Translating your website for SEO: what the free tools can and cannot do

Searching for a free website translator is really two questions at once, and they have opposite answers. Translating the words costs nothing and the tools are good. Getting the translated version to appear in search results is a different job, and no widget accomplishes it, because ranking requires pages at their own addresses and a widget creates none. The version a visitor sees after clicking a language dropdown often exists only in their browser. Search engines were never shown anything.

Do free website translators work for SEO?

By researching each language separately rather than running the pages through a translator. A new site should pick one and do it well; a mid-authority site adds the second with its own keyword list; a consolidated one declares hreflang, which only counts when both halves point back at each other.

Quick answer

• A browser widget translates for visitors, not for search engines. It produces no indexable pages.
• Search engines need separate URLs per language, which is the whole architectural decision.
• The spam policy is narrower than people quote. It targets scraped content transformed at scale.
• Do not translate your keywords. People search a different phrase, not the translation of yours.

What kind of problem this is

Three different jobs get called translation, and conflating them is what produces a Spanish version nobody ever finds.

Translating converts words. Localizing adapts the offer, the examples and the wording to how the market actually talks. Publishing for search creates addresses a crawler can reach and store.

A free tool does the first well. It does not attempt the third at all, and the third is the one that decides whether any of it appears in a search result. The tell is easy to check: click your language switcher and look at the address bar. If the URL did not change, there is no second page, and there is nothing to rank.

The three approaches, honestly

The browser widget

Fast, free, and useful for a visitor who has already arrived and cannot read the page. That is a real use and it is worth having.

What it does not do is create a Spanish page for someone who searches in Spanish and has not arrived yet. Those two problems look similar and share almost no solution.

Machine translation published as real pages

This one does create indexable pages, and it is where the risk lives, though not for the reason usually given. Google's spam policies do not name unreviewed machine translation of your own content. What they name, under scaled content abuse, is "scraping feeds, search results, or other content to generate many pages", including through automated transformations such as translating. The target is scraped material transformed at volume, not a business translating its own service pages.

So the risk here is a quality risk rather than a policy one, and it is no less real for that. Raw machine output reads as raw machine output to the person deciding whether to trust you with money, and that judgment happens before any algorithm gets involved.

For a small site the review pass is not big. Ten service pages is an afternoon for someone who speaks the language, and skipping that afternoon is what turns a translation project into a liability.

Separate pages with human review

The working version, and the one this site runs. Each language has its own URLs, each page declares its counterpart, and nothing is published that a human has not read.

The maintenance cost is real and it is the honest argument against translating everything: every future edit happens twice. Whether that cost is worth paying at all is the question behind Spanish SEO services, and the answer is more often no than the people selling it admit.

hreflang, and the way it usually breaks

Once pages exist in two languages, hreflang tells search engines which is which. The mechanics are simple and the failure is subtle.

The annotations have to point both ways. Google's documentation is explicit that without return links the annotations may not be interpreted correctly, and the failure mode is quiet: the markup is valid, nothing reports an error, and the relationship is simply not used.

This site treats that as a hard rule and it is worth describing, because it is unusual. A pair is only declared when five conditions hold at once: the counterpart file exists, it is published, it is in the other language, its own pointer comes back, and both halves say so. If any one fails, neither page declares anything, on the reasoning that half a signal is worse than none. The generator refuses to emit it and the site's own audit fails the build.

Most sites do the opposite. They declare the annotation as soon as the translated page exists and never verify the return link, which is how a project ends up with markup that looks complete and does nothing.

The mistake that costs the most

Translating the keywords instead of researching them.

The words people search in Spanish are frequently not the translation of the words they search in English, and the gap is widest exactly where the money is. A term that is standard in one market is unused in another, regional variants split the demand, and the literal translation of your best English phrase can be something nobody types.

The fix is to research the second language as its own market rather than as a mirror of the first, which is the same work described in the Spanish guide on keyword research.

The shape of the difference is easiest to see in how the questions are framed rather than in vocabulary. Spanish-language searches in this market lean heavily toward getting a business to show up at all, phrased as a procedure: how to appear, how to add the business, how to appear first. English-language searches in the same categories lean toward cost, choosing a provider and specific trades. Those are not the same article with different words. A site that translated its English pages line by line would answer the second set of questions, competently, in a language whose speakers were asking the first. What that gap means for a business serving both audiences is the subject of SEO for Latino-owned businesses.

