# Translate WordPress with ChatGPT and Claude: Time and Cost Comparison > URL: https://4eck-media.de/en/blog/translate-wordpress-with-chatgpt-and-claude-time-and-cost-comparison/ > Language: en > Description: Translating a WordPress website into another language sounds simple at first: copy the text, paste it into ChatGPT or Claude, and put the result back into WordPress. That works fo… --- Translating a WordPress website into another language sounds simple at first: copy the text, paste it into ChatGPT or Claude, and put the result back into WordPress. That works for three paragraphs. But on a real website with many pages, images, buttons, SEO fields, and internal links, it quickly becomes a tedious task. The good news: You can **translate WordPress with ChatGPT and Claude** without handling every paragraph individually. What matters is not only the language model, but also a clean workflow. In this article, we explain the method in deliberately simple terms, compare ChatGPT, Claude, and WPML, and show the realistic costs. ![WordPress translation from English with XLIFF, ChatGPT, and Claude – time and cost comparison](https://4eck-media.de/wp-content/uploads/2026/07/wordpress-chatgpt-claude-time-cost-cover-16x9-v4.avif) ## XLIFF Is the Key to Our Workflow The most important component of this method is **XLIFF**. Without XLIFF, translating a large WordPress website with ChatGPT or Claude would be little more than laborious copying and pasting. XLIFF turns it into a controlled process that preserves the WordPress structure, technical identifiers, and the mapping of each text. XLIFF stands for *XML Localization Interchange File Format*. The complicated name describes something simple: an exchange file for translations. It stores small units of text. Each source unit has a designated place for its translation. The file also contains IDs and markers so that WPML later knows exactly where every translated sentence belongs on the page. Think of XLIFF as a secure transport container. WordPress and WPML place the content to be translated inside it. ChatGPT, Claude, or their coding agents edit only the designated text fields. The same container is then imported again, and WPML distributes the translations back to headings, paragraphs, buttons, and other fields. ### Why a Standard XLIFF File Is Not Enough An exported XLIFF file is structured, but a language model still needs precise instructions about which parts it may change. If the file is passed to a chat without rules, the model may rewrite tags, change IDs, translate placeholders, or omit sections in long files. The language may be correct, but the file may no longer import reliably. ### How We Modernize the XLIFF Process for AI We therefore extend the classic WPML export with a clear workflow for coding agents. Claude Code or Codex receive not only the files, but also fixed editing rules, a glossary, output folders, and automated checks. The agent translates only the target segments and treats tags, URLs, IDs, variables, and placeholders as protected technical components. - The XLIFF structure and order remain unchanged. - Only translatable target segments are edited. - A glossary and tone-of-voice rules ensure consistent language. - Checks detect empty segments, changed placeholders, and damaged XML. - Original and completed files are stored separately so that every step remains traceable. > **XLIFF is therefore our key:** The format combines WPML’s organizational structure with the language quality and flexibility of ChatGPT and Claude. ## Why Simple Copying Is Not Enough A website consists of more than visible body text. In WordPress, content is stored in Gutenberg blocks, page builders, custom fields, SEO plugins, menus, and sometimes theme settings. There are also links, image descriptions, tables, accordions, and forms. When entire pages are copied into a normal chat without checks, typical errors occur: HTML tags are changed, links disappear, placeholders are translated, or the answer is cut off because it is too long. The model can produce good language and still damage the technical structure. That is why we separate two tasks: WordPress provides the content in a controlled format, and the AI translates only the parts that should actually be translated. The result is then imported and checked. This is the core of the entire workflow. > **In short:** The AI does not need to understand WordPress. The workflow must clearly show the AI what is text and what must remain unchanged. ## The Role of WPML WPML remains a useful tool because it connects the language versions of a page. It knows which German page belongs to the English, French, or Spanish version. WPML also recognizes many translatable fields and can export content as an XLIFF file. Put simply, XLIFF is a transport format for translations. Source text and target text appear side by side. Technical identifiers, IDs, and placeholders are preserved. A coding agent such as Claude Code or Codex can process this file systematically. The difference from the classic WPML workflow is that the translation does not have to run through WPML’s automatic translation feature. WPML organizes the language versions, while ChatGPT or Claude handles the language work. This preserves the WordPress structure while providing more control over tone, terminology, and costs. ## Claude Cowork or ChatGPT/Codex: Which Works Better? For our workflow, we compare not only two language models, but two practical working environments: **Claude Cowork** and **ChatGPT with Codex**. Both can open files, apply instructions across multiple XLIFF files, save results, and perform technical checks. For a large WordPress translation, these capabilities matter more than one particularly elegant sentence in a chat. **Our honest conclusion:** After a few small adjustments and setting up our API, both systems can work according to the same principle. We provide the XLIFF files, fixed rules, a glossary, and a clearly defined output folder. The agent translates only the designated target segments and then checks that tags, IDs, URLs, and placeholders remain unchanged. We explain in detail how we established this connection and adapted the classic WPML process for AI agents in our [complete workflow for translating WordPress with AI](https://4eck-media.de/en/blog/translating-wordpress-with-ai-fast-cheap-and-without-the-wpml-frustration-our-complete-workflow/). The guide can be used in Claude Cowork, Claude Code, or Codex and adapted to your own website. - **Claude Cowork:** works very comfortably with long documents, many related files, and detailed rules. - **ChatGPT with Codex:** can automate the same process, edit files, and check the results technically. - **Both systems:** require a clean XLIFF export, clear protection rules, and a final check in WordPress. The key question is therefore not, ‘Can only Claude do this?’ Both can. In practice, they differ mainly in **speed, available usage limits, and price**. Depending on the volume of files, one agent may work faster or reach its message limit later. This can change at any time with new model versions and pricing plans. Price comparisons must also distinguish between two options. With a monthly subscription, you pay a fixed amount but work within the respective usage limits. With the API, you are charged for the text actually processed. The API can be extremely inexpensive for occasional translations. Those who use the agent daily for many additional tasks may find a subscription more convenient. We compare specific cost examples later in this article. > **Our opinion:** Do not choose based on the provider’s name, but on the current balance of speed, limits, and price. The technical workflow is essentially the same for Claude and ChatGPT/Codex. ## The Workflow in Six Simple Steps ### 1. Define Target Languages and Rules Before the first export, we define the language, form of address, writing style, and fixed technical terms. A small glossary prevents the same term from being translated differently across five pages. Brand names, product names, and legally relevant wording also belong on this list. ### 2. Export Content from WordPress The required pages are created as translation jobs in WPML and exported as XLIFF. For large websites, we work in manageable batches. This makes errors easier to find, and an interrupted run does not affect the entire website. ### 3. Give the Agent Clear Instructions The task should be specific: translate only the target segments, do not change tags, IDs, URLs, shortcodes, or variables, and return valid XLIFF files. The agent also receives the glossary and examples of the desired style. ``` Translate the XLIFF files from German into English. Preserve tags, IDs, URLs, placeholders, and the file structure. Use the glossary and check every target segment for completeness. Save the completed files in a separate output folder. ``` ### 4. Translate and Check Automatically The agent processes the files and then checks whether every source segment has a target segment. Useful additional checks include empty translations, changed links, missing placeholders, and damaged XML structures. This technical validation is at least as important as the translation itself. ### 5. Import into WordPress The validated files are imported back into WPML. The texts are placed in the correct locations and remain connected to the source page. There is no need to rebuild the page or assign paragraphs manually. ### 6. Check the Front End Finally, we inspect the completed page as a visitor would: headings, buttons, navigation, forms, line breaks, images, internal links, and the mobile layout. AI saves a great deal of work, but it does not replace the final quality check in the browser. ### The Most Important Points from the Workflow - WPML organizes the language versions. - XLIFF transports content securely. - ChatGPT or Claude translates according to fixed rules. - Automated and visual checks ensure quality. ## How Much Does Translation with ChatGPT or Claude Cost? *All prices and model names in this section are accurate as of July 20, 2026. Providers may change prices, limits, and names at any time. Always check the current price lists before starting a larger project.* There are two basic options: a monthly subscription or usage-based API billing. ChatGPT Plus costs around USD 20 per month. Claude Pro also costs USD 20 per month, or about USD 17 per month with annual billing. Depending on the exchange rate, this is roughly in the region of EUR 20. A flat-rate subscription is convenient, but not unlimited. Message and usage limits can be reached with very many or very long files. For small and medium-sized websites, a monthly subscription may still be the simplest calculation: you pay a fixed amount and complete the translation, editing, and checking within that month. The API charges for processed tokens. Tokens are small units of text. The price depends on the model, the length of the source text, and the length of the translation. For a rough example, we assume an article of around 2,000 words: approximately 4,500 input tokens including instructions and around 3,000 output tokens for the translation. | Model | Estimated API Cost per Article | | --- | --- | | GPT-5.6 Luna | approx. USD 0.02 | | GPT-5.6 Terra | approx. USD 0.06 | | GPT-5.6 Sol | approx. USD 0.11 | | Claude Haiku 4.5 | approx. USD 0.02 | | Claude Sonnet 5, promotional price | approx. USD 0.04 | | Claude Sonnet 4.6 / standard price | approx. USD 0.06 | | Claude Opus 4.8 | approx. USD 0.10 | These are estimates for model usage only. Repetitions, extensive checks, and very long system instructions increase consumption. Nevertheless, the figures show that translating a normal blog article through the API usually costs only a few cents. ## How Expensive Is WPML by Comparison? WPML’s automatic translation is billed using credits. According to the official price list, Microsoft requires one credit per word, DeepL and Google two credits each, and PTC four credits. The first 2,000 credits each month are free. Costs then increase in tiers. | Volume | Microsoft via WPML | DeepL/Google via WPML | PTC via WPML | | --- | --- | --- | --- | | 2,000 words | EUR 0* | approx. EUR 1.50 | approx. EUR 4.50 | | 10,000 words | approx. EUR 6 | approx. EUR 13 | approx. EUR 25 | **If the 2,000 free monthly credits are still available.