Your AI Visibility Changes With the Language

Ask an AI engine which supplier to use. Ask it in English, then ask the same question in Portuguese. You will not get a translated version of the same answer. You will get an answer built from a different set of sources entirely.

Most brands measuring their AI visibility are measuring it in one language, then assuming the result describes every market they sell in. It does not come close.

The short answer

Profound analysed 3.25 billion citations across seven engines and fourteen countries, filtered to native language prompts only, so a Brazilian prompt counted only if it was written in Portuguese. The source mix changes dramatically with the language.

In Google AI Overviews, YouTube took 38.0% of social citations in English and 65.0% in Portuguese. Instagram took 4.6% in English and 29.0% in Arabic. Reddit took 21.0% in English and 4.9% in Arabic. TikTok took 3.1% in English and 15.9% in Spanish.

That is the same engine answering the same kind of question. The only thing that changed was the language of the person asking.

The same question, a different internet

Here is the full comparison for Google AI Overviews, which is the surface with the widest reach of the seven measured.

Read down the columns rather than across the rows and the strategic picture appears.

English is the only column where Reddit is a major force at 21.0%. Almost every piece of advice written about AI visibility was written from this column, which is why so much of that advice recommends forum presence.

Portuguese is video first to an extreme degree. YouTube alone takes 65.0% of social citations, Reddit nearly disappears at 7.0%, and Instagram triples against the English baseline to 17.0%. LinkedIn is essentially absent at 1.1%.

Spanish spreads wider. YouTube still leads at 50.5%, but TikTok reaches 15.9%, five times its English share. A Spanish market strategy that ignores short video is ignoring a sixth of the social citations.

Arabic is the column that surprises people most. Instagram is the single biggest social source at 29.0%, LinkedIn is unusually strong at 8.0%, and Reddit is almost irrelevant at 4.9%.

How much social gets cited at all, by engine

Before acting on any of that, there is a prior question: how much does the engine lean on social sources in the first place? That varies more by engine than by language.

The spread runs from 15.3% down to 3.6%, which is more than a fourfold difference. Claude and Gemini barely touch social at all.

This is why two agencies can give a client opposite advice in good faith. One measured on Google AI Overviews and concluded social presence drives AI visibility. The other measured on Gemini and concluded it is irrelevant. Both readings were accurate about the engine they looked at.

There is a regional layer on top. In Spanish markets, social citation rates on Google AI Overviews reached 22.5% in Mexico and 20.0% in Spain, comfortably above the 15.3% overall figure. The same engine is meaningfully more social in those markets than its own average suggests.

Reddit is the one constant, but only on ChatGPT

The picture above is mostly about Google surfaces. ChatGPT behaves differently, and far more uniformly.

Reddit accounted for between 51% and 76% of ChatGPT's social citations in every single country measured.

Look at the top of that table. Reddit's grip on ChatGPT's social citations is stronger in India, Brazil, Sweden and Germany than it is in the United States or Britain. That runs against the intuition that Reddit is a mainly American platform whose influence fades abroad.

So the practical rule splits by engine. On Google surfaces, the local platform mix matters enormously and changes with the language. On ChatGPT, Reddit is close to a universal constant regardless of where the question comes from.

Why the language changes the sources

Three mechanisms explain most of it, and none of them are arbitrary.

The corpus in that language is genuinely different. There is far more Portuguese video content than Portuguese forum content, because that is how the Brazilian internet developed. The engine is not preferring YouTube in Portuguese out of policy. It is reflecting where the answers actually live.

Platform adoption differs by market. Instagram's dominance in Arabic tracks real usage patterns across the Gulf and the wider region. TikTok's strength in Spanish does the same. The engine is downstream of the market, not upstream of it.

Retrieval happens in the language of the prompt. When you ask in Portuguese, the engine looks for Portuguese sources to ground the answer. Your English page is often not in the candidate set at all, no matter how authoritative it is. This is the part that catches exporters: your authority in one language does not transfer.

That third point deserves emphasis because it is the one that costs money. A brand can be the most cited source in its category in English and be completely invisible in Spanish, not because it lost to a competitor, but because it never entered the pool.