When NOT to translate your site

When nobody searches your category in the second language in your area. Translation costs money forever and creates demand never.

When you cannot serve the customer in that language. A Spanish page that produces a call nobody in the office can answer is worse than no page, because it converts a stranger into a bad experience.

When the site is small and the English version is not working yet. Two underperforming languages is not progress, and the diagnosis usually applies to both.

The claim that keeps circulating, and what the policy says

Search this topic and you will read, in more than one of the first results, that unedited machine translation triggers a penalty, or that Google detects it with high accuracy and demotes it.

Google's spam policies do not say that. What they describe under scaled content abuse is "scraping feeds, search results, or other content to generate many pages", including automated transformations such as synonymising or translating. The target is scraped material transformed at volume. A business translating its own service pages is not what that describes.

This matters because the two readings lead to opposite decisions. Under the penalty reading, you avoid machine translation entirely and either pay for everything or skip the second language. Under what the policy actually says, machine translation is a starting point that needs a person who speaks the language to make it answer the question the reader asked, which is a different and much smaller budget.

There is a second figure worth treating carefully. Several of those results state that 75 % of hreflang implementations contain errors, attributed only to "studies". No study is named. What is verifiable, and matters more, is the mechanic: hreflang requires return tags, and a single error in a cluster makes the whole cluster get ignored. Half a signal is not half a benefit, it is none, which is why declaring a pair that does not point back both ways is worse than declaring nothing.

What a second language costs, honestly

Keyword research twice, because the second list is not a translation of the first: free keyword research.

Writing rather than converting. A translated page answers the question the original was written for, which is often not the question people ask in the other language.

And maintenance twice, which is the part nobody budgets: every update, every price change and every new service now has two homes. That is the real reason most bilingual sites end up with one half stale.

The market this matters most in is covered in SEO for Latino businesses in the USA and Spanish SEO services, and the local layer that usually decides more than any of it, in local SEO for small business. What the whole thing costs is in SEO cost for a small business.

Mistakes that repeat

• Using a widget and expecting search results. It renders for visitors and indexes nothing.
• Publishing machine output unread. Not a policy violation by itself, and it still reads like one.
• Translating keywords instead of researching them. The phrase people type is different.
• Declaring hreflang in one direction. Without the return link the annotation may be ignored.
• Translating everything. Every future edit now happens twice, forever.
• Offering a language you cannot answer the phone in. It converts interest into a bad experience.

Data and transparency

That Google's spam policies describe scaled content abuse as scraping feeds, search results or other content to generate many pages, including automated transformations such as translating, and do not name unreviewed machine translation on its own as a violation, was checked against the policy page on 12 August 2026. A previous version of this article said otherwise, following a wording that an earlier version of the guidance used. The 75 % figure for faulty hreflang implementations is quoted as an example of an unsourced statistic, not repeated as fact: the results carrying it attribute it only to "studies".

Google documents the requirement that hreflang annotations include return links, and states that without them the annotations may not be interpreted correctly. Google's spam policies do not name unreviewed machine translation of a site's own content; verified against the spam policies page in August 2026, what they name under scaled content abuse is scraping content to generate many pages, including through automated transformations such as translating. An earlier version of this article stated the policy covered unreviewed machine translation directly, which reflected older guidance and was corrected. The five-condition pairing rule and the build that fails on a broken pair are this site's own implementation, described here because it is verifiable in the pages themselves, not because it is a standard anyone publishes. The description of how Spanish-language and English-language searches are framed differently in the US market is a qualitative reading of this project's own harvest of 1,510 autocomplete suggestions collected in August 2026, and it is deliberately given without counts: the two language halves of that harvest were seeded from different starting phrases, so their totals are not comparable and any percentage drawn from them would be an artefact of the seeds rather than a measurement of demand. No figures appear here for how much traffic a translated site gains: that number depends entirely on whether demand exists in the second language, and any general figure is invented. Verified as of August 2026.

What this changes

The way this decision is usually framed is as a translation project with a budget and an end date.

It is closer to opening a second location. Everything published now exists twice, every future change happens twice, and the version nobody maintains slowly becomes the version that embarrasses you. The businesses that do this well tend to translate less than they intended: the pages that sell, in the language they can actually answer the phone in, and nothing else.

Frequently asked questions