Choosing a translator or translation company is not a shopping exercise in which the lowest price automatically wins. Professional translator selection is a risk decision: you are choosing who will interpret meaning, handle terminology, protect confidential material, resolve ambiguity, work inside your technology, meet deadlines, accept review, and ultimately put words in front of your customers, students, patients, employees, regulators or public audience. A strong selection process therefore looks for evidence of fit rather than a vague promise of “high quality.”
Searches for how to choose a translator, how to choose a translation company, translation provider selection, translation vendor evaluation, translator test translation, translation quality criteria, translation agency comparison, professional translation services and translation procurement checklist all point to the same practical problem. Buyers need a way to compare providers whose quotations may look similar while their actual processes, subject expertise, review discipline, security controls and ability to learn the project are very different.
This guide gives you that way. It explains how to define the work before asking for quotations, screen translators for real domain competence, design a fair test translation, examine revision and quality-assurance processes, assess technology and security without being dazzled by tool names, compare prices on the same scope, run references intelligently, score evidence consistently, and begin with a pilot before committing a large multilingual programme. The aim is not to find the theoretically “best” translator in the world. It is to select the provider whose capability matches the job you actually have.
50-second quick read
If you remember only one idea, remember this: select for the work, not for the sales pitch. Start by specifying audience, content type, language pair, subject matter, volume, deadlines, formats, risk, security and required review. Ask candidates to show evidence against those needs. Use a representative test only after you have defined what the test is measuring. Compare total scope, not headline per-word rates. Check how candidates handle questions, terminology, corrections and source changes. For important work, begin with a bounded pilot and evaluate the real workflow before scaling.
A professional provider should be able to explain who translates, who revises, how terminology is controlled, what happens when the source is ambiguous, how files and confidential data are handled, how quality is checked, what evidence is recorded, and how corrections propagate. The answers matter more than the number of logos on a capability deck.
1. Begin with a translation specification, not a vendor list
The quality of a selection process is limited by the quality of the question you ask the market. “We need 40,000 words translated into three languages by Friday” describes volume and deadline but not the communicative problem. A provider cannot prove fit if you have not defined what fit means.
Write a short translation specification before contacting anyone. State the source and target languages, target locale where relevant, document types, subject matter, audience, purpose, approximate volume, recurring or one-off nature, file formats, content-management environment, required delivery format, deadline pattern, source stability, confidentiality level, and the consequence of error. Then add the quality expectations: terminology control, bilingual revision, subject-matter review, monolingual editing, final proofreading, in-context testing, or any other stage genuinely required.
This is where the wider Master Art of Translation architecture matters. Translation is a system. Provider selection should therefore test whether the provider can operate the system you need rather than merely produce sentences.
2. Separate language-pair competence from subject competence
Being fluent in two languages is necessary for many translation tasks, but it is not sufficient. A translator can understand ordinary language perfectly and still misunderstand a patent claim, a financial covenant, a medical instruction, a software string, a curriculum standard or a public-policy term. Domain competence changes what the translator notices.
Ask how the candidate acquired relevant subject knowledge and how current that knowledge is. Evidence can include prior projects in the domain, formal education, professional experience, specialised training, published work, client references, terminology resources, or a demonstrated ability to research unfamiliar concepts. The goal is not credential collecting for its own sake. The goal is to reduce the probability that the translator will produce a plausible sentence that is conceptually wrong.
For highly specialised work, ask candidates to explain two or three domain problems they would expect to encounter. Someone who has genuinely translated clinical-trial material, for example, can usually describe recurring terminology, register, consent, protocol and consistency issues without revealing confidential client information. That explanation often tells you more than a generic claim of “medical expertise.”
3. Decide whether you need an individual translator, a team or a language-service provider
Different operating models solve different problems. An experienced individual translator can offer continuity, direct communication and deep ownership of a text. A small specialist team can add revision and resilience while preserving close collaboration. A larger language-service provider may be better when you need many languages, production management, engineering, desktop publishing, continuous localisation or rapid scaling.
Do not assume that bigger means better or that direct freelancer relationships are always cheaper. Compare the coordination burden as well as the linguistic work. If your internal team would need to recruit five languages, manage deadlines, merge files, maintain termbases and resolve reviewer feedback, a provider that handles those layers may be creating real value. Conversely, paying for a complex production organisation can be unnecessary when one expert translator can handle a stable specialist publication from start to finish.
The key is to match operating model to project shape. Ask who owns the relationship, who owns terminology, who answers queries, who substitutes if a translator becomes unavailable, and how continuity is protected when people change.
