Last updated: 26 July 2026. Every statistic below is attributed to its original publisher with a publication date. Where a figure is a forecast, vendor-reported telemetry, or an older study still in circulation, it is labelled as such. Figures we could not trace to a primary source have been removed — see Myths and misquoted statistics for what was cut and why.
Voice search is changing how consumers interact with search engines, local businesses, and digital services. But one of the biggest shifts in this space is not voice search itself — it is multilingual voice search.
Consumers are increasingly searching in their native languages, regional dialects, conversational phrases, and localized speech patterns. As AI assistants become more accurate and more widely available, multilingual voice search is opening real opportunities for local businesses, international brands, travel companies, and multilingual SEO strategies.
The evidence base for this shift is uneven. A great deal of what circulates as “voice search statistics” is recycled from a handful of 2017–2019 English-language studies, and some of the most-quoted numbers are misquoted or fabricated. What follows separates what is genuinely documented from what is not.
Key multilingual voice search statistics at a glance
| Statistic | Figure | Original publisher | Year | Type |
| Websites with English-language content | 49.6% | W3Techs | Updated 26 Jul 2026 | Ongoing census |
| Living languages worldwide | 7,100+ | Ethnologue | 2025–26 editions | Reference database |
| Online shoppers who prefer to buy in their native language | 76% | CSA Research, Can’t Read, Won’t Buy | 2020 (n=8,709, 29 countries) | Consumer survey |
| Consumers who will never buy from a website in another language | 40% | CSA Research | 2020 | Consumer survey |
| EU internet users who prefer to browse in their own language | 90% | European Commission, Flash Eurobarometer 313 | 2011 | Government survey |
| Indian internet users accessing the web in Indic languages | 870m of 886m (98%) | IAMAI–Kantar, Internet in India 2024 | Jan 2025 (n≈90,000) | Industry survey |
| Indian internet users relying on voice-based commands | ~1 in 5 | IAMAI–Kantar, Internet in India 2024 | Jan 2025 | Industry survey |
| Malayalam voice-to-text query ratio on MakeMyTrip’s Myra assistant | 46:1 | MakeMyTrip, reported by CNBC-TV18 | Mar 2026 | Vendor telemetry |
| Languages supported by Google speech recognition | 119 language varieties | Google, via The Verge | Aug 2017 baseline | Product announcement |
| Languages in Meta’s single multilingual ASR model | 1,107 | Meta AI, Scaling Speech Technology to 1,000+ Languages | 2023 | Peer-reviewed research |
| Google Cloud Chirp speech recognition accuracy, English | 98% | Google Cloud | 2023 | Vendor benchmark |
| Average AI Mode query length vs. traditional Search | 3× longer | May 2026 | First-party platform data | |
| US voice assistant users (forecast) | 157.1m in 2026 | EMARKETER | Sept 2025 | Forecast |
| UK adults who used a voice assistant in the past three months | 54% | Ofcom, Audio Listening in the UK 2025 | May 2025 | Regulator survey |
Multilingual voice search is expanding across global markets
Voice assistants are no longer limited to English-language search experiences — but language coverage varies far more between platforms than most coverage of this topic suggests.
Google has supported speech recognition in 119 language varieties since August 2017, when it added 30 languages including Amharic, Swahili, Bengali, Tamil, Telugu, Malayalam and Urdu in a single update, as reported by The Verge and Search Engine Land at the time. That figure applies to speech recognition — the dictation layer. Conversational assistants have historically supported far fewer languages, which is the distinction that matters for search visibility.
Language support by platform (as of July 2026):
| Platform | Voice/conversational language support | Source and date |
| Google speech recognition (voice typing, Voice Search) | 119 language varieties | Google, Aug 2017 announcement — still the last published headline count |
| Gemini Live (consumer voice conversations) | 40+ languages, up to two simultaneously per device | Google, Oct 2024 |
| Gemini Live API (developer surface) | 97 languages | Google AI for Developers documentation, current |
| Google Assistant (legacy, being retired) | ~30–40 languages | Google, Sept 2019 |
| Apple Intelligence (current Siri) | 16 languages | Apple Support, iOS 26.1 |
| Apple’s next-generation Siri AI | English only at launch | Apple Newsroom, 8 June 2026 |
| Amazon Alexa | 9 native languages; multilingual mode works only in English-anchored pairs | Amazon Developer documentation |
| ChatGPT (interface and voice) | 50+ languages; voice red-teamed in 45 at GPT-4o launch | OpenAI Help Center; OpenAI, May 2024 |
Two things stand out. First, the gap between recognition and conversation is wide: a system can transcribe a language without being able to hold a search conversation in it. Second, the leading assistants are moving in opposite directions — Google is expanding aggressively while Apple launched its rebuilt Siri in English only.