* Compared with direct API usage, WPML is significantly more expensive for large volumes of text. A 10,000-word package costs roughly USD 0.09 to 0.28 with an inexpensive AI model used directly through the API, while WPML costs around EUR 6 to 25 depending on the engine. Purely mathematically, WPML can therefore be approximately 20 to more than 200 times more expensive. That does not mean WPML is bad. The price includes convenience, WordPress integration, management, and a ready-made translation process. Those who translate only a few pages occasionally may deliberately pay for this convenience. However, anyone with a great deal of content, multiple languages, or recurring translations can save considerably with a dedicated ChatGPT or Claude workflow. ## Which Model Should You Choose? An inexpensive model is usually sufficient for standard texts. A good glossary, clear rules, and a review are more important than the most expensive model class. A stronger model is worthwhile for complex technical language, sophisticated brand tone, or when sentence structure and style must sound particularly natural. - **Many simple product texts:** inexpensive model plus automated checks. - **Blog articles and corporate pages:** a mid-range model offers a good compromise. - **Legal, medical, or highly sensitive content:** a powerful model plus human expert review. ## Why WordPress Is Particularly Well Suited to This Approach WordPress’s greatest strength is its adaptability. Content can be exported, structured through plugins, and processed with small custom tools. You are not restricted to the single method specified by one provider. This is exactly why WordPress can be adapted to a specific task: store a glossary, exclude certain fields, add checks, or automate output for several languages. The technical setup requires some preparation. Afterwards, many manual individual steps become a repeatable process. ## Do Not Forget SEO, Images, and Internal Links A multilingual website is complete only when more than the main body text has been translated. Page titles, meta descriptions, URLs, image alt text, and internal links are especially important for search engines. This information often sits outside the visible Gutenberg text. Before exporting, you should therefore determine which fields WPML includes and which must be processed separately. The main keyphrase should not necessarily be translated word for word from German. People search differently in every language. A good translation therefore combines language quality with a brief keyword check in the target market. For example, the German term ‘Website übersetzen’ may have different common equivalents in English than a direct translation would suggest. Internal links also require attention. If the German page links to a German-language article, the English version should link to its English translation whenever possible. Automated checks can detect incorrect language paths, but the contextually appropriate target page sometimes requires an editorial decision. ## Common Mistakes and How to Avoid Them - **Packages that are too large:** divide content into smaller, traceable groups. - **No glossary:** define key terms before translating. - **Focusing only on language:** also check XML, links, placeholders, and completeness. - **Importing blindly:** process a test page first and validate the workflow. - **No backup:** keep the export files and a WordPress backup before importing. A test run with a typical page is particularly useful: headings, images, buttons, lists, and a form. Once this test works correctly, the same process can be applied to larger volumes. This keeps the risk low and the work manageable. ## Frequently Asked Questions Can ChatGPT Translate an Entire WordPress Website? Yes. For a complete website, however, ChatGPT or Codex should not rely on simple copying and pasting, but on structured XLIFF files, clear rules, and subsequent checks. Does the Workflow Work Only with Claude? No. Claude Cowork, Claude Code, and ChatGPT with Codex can all implement the workflow. Quality depends primarily on the model, the instructions, the glossary, and the review process. Do I Need WPML? Not necessarily, but WPML simplifies the mapping of language versions and the XLIFF export. With other multilingual plugins, the export and import process must be adapted accordingly. Is an AI Translation Immediately Ready to Publish? Technically, almost; editorially, not always. Important pages should be checked for language, layout, links, forms, and technical terms. ## Our Complete Guide as a Template This article provides a simple overview. If you want to reproduce the entire process with rules, checks, and practical details, use our [complete guide to translating WordPress with AI](https://4eck-media.de/en/blog/translating-wordpress-with-ai-fast-cheap-and-without-the-wpml-frustration-our-complete-workflow/). You can copy the guide and paste it into Claude Code or Codex. The agent will then assist you step by step with exporting, translating, checking, and importing. This is where WordPress demonstrates its greatest strength: the platform can be adapted simply and specifically to your own workflow. ## Have a Multilingual Website Built Professionally Would you prefer not to set up the workflow yourself, or do you need a fully multilingual website? We develop and optimize multilingual WordPress websites—from the technical structure and WPML configuration to translation, international SEO, hreflang, internal linking, and quality assurance. Our service page explains how we [**create multilingual websites**](https://4eck-media.de/en/competencies/build-a-multilingual-website-more-reach-more-customers/) and which solution fits your target markets, languages, and existing WordPress system. **Sources and pricing date:** OpenAI Models and ChatGPT Pricing, Anthropic Pricing, and WPML Automatic Translation Pricing; accessed July 20, 2026.