliverable. The annual value of that account came out slightly above where it had been. What changed was that variance in document difficulty stopped being absorbed silently by her.

A second example runs the other way. A Spanish to English generalist we spoke with in 2025 held per-word pricing through two rate cuts because switching felt risky. His volume rose and his income fell, which is the failure mode of per-word billing when effort per word is climbing.

This works best when you can estimate your own hours reliably, which requires having tracked them. It does not apply cleanly to agency work where the purchase order system only accepts a word rate, and plenty of good agency relationships are built on that constraint. Even there, a minimum charge and a separate line for terminology or formatting work recovers part of the gap. The principle holds regardless of the mechanism: bill for the thing that actually consumes your time.

Narrower specialization did more than broader service menus

The instinct under pressure is to widen: add subtitling, add copywriting, add another language pair. Most of the translators we have watched do well went the opposite direction.

Narrow specialization is defensible against automated output for a specific reason. A general-purpose model produces fluent text in almost any domain. What it does not have is the client's history, the regulator's current position, or the reason a term was fought over three years ago and settled a particular way. That knowledge lives in a person or in a maintained glossary, and it is what buyers of high-stakes content are paying for when they pay above market.

The practical version is unglamorous. It means picking a domain narrow enough to describe in one sentence, then building the terminology asset that proves you have been there. One medical device translator we know keeps a per-client glossary with the approval history attached: which term was rejected by which reviewer and when. When a new project manager arrives at that client, that document makes replacing her expensive. Her rate has not moved down since 2023.

The limitation is honest and worth stating. Narrowing reduces your addressable market, and if your chosen domain contracts you have less to fall back on. Translators who specialize into a single client's subject matter rather than a market's subject matter are exposed. The distinction between "I work on cardiac device documentation" and "I work for one cardiac device manufacturer" is the difference between a niche and a dependency.

Specialization also takes longer than people expect to show up in revenue. Eighteen months is a reasonable planning horizon before a repositioned profile changes who contacts you. That lag is a large part of why the people who considered leaving and then did not are usually the ones who started the shift earlier.

Post-editing is priced as a service or it is priced badly

MTPE is where a lot of the career doubt concentrates, and often for a fixable reason. The standard discount grid was designed for translation memory reuse, where a 90 percent fuzzy match genuinely takes less work. Machine output does not behave that way. Effort is not proportional to surface similarity, because a fluent wrong sentence takes longer to fix than an obviously broken one.

Translators who kept post-editing profitable did two things. They separated light post-editing from full post-editing in writing, with a definition of each attached to the quote, and they priced full post-editing hourly or per document rather than at a fixed discount off their translation rate. The second part only works if the first part exists, because without a written scope the client will assume the higher standard at the lower price.

The other change is upstream. Post-editing effort depends heavily on the quality of the raw output, and the raw output depends on the glossary and the instructions used to produce it. Translators who get a say in that preparation step spend measurably less time fixing terminology afterward. Those who receive output generated with no glossary are being asked to repair a preventable problem at their own cost. Asking to see or supply the glossary before the run happens is a reasonable condition of the engagement, and in our experience clients agree more often than translators expect, because the request is obviously in their interest too.

This does not apply if you are working through a platform that hands you segments with no visibility upstream. In that setup the only levers are the rate and whether you accept the job. Some translators have concluded that the rate is not recoverable and stopped accepting that category of work entirely, which is a legitimate answer rather than a failure.

Direct clients changed the math more than any tool did

Nearly every translator we have talked to whose income grew since 2024 has more direct clients than they had in 2022. This is the least surprising finding and the hardest one to act on.

The margin argument is simple arithmetic. An agency taking a normal cut means the same end-client budget supports a substantially higher rate when there is no intermediary. The work you take on in exchange is real: quoting, scoping, chasing payment, and explaining to a non-specialist why their file cannot be turned around by tomorrow. Some translators find that work draining enough that the agency cut is worth paying, and that is a defensible position rather than a lack of ambition.

What has shifted is that the barrier to being found by direct clients dropped. Companies that never had a translation budget now have documents they need handled and no idea who to ask. A supplier manual, an HR policy, a technical specification arriving from a partner. These buyers search in plain language and they do not know the vocabulary of the industry. A translator with a page that describes the exact document type they handle, in the words a non-specialist would use, gets found by people who would never have appeared through an agency channel.

The counterweight: direct clients are slower to acquire and lumpier in volume. Building that side of a practice while agency work is still paying the bills is the version that works. Doing it after the agency work has already dried up is a much harder position, which is the argument for starting during a decent year rather than a bad one. Our own reading of how AI translation tools are changing the way translators work points the same direction, since the tasks being automated first are the ones agencies were charging most for coordinating.

The admin hours are where the exit decisions get made

When translators describe wanting out, the reason they give is usually rates. When they describe their week, the thing that has grown is administration. Those two facts are related and the second one is more fixable.

The pattern is that fragmentation has increased. More clients, smaller jobs, more file formats arriving, more back-and-forth on scope. A practice with four agency clients sending steady volume has a low coordination cost. The same revenue spread across fifteen direct clients sending occasional documents does not, and nobody bills for the difference.

Two things reduce this without much effort. The first is a written intake process, so that a new job arrives with the source file, the target language, the deadline, and the reference material in one message instead of six. A short form or even a saved email template does most of the work. The second is a minimum charge that makes small jobs worth accepting. A 300-word certificate takes nearly as much administrative time as a 3,000-word report, and pricing it purely by word guarantees a loss on the coordination.

Track this before you conclude your rates are the problem. Log a normal month with a timer running on everything, including email and file handling, then divide actual revenue by total hours. In most cases we have seen the number is well below what people assume, and the gap sits in unbilled coordination rather than in the translation rate. That measurement changes the decision. A translator earning a decent rate but losing a third of their week to unbilled work is not in the same position as one whose rate has genuinely collapsed, and the two situations call for different responses.

What to do in the next quarter if you are on the fence

Pick one month and measure it properly. Every hour, tagged by client and by activity, including the parts that feel too small to record. At the end, calculate your real hourly rate per client rather than your nominal per-word rate. That single number tells you whether your problem is pricing, positioning, or coordination, and those have different fixes.

Then act on the worst client rather than on the whole practice. Requote them per project, add a minimum charge, or replace them. Doing this once with a single account is reversible and teaches you more than a year of reading about industry trends. The translators who stayed did not make one decisive move. They made a series of small ones, starting before the situation forced it, which is the only part of this that is genuinely hard to copy after the fact.

If you want a wider view of how the work itself is shifting, our blog covers the tooling and workflow side of the same question in more detail.