The platform landscape is also mid-transition. Google announced in March 2025 that Google Assistant on mobile would be upgraded to Gemini, and confirmed in a December 2025 support update that the migration would continue into 2026. Any strategy built on Google Assistant-era assumptions is being rebuilt underneath it.
Generative search is expanding multilingually at the same time. Google brought AI Overviews to more than 200 countries and 40+ languages in May 2025, adding Arabic, Chinese, Malay and Urdu, and launched AI Overviews across MENA and in Arabic globally the same month.
For businesses in multilingual markets, the practical implication is that visibility is no longer confined to typed English-language keywords — but the opportunity is uneven by language and by platform, and needs to be assessed market by market rather than assumed.
Regional-language voice search is growing faster than text search
The clearest evidence that voice unlocks native-language search behavior comes from India, and it comes from a company’s own product data rather than an independent study.
MakeMyTrip, India’s largest online travel company, released usage data from its GenAI assistant “Myra” in March 2026. According to figures the company gave to Indian business press and reported by CNBC-TV18, the Economic Times and India Outbound:
| Language | Voice-to-text query ratio on Myra |
| Malayalam | 46:1 |
| Tamil | 36:1 |
| Telugu | 32:1 |
MakeMyTrip reported these ratios against a base of over 50,000 daily conversations on Myra, across eight supported languages: Hindi, Bengali, Marathi, Tamil, Telugu, Kannada, Malayalam and English (Economic Times).
How to read these numbers. This is vendor-reported product telemetry, not independent research. MakeMyTrip did not publish a methodology, sample composition, or confidence intervals, and the ratios are affected by the product itself: regional-language text entry is comparatively rare on the platform, which inflates the denominator gap. The ratios demonstrate that voice removes a barrier for these users — they do not measure how Indians search generally.
The company’s own framing is appropriately cautious. Group CEO Rajesh Magow said in the March 2026 release: “For someone in Kochi or Coimbatore who thinks in Malayalam or Tamil, being able to simply speak their requirements, rather than type them in English, changes the experience meaningfully. It is still early, but these initial signals point to voice having the potential to make travel planning more inclusive and accessible across India” (India Outbound).
Independent survey data supports the underlying direction. The IAMAI–Kantar Internet in India 2024 report, published in January 2025 and based on roughly 90,000 respondents across 35 states and union territories, found that 870 million of India’s 886 million active internet users — 98% — accessed the internet in Indic languages, and that around one in five internet users relied on voice-based commands. The follow-up 2025 edition put India’s active internet base above 958 million (Business Standard).
The structural shift behind this predates AI assistants. The KPMG–Google report Indian Languages: Defining India’s Internet, published in April 2017, found Indian-language internet users had already overtaken English users (234 million vs. 175 million in 2016), that nine out of ten new internet users were Indian-language users, and that 99% of Indian-language internet users accessed the internet on mobile.
Google’s own data points the same way. In October 2025 the company stated that “People in India are power users of multimodal search, forming our largest user base for both voice and visual search globally” while expanding AI Mode and Search Live into Bengali, Kannada, Malayalam, Marathi, Tamil, Telugu and Urdu. In July 2026 Google announced that Gemini Live now converses in more than 25 Indian languages and dialects, including Sanskrit, Bhojpuri and Maithili, and that Project Vaani has open-sourced speech datasets covering 109 Indic languages.
Why voice outperforms text in these markets is straightforward and well documented: non-Latin script entry on mobile keyboards is slow, mobile is the dominant and often only access point, and speech requires no literacy in a second script. The KPMG–Google finding that 99% of Indian-language internet users are mobile-only is the mechanism, not a coincidence.
Voice search is becoming more conversational across languages
Spoken queries are longer and more conversational than typed ones. This is well supported — but it is also the area where the most-quoted statistic in the industry is wrong.