4. Understand what professional standards can and cannot tell you
Standards can help buyers ask better questions, but a certificate should not replace judgment. ISO 11669:2024 provides general guidance across translation projects, including specification, communication, needs analysis, risk and workflow. ISO 17100:2015 remains the published edition of the translation-services requirements standard as of September 2026, while ISO is developing a second edition. These references are useful because they make processes discussable: resources, competences, specifications, revision and project management can be examined rather than left as vague assurances.
However, conformity does not automatically prove that a provider is right for your content. A well-controlled general process does not create subject expertise where none exists. Likewise, a superb specialist translator may produce excellent work without presenting a certification badge. Use standards as evidence about process maturity, not as a substitute for testing fit.
Ask what the provider actually does on your project. Which steps are mandatory? Which are optional? Who performs revision? How is the translator selected? What happens to urgent work? Which quality checks are automated, which are human, and which are not part of the quoted scope?
5. Examine who will actually do the translation
One of the most important procurement questions is often skipped: who is the linguist? A strong account manager can understand your business beautifully, but the translator will make the sentence-level decisions. If the provider uses a network, ask how linguists are recruited, tested, assigned and retained.
You do not need personal data that is irrelevant to the project. You do need confidence that the assigned people have appropriate language, domain and tool competence. Ask whether the same core linguists will remain on recurring work. Continuity is especially valuable because translators learn your terminology, products, risk points and preferences over time. A rotating pool may be efficient for commodity text but costly for nuanced or highly controlled content.
If subcontracting is permitted, make that transparent. Ask whether further subcontracting can occur, whether confidentiality obligations flow down, and whether you can approve or reject changes to the core team for high-risk work.
6. Ask how revision works, not merely whether it exists
“Includes proofreading” can mean almost anything. The reviewer may compare source and target carefully, skim the target for grammar, run an automated checker, or simply look at formatting. These are different activities.
Ask the provider to describe the review path in operational language. Does a second qualified person compare source and target? Is terminology checked separately? Are names, numbers and units verified? Is there a target-language editing pass? Who resolves disagreements between translator and reviewer? Does the final translator see the review and learn from it? What changes for safety-critical or regulated content?
A good answer distinguishes error prevention, bilingual revision, monolingual editing, proofreading and technical QA. The provider does not need every possible stage on every job. It does need to know which stage solves which risk.
7. Design a test translation that resembles the real work
A test translation is useful only when it measures the job you intend to buy. A 200-word marketing passage is a poor test for a 500-page technical manual. A clean paragraph with no terminology or formatting constraints tells you little about how a candidate will manage messy real files.
Select a short but representative sample containing the decisions that matter: domain terminology, ambiguity, recurring phrases, numbers, headings, UI labels, tables, tone shifts or culturally specific references. Supply the same brief, glossary and context that a real translator would receive. If the real project requires screenshots or product access, the test should too. Otherwise you are measuring performance under artificially bad conditions.
Tell candidates how the test will be evaluated. Do not build a trap full of secret preferences. A fair test distinguishes genuine errors from acceptable alternatives. If your organisation has mandatory terminology or style choices, give them in advance.
8. Score the test on dimensions, not a single feeling
“This one reads better” is not a sufficient evaluation method. Break the test into dimensions. A practical rubric can include accuracy of meaning, completeness, terminology, names and factual details, grammar, naturalness, register, consistency, formatting, adherence to instructions, and quality of any translator queries.
Weight the dimensions according to the project. A brand campaign may place more weight on voice and persuasive function. A safety procedure should place far more weight on unambiguous meaning, numbers, conditions and terminology. A software interface needs action clarity and context fit. The test should reflect the consequences of error rather than an abstract ideal of elegance.
ISO 5060:2024 provides current general guidance on evaluating translation output and discusses analytic evaluation using error types and penalties. You do not have to reproduce a formal standard to benefit from the principle: define categories, define severity, and judge against evidence rather than taste.
9. Include the translator’s questions in the evaluation
Strong translators ask good questions. A candidate who identifies that a term has two plausible senses, notices that a table contradicts the body text, or requests a missing screenshot is demonstrating professional reading. Silence is not always efficiency.
Evaluate whether queries are specific, evidence-based and proportionate. “What does this mean?” creates work for the client. “In section 4, does ‘charge’ mean service fee or electrical charge? The neighbouring pricing table suggests fee; please confirm” is a useful query. It narrows the problem and offers a reasoned provisional interpretation.