What this breaks in a normal reporting setup

Nearly every AI visibility report starts the same way. Somebody writes twenty buyer questions in English, runs them, and produces a score. If the company sells in one English speaking market, that report is fine.

If it sells in more than one, the report is wrong in a specific and expensive way. It is not slightly imprecise. It is describing one market and being read as describing all of them.

Three failures follow from it.

You overstate visibility in markets where you have none, because the English number carries the whole business.

You misdiagnose the fix, because the tactic that would work in that market is not the tactic your English data points to. A team that reads the English column will go and earn forum mentions. In Portuguese that effort lands on a source taking 7.0% of social citations instead of 21.0%.

You cannot see a market moving, because the signal from a smaller market is averaged away inside a number dominated by your largest one.

If you sell in the Gulf

This is worth its own section because the Arabic column is the least like the English one, and the Gulf is a market a lot of companies are actively expanding into.

Instagram is the primary social source for Google AI Overviews in Arabic at 29.0%, more than six times its English share. LinkedIn, usually a minor player, reaches 8.0%, its strongest showing of the four languages. Reddit, the backbone of most English language advice, is down at 4.9%.

Meanwhile on ChatGPT, Saudi Arabia sits at 67% Reddit and the United Arab Emirates at 51%, the lowest Reddit share of any country measured.

So the two engines point in almost opposite directions in the same region. If your buyers use Google, your Arabic visibility work runs through Instagram and video. If they use ChatGPT, Reddit still matters even in a market where Arabic Google results barely touch it. You need to know which engine your buyers actually use before you spend anything.

Translating your playbook is not the same as translating your content

There is a difference between these two sentences, and it is the whole point of the article.

Translating your content means the same pages, in another language. Useful, and often necessary, but it only addresses whether the engine can read you.

Translating your playbook means changing which kinds of sources you go and earn, because the engine in that language prefers different ones. That is the part almost nobody does.

A worked version for a brand moving from English into Brazilian Portuguese:

None of that means abandoning written content. Written pages still do the heavy lifting for the non social share of citations, which is the large majority on every engine. It means the social portion of your plan, the part you spend outreach effort on, should be pointed at different platforms per market.

Social is the smaller half of the story

Everything above is about social platforms, because that is what the data breaks out. It is worth being honest about how much of the picture that actually covers.

Even on Google AI Overviews, the most social of the seven engines, social sources account for 15.3% of citations. On Gemini it is 3.6%. Which means that between roughly 85% and 96% of what these engines cite is something else: ordinary web pages, documentation, news, retailers, industry publications and the sites of the companies being discussed.

So the language effect described here is real, but it applies most visibly to a minority slice of citations. Two things follow.

The first is that your own pages remain the main event in every language. If you want to be cited in Spanish, having genuinely useful Spanish pages that answer the question directly will do more than any social strategy. The social mix tells you where to spend outreach effort, not where to spend most of your effort.

The second is that the non social share is where the language barrier bites hardest. A social post in Portuguese exists in Portuguese by default. Your product documentation probably does not. If your entire knowledge base is English only, you are absent from the large majority of the citation pool in every non English market, and no amount of Instagram presence compensates for that.

Language and geography are two separate dials

These get conflated constantly, and they are not the same thing.

Language is what the question is written in. It determines which corpus the engine retrieves from.

Geography is where the person asking is. It influences results through local ranking signals, regional availability and what the engine knows about the user's location.

You can move either one independently. An English question asked from Dubai is not the same as an Arabic question asked from Dubai, and neither is the same as an English question asked from London. The Profound data deliberately isolates language by filtering to native language prompts, which is the right way to study the language dial, but in real life both are moving at once.

This matters for testing. If you run your prompts in English from a London office and conclude something about your Gulf visibility, you have changed both dials and learned little. To get a usable reading, ask in the language your buyers use and, where your tooling allows it, from the market itself. That is why our own checks build prompts with the market in mind rather than running one generic English set for every customer.

A worked example: entering Brazil

Take a mid sized company that sells software, is well established in English, and has just opened a Brazilian operation. Their English AI visibility is good. Their pipeline in Brazil is empty and nobody can explain why.