The correction. The frequently cited claim that “the average voice search query is 29 words” is a misquote. Backlinko’s 2018 analysis of 10,000 Google Home results found that “the typical voice search result is only 29 words in length” — that is the length of the answer Google reads back, not the length of the user’s question. The same study found that 40.7% of voice answers came from a featured snippet, that roughly 75% came from pages ranking in the top three organic results, and that the average voice result page contained 2,312 words. Those findings remain useful; the query-length claim built on top of them does not exist.
For actual query length, better evidence now exists.
MakeMyTrip’s Myra data provides a direct voice-versus-text comparison within one product. According to the March 2026 release reported by India Outbound and CNBC-TV18:
| Behaviour | Voice | Text |
| Queries longer than 11 words | 23% | 7% |
| Typical query | Full sentences with budget, group size, amenities | 3–4 words (“Goa hotels cheap”) |
| Date-specific queries | 3.3× higher than text | baseline |
| Informational queries (visa, documentation) | 2.7× higher than text | baseline |
| Location-specific queries | 25.1% of queries, 1.5× text | baseline |
| Average length of Hinglish (Hindi–English) queries | 10.5 words | — |
Google’s platform-wide data confirms the same trend in generative search. In April 2025, Google reported on its Q1 earnings call that AI Mode queries were “twice as long as traditional Search queries”. By May 2026, Google’s one-year AI Mode retrospective stated: “the average AI Mode search is triple the length of a traditional Search query” (Google). When Google expanded AI Mode to Arabic and 35 other languages in October 2025, it reported the same pattern in those markets.
For a typed baseline, Backlinko’s analysis of 306 million keywords (December 2020) found the average keyword is 1.9 words. The widely repeated “typed searches are 3–4 keywords” figure could not be traced to a primary, dated study; MakeMyTrip’s product-level observation of 3–4 word text queries is the closest verifiable comparison, and applies to travel search specifically.
The practical consequence is that multilingual voice queries carry conversational wording, local expressions and explicit intent. Instead of typing “hotel Paris cheap,” a user may ask “what’s the best affordable hotel near the Eiffel Tower?” — and increasingly, they ask it in their own language, or in a mix of two.
Code-switching is the behaviour most strategies miss. MakeMyTrip reported at its Myra 2.0 launch that around 70% of voice queries arrive in Hinglish, that voice prompts are roughly 40% longer and more complex than text, and that voice usage runs about 50% higher in non-metro markets than in metros (Metro India News, May 2026). Academic research has flagged this as a systemic gap: a 2022 study presented at the Conference on Conversational User Interfaces found voice assistants are “designed to be monolingual by default, lacking support for the bilingual dialogue experience.”
Mobile devices are driving multilingual voice adoption
Voice adoption is mobile-led, particularly in markets where the smartphone is the primary or only internet access point.
The most current global measure comes from DataReportal and We Are Social’s Digital 2025 report (February 2025), drawing on GWI survey data, which puts weekly voice-assistant use among internet users aged 16–64 at roughly 30% worldwide, with the UAE (35.4%), Mexico (35.0%) and India (33.6%) above the global average — a useful signal that adoption is not concentrated in English-speaking markets.
Two older figures still in wide circulation should be handled carefully:
- “27% of the global online population uses voice search on mobile” originates with GWI’s Voice Search report, based on a Q1 2018 survey of 90,021 internet users aged 16–64. Google later repeated the figure. It is a 2018 statistic and should be cited as such.
- “More than 50% of mobile searches are voice-based” is not supported by any source and has been removed from this article. It traces to a 2014 Fast Company interview in which Andrew Ng, then at Baidu, predicted that within five years at least 50% of searches would be “through images or speech” — a combined voice-and-image prediction that was later stripped of its qualifier and misattributed to ComScore, which says it never published it. The debunking is documented by Brodie Clark and Econsultancy. Google’s last official statement on voice share was Sundar Pichai’s 2016 remark that 20% of US mobile app queries were voice, and it has not been updated since.
Where mobile-first behaviour is documented, it is strong. The KPMG–Google 2017 study found 99% of Indian-language internet users access the internet via mobile. IAMAI–Kantar found rural India — 488 million users, 55% of the total — growing faster than urban. MakeMyTrip reported that over 45% of Myra’s usage comes from tier-2 and smaller cities.
In these markets, mobile-first, conversational, multilingual optimization is not an incremental refinement. It matches how the majority of users already reach the internet.
“Near me” searches and local intent in multilingual voice search
Voice search and local intent overlap, but the specific numbers used to make that case are usually misstated.