On recurring work, query quality predicts how well the translator will handle imperfect sources. It also reveals whether the provider treats uncertainty as something to surface or something to conceal behind fluent wording.
10. Test research behaviour without asking for confidential methods
Translation requires research, especially when terminology, names and institutional language have authoritative forms. Ask candidates how they verify a term when several plausible target-language versions exist. A good answer often includes source hierarchy: official legislation, manufacturer documentation, standards, institutional websites, client terminology, reputable corpora and subject references can carry different authority depending on the task.
You are not looking for a theatrical list of websites. You are looking for disciplined evidence. Can the translator distinguish an official name from a literal rendering? Can they identify when a legacy translation is obsolete? Do they record a decision so it remains consistent later?
This also connects to the existing translation decision-log workflow: strong providers make difficult choices traceable instead of forcing future linguists to rediscover them.
11. Examine terminology governance
If terminology matters, ask who owns it and how it changes. A glossary is useful, but a list of source words and target words without definitions, status or context can create as many problems as it solves.
Look for a method that distinguishes mandatory terms from preferred variants, records definitions, tracks approval, propagates changes and prevents deprecated terms from reappearing. Ask how new terms discovered during translation are proposed and who approves them. If different target languages require different conceptual distinctions, the provider should be able to support language-specific notes rather than forcing every language into one simplistic table.
On large projects, terminology governance is one of the strongest indicators that a provider thinks beyond individual files.
12. Check technology fit without buying a tool name
Computer-assisted translation tools, translation memories, termbases, quality checkers, content connectors and localisation platforms can improve consistency and throughput. But software names are not a quality argument by themselves.
Ask what problem each tool solves in your workflow. Can the provider handle your file types without flattening structure? Can they preserve tags and placeholders? Can they import an existing translation memory with provenance controls? Can they return files in a form your content team can use? Can they support version changes without retranslating everything? Can they export your linguistic assets if the relationship ends?
Technology fit includes interoperability and ownership. Avoid creating a situation in which years of terminology and approved translations become inaccessible because they live only inside a supplier-controlled system with no practical export.
13. Ask direct questions about AI and machine translation
Do not assume a provider uses no machine translation, and do not assume that using it is automatically poor practice. The important questions are suitability, disclosure, security, review and accountability.
Ask whether source content may be sent to external AI or machine-translation services, under what contractual terms, whether data is retained or used for model improvement, which content classes are excluded, and what human review is applied. If your project prohibits certain tools or requires private deployments, state that before pricing.
Also ask how the provider decides when machine assistance is appropriate. High-volume repetitive support content may have a different risk profile from unpublished intellectual property, a legal instrument or a sensitive patient document. Professional practice is not a universal yes or no; it is controlled use tied to risk and specification.
14. Evaluate security in proportion to the content
Security questionnaires can become ceremonial unless they connect to actual data. Start by classifying what the provider will receive: public web copy, pre-release product information, personal data, legal files, student records, financial information, source code, health information or other sensitive material.
Then ask relevant questions about access controls, approved devices, storage location, encryption, subcontractors, retention, deletion, incident response and secure file transfer. The goal is not to demand the same controls for a public brochure and a confidential acquisition document. The goal is to ensure controls match the harm that disclosure could cause.
For the wider operational layer, see Protect Confidential and Sensitive Content Throughout Translation.
15. Compare prices on identical scope
A low quotation can simply exclude work that another provider included. One supplier may quote translation only; another may include bilingual revision, terminology management, project management, engineering and final QA. Comparing headline per-word rates without normalising scope is like comparing airfares without checking whether taxes, baggage and connections are included.
Create a comparison sheet that lists each component: translation, revision, editing, proofreading, terminology, engineering, desktop publishing, project management, rush charges, minimum fees, repetition discounts, translation-memory match discounts, reviewer integration and change requests. Ask how source updates after kickoff are billed.
Then consider the cost of internal management. A provider that costs slightly more but eliminates repeated rework, unclear files and terminology disputes may reduce total project cost. The cheapest sentence is not necessarily the cheapest finished publication.
16. Check capacity with evidence, not promises
Every provider can say that they can “scale.” Ask what scaling means operationally. How many qualified linguists are already available for the required pair and domain? How quickly can new linguists be onboarded? How is style kept consistent when the team grows? What review bottlenecks appear at peak volume?
