Answering in Spanish: Where AI Translation Helps, and Where It Costs You

A large share of homeowners and patients in the US and Canada speak a language other than English at home — most often Spanish, and in parts of Canada, French. Machine translation is now good enough that a two-person office can answer a Spanish-language inquiry in Spanish within a minute. That is a real advantage, and it is also a real way to lose trust if you point the tool at the wrong things. Here is where AI translation earns its place in a service business, and where it should never be the last step.

Translate the top of the funnel first

The highest-value place to add another language is the part of the journey where the customer is deciding whether you are worth calling: the ad, the landing page, the form, the first reply, the appointment reminder. This is repetitive, low-risk, high-volume text. It is also the part most businesses never touch, because translating a whole website feels like a project and translating one landing page feels pointless.

Start smaller than a website. Take your three highest-traffic service pages and your booking form, and produce a Spanish version of each at a clean URL — not a browser auto-translate widget, which search engines treat as the same English page and which produces different text every time someone loads it. A real second page can rank, can be linked in an ad, and can be reviewed by a human once instead of re-rolled on every visit.

Then do the same for your automated messages: confirmation texts, reminder texts, the "we got your request" email, the after-hours reply. These get sent thousands of times and change almost never, which makes them the cheapest thing in your whole operation to translate well. Write them once, have a fluent speaker check them once, and they keep paying off.

Regional Spanish is not one language

Generic machine Spanish tends to come out in a formal, neutral register that reads as translated. That is not fatal on a reminder text, but it is noticeable in marketing copy, and it can be flatly wrong in trade vocabulary. Words for common household and construction items differ between Mexican, Caribbean, and South American Spanish, and your service area probably skews toward one. A roofer in San Antonio and a roofer in Miami are writing to different readers.

The practical fix is to tell the model who it is writing for — region, register, and reading level — and then have someone from that community read it. If you have a bilingual technician, receptionist, or injector on staff, twenty minutes of their time on your core pages is worth more than any translation vendor. If you don't, budget for a one-time human review of the pages and templates you use most. It is a fixed cost, not a recurring one.

The same applies to the formal-versus-informal choice. Spanish forces you to pick between usted and tú in nearly every sentence, and the right answer depends on your service and your customer. A med spa talking to returning patients often wants warmth; a commercial HVAC contractor writing to a property manager usually does not. Make the call once, write it into your instructions, and stay consistent — drifting between the two inside one email is the tell that nobody read it.

Where translation stops and a human takes over

Draw the line at anything with money, medical content, or legal consequence attached. Estimates, contracts, warranty terms, cancellation policies, consent forms, pre- and post-treatment instructions, and anything describing risk should not go out in a language you cannot read, translated by a system you cannot question. A mistranslated line in a warranty is a dispute you will lose. A mistranslated line in aftercare instructions is worse than that.

Live conversation deserves the same caution, for a different reason. AI can hold a competent Spanish exchange about hours, pricing ranges, and availability. What it cannot do is notice that the customer has switched to English mid-sentence because they are frustrated, or that the person on the other end is describing an emergency in words the model is smoothing over. Build an explicit rule: if the conversation moves from information to commitment — a price, a date, a promise — it hands off to a person.

And be honest about capacity. If your ad runs in Spanish, someone will call in Spanish. Answering that call in English after promising otherwise costs you more goodwill than never advertising in Spanish at all. Do not open a door you cannot staff. If you have one bilingual person and they work Tuesday through Saturday, either your ads run on those days or your intake flow routes those calls to them, plainly and quickly.

What to automate, what to decide

Automate the repeatable text: transactional messages, reminders, form labels, FAQ answers, review replies, and first drafts of marketing copy. Automate the routing too — detecting the customer's language from their inquiry and tagging the record so every subsequent message stays in that language, and so the right person gets the callback. Automate consistency checks, like flagging when a page has drifted out of sync with its English version.

Keep the judgement calls human: which languages you commit to at all, who reviews the core pages, what register you use, where the hand-off line sits, and every document a customer signs. Translation is one of the clearest examples of a wider rule — the machine is excellent at volume and terrible at knowing when the stakes just changed. Your job is to decide in advance where those moments are, and make sure a person is standing there when they happen.