Translate content to English to capture ChatGPT’s hidden queries

What happens when a German user asks ChatGPT about IT onboarding software, and the system cites both the German and the English version of the same page in a single response? That’s exactly what we observed with our client deeploi. And it pointed us toward an underrated AI search tactic available to non-English companies right now: translating your existing content into English to capture citation spots you didn’t know you were missing.

Key takeaways

• 43% of ChatGPT’s background searches run in English, even when the user prompts in another language. In 78% of non-English sessions, at least one fan-out query switches to English. Companies without English content are invisible to nearly half of ChatGPT’s retrieval process.
• Translating existing content into English can double your AI citation spots without creating anything new. We observed this firsthand with deeploi, where both the German original and the English translation were cited separately in the same AI response.
• This isn’t about reaching new markets. It’s about dominating your existing market in AI search by matching the English-language sub-queries that ChatGPT generates behind the scenes for your local audience.
• The tactic is fast, low-risk, and measurable. Start with your top 5 most-cited pages, deploy translations with proper technical setup, and monitor citation impact within 30 days.

Looking for a shortcut to drive more organic growth from your content, SEO & AI Search efforts? Request a free growth audit from Radyant to get an honest assessment of your organic growth potential.

The discovery: same content, two citation spots

We started supporting deeploi in organic growth efforts in October 2025, with a specific focus on AI search visibility. As part of the strategy, we produced content in German first, then published English translations on the same site.

When we checked Google AI Overviews in Peec AI for “best IT onboarding software,” something unexpected showed up: both the German original and the English translation were cited as separate sources in the same response. Same content. Two languages. Two citation slots.

deeploi is now among the top recommended IT onboarding tools in AI search, powered by a multilingual owned content citation base. The English version didn’t just serve international users. It captured citation spots that the German version alone could never reach, because the AI system was running background queries in English.

This observation sent us down a research rabbit hole. And the data we found explains exactly why this works.

How ChatGPT’s query fan-out creates the opportunity

When you type a prompt into ChatGPT, the system doesn’t just fire off a single search query. It decomposes your prompt into multiple sub-queries, a process called query fan-out. Google popularized the term when introducing AI Mode, but ChatGPT has been doing it for months.

The purpose is straightforward: by breaking one broad question into 8-12 specific sub-queries, the system retrieves more precise, comprehensive information to build its response. Think of it as a research assistant running parallel searches rather than one big Google query.

Here’s where it gets interesting for non-English companies.

The English-language bias in fan-out queries

Peec AI analyzed over 10 million prompts and 20 million fan-out queries and found a consistent pattern: 43% of ChatGPT’s background searches ran in English, even when the original prompt was in another language.

The behavior follows a predictable sequence. ChatGPT starts its fan-out queries in the user’s language, then switches to English as it builds the response. In 78% of non-English sessions, at least one fan-out query runs in English.

This isn’t a minor edge case. The English fan-out rate varies by language, but no language falls below 60%:

• Turkish: 94% of sessions include English fan-out queries
• German: Approximately 80%
• Spanish: 66% (the lowest observed)

Even in Spanish, the language with the lowest English fan-out rate, two-thirds of ChatGPT sessions include English-language research. For German-speaking markets (where many of our clients operate), the number is closer to 80%.

Why does ChatGPT switch to English?

Peec AI’s Tomek Rudzki identifies two root causes:

1. Authority signals favor global content. English-language pages typically have more backlinks and citations, which are signals AI models use to determine source authority. Sites with over 32K referring domains are 3.5x more likely to be cited by ChatGPT than those with fewer than 200.
2. Risk minimization. Roughly 50% of internet content is in English. Querying in English increases the probability of finding quality sources, so the system defaults to it when it needs to supplement its initial findings.

Whether this is an intentional design choice or an emergent behavior remains unclear. OpenAI’s documentation doesn’t describe how language is chosen for rewritten queries. But the practical impact is the same regardless of intent: if you only publish in your local language, you’re invisible to nearly half of ChatGPT’s retrieval process.