Give candidates a realistic scenario. Suppose volume doubles for two weeks and the source continues to change. Ask how they would protect terminology, reviewer capacity and delivery sequence. Their answer should reveal trade-offs. A provider who claims that nothing changes under unlimited volume is not describing a real production system.
For very large programmes, ask whether they can prioritise content by business risk instead of simply pushing every file through the same queue.
17. Use references to investigate comparable work
References are most useful when the reference project resembles yours. Instead of asking only “Were you happy?”, ask how the provider handled difficult feedback, source changes, urgent requests, terminology disputes, continuity and mistakes. The quality of a relationship is often visible in recovery, not in the absence of all problems.
Respect confidentiality. A provider may be unable to name clients or disclose details. In that case, case studies with enough operational detail can still be useful. Look for evidence that the provider has solved the same class of problem, not necessarily worked for a famous logo.
18. Run a bounded pilot before a large commitment
For recurring or high-value work, the strongest selection evidence comes from a small real project. A pilot tests things that a sample translation cannot: kickoff, file handling, query flow, deadline behaviour, terminology learning, review response, invoicing, change control and final delivery.
Choose a pilot large enough to expose the workflow but small enough to change direction. Define acceptance criteria before it begins. Measure not only final text quality but also communication, traceability, adherence to instructions and how quickly the team absorbs feedback.
The existing translation pilot guide develops this stage in detail. Provider selection and pilot design belong together: selection narrows the candidates; the pilot tests the relationship under real conditions.
A practical provider-selection scorecard
A scorecard does not make judgment automatic. It prevents one impressive presentation or one low price from dominating the decision. Adjust the weights to your risk.
- Language and subject fit — 20%: evidence that assigned linguists understand the domain and target audience.
- Test translation — 20%: accuracy, completeness, terminology, naturalness, register and instruction-following.
- Revision and QA — 15%: clear ownership of bilingual review, factual checks and final quality control.
- Terminology and knowledge continuity — 10%: ability to build and maintain reusable decisions.
- Project management — 10%: communication, queries, scheduling, source-change handling and escalation.
- Technology and interoperability — 5%: safe handling of files, memories, tags, automation and exports.
- Security and confidentiality — 10%: controls proportionate to the content.
- Commercial fit — 10%: transparent pricing, sensible assumptions and sustainable capacity.
A medical translation buyer might increase domain and security weight. A marketing buyer might increase target-language style and creative adaptation. A continuous software team might increase workflow integration. The point is not the exact percentages; the point is explicit priorities.
Red flags worth investigating
- A quotation arrives before the provider asks what the translation is for.
- Every language and every subject is described as equally easy.
- The provider cannot explain who translates or who revises.
- A test is strong, but there is no assurance that the tested linguist will work on the project.
- “Proofreading” is promised without a defined activity.
- Machine translation or AI use is evasive rather than governed.
- The provider will not explain ownership or export of translation memories and termbases.
- Security answers are generic and disconnected from actual data flow.
- Prices rely on undefined future surcharges or unclear minimums.
- All feedback is accepted instantly without checking whether it conflicts with earlier approved terminology.
- There is no clear escalation route for source ambiguity or contradictory client instructions.
- The provider promises effectively unlimited capacity with no explanation of reviewer or linguist scaling.
Worked scenario: selecting for a technical product launch
Imagine a company launching a connected industrial device in five markets. The content includes a website, installer guide, in-app strings, warnings, packaging and a troubleshooting knowledge base. Three providers quote similar total prices.
Provider A offers the lowest unit rate and promises fast turnaround but cannot identify the proposed linguist team before award. Revision is described as “quality checking.” Provider B costs more, assigns named technical linguists, proposes a terminology workshop, asks for screenshots and engineering files, and separates safety content for stronger review. Provider C has excellent marketing references but little industrial experience and proposes one workflow for every asset.
A price-only comparison favours A. A domain-only comparison might still be ambiguous. A specification-based scorecard reveals more. The project contains interface constraints, safety language, technical terminology and frequent source updates. B’s questions demonstrate that it sees the actual problem. The added terminology and review work is not overhead unrelated to translation; it is part of producing a reliable multilingual product.
This does not mean B must win every project. If the content were a short public campaign with no technical risk, C’s creative strengths might matter more. Selection quality comes from matching evidence to the work.
Worked scenario: choosing an individual translator for a specialist report
A research organisation needs a 12,000-word report translated into one language. It will be published once, the subject is specialised, formatting is simple, and the internal team can manage the schedule. A large agency proposes a multilingual production environment that the project does not need. An individual translator with subject expertise provides relevant samples, asks precise questions about terminology and authorship, and has an independent reviser available.