A single English report tells them they are visible, so the problem gets blamed on sales. Here is what a market split report would have shown instead.

The fix in that scenario is not a bigger sales team. It is Portuguese pages that answer real buyer questions, plus video, because video is where 65.0% of the social citations in that language come from. Neither of those is visible from an English report, which is the entire problem.

The failure mode is worth naming plainly, because it repeats across companies. An English number that looks healthy provides false comfort. Nobody investigates a metric that appears green, so the market gap survives for quarters.

How to measure your own visibility per market

The method is not complicated, it is just rarely done.

1. Write your buyer questions in the target language, from scratch. Do not translate the English ones. People ask differently, not just in different words.
2. Run them from the market where possible, because geography influences results independently of language.
3. Keep each market in its own report. Never average them into one score.
4. Record which platform each citation came from, not only whether you were cited. The platform mix is the thing that tells you where to spend.
5. Repeat on a schedule, because the cited set rotates regardless of language.

If you only have the budget to do one market properly, do the one where you are already selling and cannot explain the pipeline. That is usually where a hidden visibility gap is quietly costing you.

What we see in our own logs

A small first hand observation, offered as a hint rather than a study, because the numbers are tiny.

This site is written entirely in English and targets an English speaking audience. Even so, a noticeable slice of the queries Bing reports for us are not in English. Over the last three months they include Portuguese searches for monitoring tools using the word citou, Spanish phrasing asking for a plataforma to monitor citations across ChatGPT, Perplexity and Google AI Overviews, and explicit requests for coverage of Brazilian Portuguese.

We did nothing to earn those. There is no Portuguese page on this site. The demand simply exists and is currently being served, badly, by English pages that happen to be topically close.

If a specialist English site with no localisation at all is being surfaced against Portuguese and Spanish queries, the reverse is almost certainly true for your category too. There is demand in languages you are not measuring, and somebody is being cited for it. At the moment it probably is not you, and you would have no way of knowing from an English report.

The bottom line

There is no such thing as your AI visibility. There is your AI visibility in a given language, on a given engine, in a given market, and those numbers are not close to each other.

The data is unambiguous on this. Between English and Portuguese, YouTube's share of Google AI Overviews social citations nearly doubles while Reddit's falls by two thirds. Between English and Arabic, Instagram goes from a rounding error to the single largest social source. Meanwhile ChatGPT leans on Reddit everywhere, most heavily in the markets you would least expect.

If you sell in one language, none of this changes your plan. If you sell in several and you are running one English report, you are not measuring most of your business. Split the report by market, record the platform mix and not just the score, and let each market's data pick its own tactics.

Sources and references

• Profound, how query language reshapes AI citations. 3.25 billion citations across ChatGPT, Claude, Google AI Mode, Google AI Overviews, Google Gemini, Microsoft Copilot and Perplexity, covering fourteen countries filtered to native language prompts only, March sample. Includes the Google AI Overviews platform mix by language, social citation rates by engine, and Reddit's 51% to 76% share of ChatGPT social citations by country. tryprofound.com
• Markets covered in that dataset: United States, Britain and Australia in English, Germany, France, Italy, Spain and Mexico, Brazil, Japan, Sweden, India, Saudi Arabia and the United Arab Emirates.

Cite this article

According to AI Citation Monitor's 2026 analysis of AI visibility by language, drawing on Profound's study of 3.25 billion citations across seven engines and fourteen countries, AI engines cite fundamentally different sources depending on the language of the question. In Google AI Overviews, YouTube took 38.0% of social citations in English but 65.0% in Portuguese, Instagram took 4.6% in English but 29.0% in Arabic, and Reddit fell from 21.0% in English to 4.9% in Arabic. Social citation rates also vary by engine, from 15.3% on Google AI Overviews down to 3.6% on Gemini. On ChatGPT, Reddit accounted for between 51% and 76% of social citations in every country measured, peaking at 76% in India and 71% in Brazil.

Source: AI Citation Monitor. Your AI Visibility Changes With the Language (2026). https://aicitationmonitor.com/blog/ai-visibility-by-language