The real-world consequences are already visible

The Peec AI research includes striking examples of what this bias produces:

• When prompted in Polish from a Polish IP about the best auction portals, ChatGPT either omitted or buried Allegro.pl (Poland’s dominant marketplace) in favor of eBay and other global platforms.
• When prompted in German about German software companies, the response listed no German companies.
• When prompted in Spanish, ChatGPT’s fan-out added the word “globales” (global) to the query, a qualifier the user never used, broadening the search toward English-language results.

These aren’t edge cases. They’re the default behavior for the majority of non-English ChatGPT sessions.

Why this is an opportunity, not just a problem

Most commentary frames the English fan-out pattern as a disadvantage for non-English businesses. The dominant narrative is defensive: “you’re losing citations to English content.”

That framing misses the bigger picture.

If ChatGPT runs fan-out queries in both your language and English, and your content exists in both languages, you can be cited from both versions in the same response. That’s not a problem. That’s a citation multiplier.

The deeploi case proves this directly. We didn’t create new content for the English version. We translated what already existed. Same arguments, same data, same structure. The result was two separate citation slots from one piece of content.

This aligns with what we’ve seen across our client work: owned content authority beats external mentions as Plan A. With Planeco Building, we grew citation share from 55% to over 130% and 5x’d organic leads in 10 months, all from owned content alone. No outreach, no Reddit commenting, no backlink chasing. The translation strategy is a natural extension of this philosophy: expand the surface area of your owned content without relying on any external signals.

A Weglot study analyzing 1.3 million AI citations supports this at scale. They found that websites with translations gain 327% more visibility in Google AI Overviews and ChatGPT. Translated sites didn’t just perform better in the secondary language. They were more visible overall, even in their original language, with 24% more citations per query across both languages.

Most companies think about multilingual content as “reaching new markets.” The real arbitrage right now is using multilingual content to dominate your existing market in AI search.

The multilingual citation multiplier: 5 steps

Here’s the framework we’ve developed based on the deeploi work, validated against the Peec AI and Weglot data. It’s designed for B2B companies operating in non-English markets that want to increase AI search citations without producing net-new content.

Step 1: Audit your current AI citation landscape

Before translating anything, you need to know which pages are already getting cited in AI search. Use tools like Peec AI or Profound to identify:

• Which of your pages appear in ChatGPT and Google AI Overview responses
• For which queries you’re currently being cited
• Which competitors are cited alongside you (and in which languages)

This audit gives you a baseline. If you’re already getting cited for 15 queries in German, you now have a clear list of pages where English translations could add a second citation slot.

Step 2: Identify fan-out language patterns for your core queries

Not all queries trigger the same fan-out behavior. Commercial intent prompts trigger web search in ChatGPT 53.5% of the time, compared to just 18.7% for purely informational queries. Local intent prompts are even higher at 59%.

This means your commercial and locally-relevant content has the highest probability of benefiting from translation. Extract fan-out queries for your core topics using browser developer tools or platforms like Peec AI, and check how frequently English sub-queries appear for your vertical.

Fan-out queries are also getting more specific over time. Average word count per fan-out query roughly doubled from ~6 words in October 2025 to ~12 words by January 2026. Longer, more specific sub-queries mean the precision of language matching matters more, not less.

Step 3: Prioritize content for translation

Don’t translate everything at once. Prioritize based on three criteria:

1. Already cited in AI search. Pages that are already getting cited in your local language are the highest-priority targets. An English translation gives them a second citation slot immediately.
2. Commercial intent. Product comparisons, “best X” guides, feature pages, and use case content are more likely to trigger fan-out web searches. Translate these before informational blog posts.
3. Competitive differentiation. Content where you have genuine expertise or unique data (not generic overviews) is more likely to be cited regardless of language. Translation amplifies your strongest assets.

For deeploi, we started with the content that was already performing in German AI citations: their IT onboarding and IT asset management guides. These were the pages where translation had the most immediate citation multiplication potential.