Here, the smaller operating model may be stronger. The buyer gets direct access to the person making the language decisions and avoids unnecessary coordination layers. The lesson is not “freelancers are better.” The lesson is that procurement should not buy complexity merely because complexity looks professional.
Worked scenario: a multilingual public-service programme
A public organisation publishes forms and eligibility guidance in twelve languages. Content changes every month, and a mistranslation can affect whether someone understands access to a service. No single translator can provide the required coverage and continuity.
The buyer therefore evaluates governance as heavily as individual linguistic skill. It asks providers to demonstrate team assignment, terminology control, review, source-update handling, record keeping, reviewer feedback, security, accessibility and release processes. A test translation is only one part of the selection. The pilot includes a real source update halfway through so the organisation can observe change control.
This is what professional selection looks like at scale: you are selecting a living production system.
Questions to ask in a provider interview
- Who would actually translate and revise this work?
- How do you decide whether a linguist has sufficient domain competence?
- What happens when the source is ambiguous or internally inconsistent?
- How do you create, approve and update terminology?
- Which review activities are included in this price?
- How do you prevent reviewer preference from becoming uncontrolled terminology drift?
- What translation-memory or AI systems may receive our content?
- How do you handle confidential data and subcontractors?
- What assets can we export if the relationship ends?
- How do you manage source changes after translation has started?
- How do you scale volume while keeping the same voice and terminology?
- Show us an example of a difficult issue your process is designed to catch.
- What information do you need from us to do good work?
Frequently asked questions
Should I always ask for a free test translation?
No. A meaningful test consumes professional time. For serious specialist work, paying for a representative test can produce better participation and allows you to request a realistic task. What matters is fairness, clarity and whether the test predicts the real work.
How long should a translation test be?
Long enough to contain representative decisions, short enough to evaluate carefully. A few hundred well-chosen words can be more informative than several thousand generic words. Include the features that create risk in the real project.
Should I choose a native speaker?
Professional target-language competence matters, but birthplace is not a quality metric. Focus on demonstrated writing competence, subject understanding, translation skill, target-locale fit and the requirements of the specific content.
Is the cheapest provider usually risky?
Price alone tells you little. A low price can reflect efficient systems and an appropriate operating model, or it can reflect excluded review, inexperienced linguists or unsustainable workload. Normalise scope before drawing conclusions.
Is an ISO-certified provider automatically better?
No. Standards can provide useful evidence of process discipline, but they do not eliminate the need to evaluate language pair, subject expertise, team fit, security and actual output.
Should I evaluate translation agencies and freelancers differently?
Use the same core questions—competence, revision, terminology, communication, security, technology and accountability—but recognise that responsibility may be distributed differently. With an individual, ask how backup and independent revision work. With a company, ask who the assigned linguists are and how continuity is maintained.
How important are references?
Useful, especially for recurring work, but only when you ask about comparable problems. A reference from an unrelated content type proves less than a detailed account of similar domain, scale and review conditions.
What if two candidates produce equally good tests?
Use the rest of the system. Compare questions, turnaround realism, terminology process, reviewer integration, security, commercial transparency and pilot performance. Translation quality is necessary, but a long-running programme succeeds through repeatable delivery.
Can AI help evaluate providers?
It can help organise scorecards, compare stated processes and flag textual differences, but it should not be the sole judge of translation quality. Meaning, domain correctness, pragmatic fit and acceptable variation require informed human evaluation.
How often should a provider be re-evaluated?
Continuously through evidence rather than through constant formal tenders. Track serious errors, feedback cycles, deadlines, terminology stability, query quality and stakeholder outcomes. Reassess formally when scope, language coverage, risk, technology or performance changes materially.
The larger lesson: procurement is part of translation quality
A good translation does not begin when the translator opens the file. It begins when the organisation defines what it needs and selects people capable of meeting that need. Poor procurement can make excellent translation unlikely by choosing the wrong expertise, stripping out necessary review, imposing unrealistic timelines or treating every language task as interchangeable.
The most reliable selection process is evidence-based and proportionate. Define the job, assess language and subject fit, test representative work, examine the real workflow, compare equivalent scope, verify security, check references, and run a pilot where the stakes justify it. Price still matters. It simply belongs inside a larger decision about value and risk.
That is how professionals choose translation providers: not by asking who promises the most, but by asking who can show how the right meaning will survive the entire journey from source to reader.