From search terms to conversations: Multilingual GEO for LLMs

As a specialist in multilingual organic optimisation, I’m particularly interested in how this evolution plays out across languages and cultures. This adds a new layer of complexity: not only do we need to understand conversational prompts, but we also need to ensure they are accurate, culturally nuanced, and locally relevant. This field is still very much in its infancy, so what follows are my observations based on experience; anyone claiming to have definitive answers is likely overconfident.

What’s the difference between prompt research and keyword research?

“Answer engines” like ChatGPT and Gemini are changing how users seek, process, and act on information. Traditional SEO relies heavily on keyword volumes, user intent, and ranking strategies. With LLMs, however, content is surfaced based on prompts, longer, more nuanced, conversational inputs from users. In short, the main difference right now between keyword research and prompt optimisation is the long-tail nature of queries. We’re talking about conversational, complex questions. 

Prompt research & content strategy: What’s the best place to publish content for LLM visibility?

So how do you go about creating content optimised for prompts? It isn’t just a matter of rephrasing blog posts, it also requires rethinking site architecture. Should it go on blog pages? Category pages? Educational hubs?

For adidas, for example, we add FAQ content to product category pages because they’re directly relevant to the products displayed there.

Structured content and schema markup also play a big role. Adding the right schema (FAQ, Product, HowTo, etc.) makes content easier for both traditional search engines and answer engines to understand. Structured answers are more likely to be picked up by LLMs, especially if the model is trained on your site’s data.

The forums factor: Should global brands use Reddit or Quora for GEO in the LLM era?

Several LLMs pull from user-generated content platforms like Reddit and Quora. Should brands post there? Probably not. Most forums prohibit overt brand promotion, and moderators will remove posts that look purely commercial. These spaces are designed for genuine community conversation, our role as brands is usually to listen.

Instead of “placing content,” I use forums to identify real user questions and then create better, more authoritative answers on my own site. It’s a customer-service mindset, listen first, then respond through intentional content.

Localisation amplified: What multilingual SEOs / GEOs need to know about prompt localisation

Now let’s add the complexity of multilingual organic optimisation. You can’t just translate keywords verbatim from one language to another and the same applies to prompts. The longer a query is, the more nuanced it becomes.

LLMs interpret prompts using a combination of linguistic, cultural, and contextual clues, which makes localisation even more important. For example, in Japan, users might never search about home security, whereas it’s a big topic elsewhere. These cultural differences still apply in the LLM era, and may even increase because of the conversational nature of prompts.

In traditional multilingual keyword research, I would identify high-volume search terms in different languages and localise them to fit each market’s cultural and linguistic nuances. In answer engines, nuance becomes even more critical. A prompt like “Which trendy adidas sneakers should I buy for a summer holiday in Mexico?” comes with a completely different set of expectations for the content (and the language) compared to a short keyword like “women’s adidas sneakers.”

There’s also an open question about how LLMs source multilingual information. Do they rely on highly ranked English content and translate it, or do they reference native-language sources directly? Does ChatGPT in German use German sources or translate English ones? If it’s the latter, English bias could creep into all responses. This is a factor multilingual SEO practitioners need to account for in their strategies.

How to adapt your multilingual SEO strategy for the LLM era

My advice for evolving your multilingual SEO practice in the LLM era is simple:

1. Keep your SEO fundamentals in order.
2. Create better content, and localise it properly.
3. Don’t rely solely on AI to write it.

This shift isn’t about abandoning the basics, it’s about adding more granularity, nuance, and quality to meet increasingly specific queries across languages and cultures.

Key takeaways:

1. Shift from keyword volumes to conversational prompts. Target the long-tail, natural-language questions users ask LLMs.
2. Prioritise high-quality multilingual content. Localise deeply rather than translating directly.
3. Use forums and customer feedback to guide prompt strategies.
4. Embed answers into structured content with schema markup.
5. Test manually to see how your brand appears in LLM environments.

Tracking multilingual GEO performance in the age of AI answers

Traditional SEO boils down to data, keyword rankings, traffic, search volumes, CTR, bounce rates… In LLM interactions, that clarity is gone. Right now, a lot of this is guesswork, and it’s far harder to measure.

We can still verify how users arrive at our sites via tools like Google Analytics, but it’s almost impossible to know how often our brand appears in LLM answers, or what prompts triggered those mentions. Tools are emerging to track LLM mentions, but this space is still immature.

Until analytics improve, I recommend being practical:

1. Check forums like Reddit and Quora to see what questions your potential customers are asking.
2. Conduct traditional FAQ research using SEO tools like Semrush.
3. Use “People Also Ask” questions in SERPs as inspiration.

The future: A feedback loop between traditional SEO and LLM optimisation

The line between traditional SEO and LLM optimisation is blurring. Rank well in Google, and you’re more likely to be referenced by answer engines. Be well-structured and authoritative, and your site may even become part of an LLM’s training data. If you rank number one in Google, you already have a higher chance of getting a mention in an LLM.

This isn’t the death of SEO, it’s an expansion. Multilingual prompt research is emerging as the next stage, requiring SEO teams to work more closely with content writers, localisation experts, data analysts, and customer experience teams. The fundamentals remain, but the playing field just got a whole lot bigger.

Author: Lorraine Van den Biggelaar, Head of SEO at Locaria