Step 4: Deploy translations with proper technical setup

The translation itself is the easy part. AI tools (DeepL, Claude) produce solid first drafts that a native speaker can refine in minutes. The more important decisions are structural:

• URL structure: Use subfolder structure (/en/page-name) rather than separate domains. This keeps domain authority consolidated.
• Hreflang tags: Implement properly so search engines understand the relationship between language versions. This matters for Google organic, and the structured signal likely helps AI systems understand your content architecture.
• Content parity: The English version should mirror the German version closely. You don’t need to create original English content. The deeploi case shows that translations of existing content get cited separately.
• AI-readable structure: Both versions should use clear headings, direct answers in the first 30% of text (where 44.2% of all LLM citations are extracted from), and high entity density.

This is where content engineering shines. If you have a large content library, you can build a workflow that mirrors the gold standard page methodology we use for programmatic content: create one perfect translation manually, then use it as the template for scaled translation production with AI assistance and human quality checks.

Step 5: Monitor citation impact and iterate

After deployment, track three things:

1. Citation share by language. Use Peec AI or similar tools to monitor whether your English pages start appearing in AI responses alongside your local-language pages. Look specifically for instances where both versions are cited in the same response.
2. Citation share vs. competitors. Are you capturing citation slots that previously went to English-language competitors? This is where the competitive advantage materializes.
3. Pipeline attribution. This is the hard part, and it requires Radyant’s 3-layer attribution model (more on this below).

Give it 30 days for initial signal. AI search indexes and cites new content faster than Google organic, so you won’t wait months for results. If your top 5 translated pages start appearing in English fan-out citations within a month, scale the approach to the next batch.

Choosing the right translation approach

Not every company needs the same level of investment. Here’s how the options compare:

Approach

Effort

Citation impact

Best for

Full translation of top-cited pages

Medium

High

B2B SaaS with existing blog/resource content

Selective translation of key landing pages

Low

Medium

Resource-constrained teams testing the approach

Creating original English content (net-new)

High

High

Companies entering English-speaking markets

AI-assisted translation at scale

Medium-Low

Medium-High

Companies with large content libraries

For most non-English B2B companies, option 1 (full translation of top-cited pages) offers the best ratio of effort to impact. You already know which content works. You just need it in English.

Option 4 becomes relevant when you’re operating at scale. With Planeco Building’s programmatic content, we launched 247 pages in 7 days using the gold standard page methodology. The same workflow can be adapted for translation: create one manually-perfected translation as the reference, then scale production with AI assistance and human QC across the remaining pages.

How to measure whether this drives pipeline

Here’s the attribution challenge: if a German user asks ChatGPT in German, gets a recommendation influenced by your English content, then visits your German website directly, that shows up as “Direct” traffic in your analytics. The AI citation is completely invisible to click-based tracking.

This is why we use a 3-layer attribution model with every client:

• Layer 1: Click-based attribution. Your CRM and analytics data. Useful but increasingly unreliable for discovery channels. Keep tracking it, but stop treating it as the only truth.
• Layer 2: Self-reported attribution. A “How did you hear about us?” field on forms. Best practice: mandatory free-text (not a dropdown). LLMs now make analyzing free-text responses trivial at scale.
• Layer 3: Verbal attribution from sales. What prospects actually say in discovery calls. Most of this intel dies in the call unless you create a custom CRM field that sales fills in after every conversation.

For multilingual AI search specifically, add citation tracking as a fourth signal. Monitor which language versions of your content appear in AI responses, and correlate spikes in citation appearances with changes in lead volume or self-reported attribution mentions.

With Heyflow, we found that AI-attributed trials convert at 14.3% compared to an 11% channel average. That higher conversion rate makes the investment in translation easily defensible. Even a modest increase in AI citations from English translations can move the pipeline needle if the conversion quality holds.

If you’re operating in a non-English market and want to understand your current AI search citation landscape, Radyant’s growth strategy audit includes multilingual AI visibility analysis.

What about Google AI Overviews?

The English fan-out behavior is most pronounced in ChatGPT, but it’s not limited to it. Google AI Overviews show a different pattern that’s worth understanding.

Weglot’s research on AI search and language found that Google AI Overviews are actually the most language-sensitive. In their studies using localized queries for the Mexican market, 96% of AI Overview citations came from Spanish sources. When a Spanish-language option existed, English sources were consistently pushed out of the top five positions.

This means the translation strategy has a dual benefit:

• In ChatGPT: Your English translation captures the English fan-out queries that your local-language version can’t reach.
• In Google AI Overviews: Your local-language content has a strong advantage, and having the English version ensures you’re covered if Google’s behavior shifts toward ChatGPT’s pattern over time.

The net result: you’re building resilience across both major AI search surfaces. Your local content dominates Google AI Overviews. Your English content captures ChatGPT’s English fan-outs. Both versions can be cited in the same response, multiplying your total citation share.

Is this a hack that will stop working?

No. And this is an important distinction.

Publishing content in multiple languages is a longstanding good practice for international SEO. The “new” part is understanding that it also doubles your AI citation surface area. This isn’t exploiting a loophole. It’s making your content accessible to more retrieval pathways.

Even if ChatGPT eventually reduces its English fan-out bias (which is possible but uncertain), you’ll still have English content that serves international users, strengthens your domain authority, and provides an additional citation surface. The downside risk is essentially zero.

Compare this to tactics like Reddit spam or manufactured backlinks, which can be penalized overnight. Translation creates genuine value for users while also increasing your AI search footprint. That’s the definition of a sustainable strategy. As we see it: AEO done right is just good SEO with better attribution. Translation for AI search is no different.

The quickest way to test this

If you want to validate this for your own company before committing to a full rollout, here’s the minimum viable test:

1. Identify your top 5 pages that are already cited in AI search (use Peec AI, Ahrefs Brand Radar, or manual ChatGPT testing with relevant queries).
2. Translate those 5 pages into English using AI (DeepL or Claude) with a native-speaker review pass. This should take 1-2 days max.
3. Deploy with proper technical setup: subfolder structure, hreflang tags, matching URL structure.
4. Monitor for 30 days: Check whether the English versions start appearing in AI responses alongside the originals.
5. Measure citation share change: Compare your total citation count before and after across both languages.

If you see your English pages getting cited within 30 days, you have your proof of concept. Scale from there.

FAQ

Does translated content actually get cited separately from the original?

Yes. We observed this directly with deeploi, where both the German original and the English translation appeared as separate citation sources in the same AI response. The AI system treats them as distinct pages because they are distinct URLs with distinct content (even though the substance is the same). This effectively doubles your citation slots for that query.

Which languages benefit most from adding English translations?

Based on Peec AI’s data, Turkish content benefits most (94% of sessions include English fan-out), followed by German (~80%) and French (~75%). Spanish benefits least at 66%, but that still means two-thirds of sessions. No non-English language falls below 60%. If your primary market is DACH, the opportunity is significant.

Will AI translation tools produce good enough quality for this?

For the purpose of AI citation, yes, with a caveat. AI tools like DeepL and Claude produce translations that are structurally sound and factually accurate. A native-speaker review pass catches the remaining awkward phrasing or cultural mismatches. The content doesn’t need to be literary. It needs to be clear, accurate, and well-structured. Spend your perfectionism budget on the original content, then translate efficiently.

Do I need to translate my entire site or just key pages?

Start with key pages. Your highest-cited content and highest-commercial-intent pages should be translated first. There’s no evidence that translating low-value pages (privacy policy, team bios) contributes meaningfully to AI citation share. Focus your effort where the citation multiplication potential is highest.

Does this only work for ChatGPT, or also for Google AI Overviews and Perplexity?

The English fan-out behavior is most pronounced in ChatGPT. Google AI Overviews are actually more language-sensitive, favoring local-language content when available. Perplexity’s behavior falls somewhere in between. Having both language versions covers you across all three surfaces. Your local content wins in Google AI Overviews; your English content captures ChatGPT’s English fan-outs; and both can be cited by Perplexity.

How long before I see results from translated content?

AI search systems index and cite new content faster than Google organic search. In our experience, new pages can start appearing in AI citations within days to weeks, not months. Allow 30 days for a meaningful signal, then evaluate whether to scale. If your translated pages aren’t appearing after 30 days, check your technical setup (hreflang, indexability) before concluding the approach doesn’t work for your vertical.