The correction. The claim “76% of voice searches are local or near me” does not appear in any primary study. BrightLocal’s Voice Search for Local Business Study, fielded in April 2018 among 1,012 US consumers, found that 76% of smart speaker users who search locally do so at least weekly, with 53% searching daily. That is a frequency measure among a device-owning subset, not a share of all voice queries. The same study found that 58% of consumers had used voice search to find local business information in the previous 12 months. Both are 2018 US figures and should be dated when cited.
On “near me” growth. Google has published several different growth figures under the “near me” heading, which are routinely conflated. On Think with Google, Google reported that “near me” mobile searches containing a variant of “can I buy” or “to buy” grew over 500% over two years; that “_ near me today/tonight” grew over 900%; that “_ near me now” grew over 150%; and that “open” + “now” + “near me” grew over 200%. A later Google figure covering September 2020 to August 2021 recorded “open now near me” growing over 400% year on year. None of these is a single blanket “near me searches grew X%” statistic.
What is not in dispute is that spoken local queries are conversational and immediate — “best dentist near me,” “hotel near airport,” “restaurant open now” — and that in multilingual markets those queries increasingly arrive in Arabic, Hindi, Tamil, Spanish, French and regional dialects rather than English. Google’s expansion of AI Overviews into Arabic globally and AI Mode into seven Indic languages is the platform-side acknowledgement of that demand.
Businesses optimized only for English-language local SEO are, in these markets, invisible to a growing share of local intent. But the size of that share has not been measured publicly by language, and any agency claiming otherwise is extrapolating.
AI assistants are accelerating multilingual search behavior
Assistant usage is substantial and growing, though the numbers most often quoted are older forecasts rather than measurements.
Current, defensible figures:
- EMARKETER’s Voice Assistant User Forecast 2025 (September 2025) projects 153.5 million US voice assistant users in 2025, rising to 157.1 million in 2026 and 168.2 million by 2029. These are forecasts, not counts.
- EMARKETER also puts US brand usage at Google Assistant 91.9 million, Siri 86.5 million and Alexa 77.2 million users in 2025 (EMARKETER).
- In the UK, Ofcom’s Audio Listening in the UK 2025 (May 2025) found 54% of UK adults had used a voice assistant in the past three months and 41% of households own a smart speaker, with Alexa used by 66% of assistant users, Google Assistant 31% and Siri 28%.
Two widely circulated figures that need labels:
- The “8.4 billion voice assistants globally” figure is Juniper Research’s April 2020 forecast for the year 2024 — a projection of devices, not users, made six years ago for a year that has since passed. No verified first-party actual count for 2024 or 2025 was found. It should never be presented as a current measurement.
- The claim that “Google Assistant and Siri each hold 36% market share” comes from Microsoft’s 2019 Voice Report, a survey of roughly 7,000 respondents measuring self-reported usage — not market share, and now seven years old. It is contradicted by the EMARKETER and Ofcom figures above and has been removed from this article.
What is genuinely new is capability rather than count. Google’s Gemini Live supports 40+ languages for consumer voice conversations and 97 languages via its developer API, and more than 25 Indian languages and dialects as of July 2026. Google Assistant’s Arabic service, which launched in Saudi Arabia and Egypt in April 2019, expanded to 15 further MENA markets in 2021.
Where multilingual voice search still falls short
The honest counterweight to the growth narrative is that speech recognition quality varies enormously by language, and the gap is documented in primary research rather than marketing material.
Accuracy is not uniform. Google Cloud reported that its Chirp model, the productionized version of its Universal Speech Model, “delivers 98% speech recognition accuracy in English and over 300% relative improvement in several languages with less than 10 million speakers” (Google Cloud). A 300% relative improvement implies a starting point far below English parity. That single sentence is the clearest first-party statement of the high-resource/low-resource divide.
The research base confirms it. OpenAI’s Whisper documentation states plainly that “Whisper’s performance varies widely depending on the language”, publishing per-language error rates rather than a single accuracy figure. Meta’s MMS project (2023) built a single ASR model covering 1,107 languages and reported that it “more than halves the word error rate of Whisper on 54 languages of the FLEURS benchmark.” Google’s Universal Speech Model covers 100+ languages, pre-trained on 12 million hours across 300+ languages, as part of the company’s 1,000 Languages Initiative. The FLEURS benchmark that these models are measured against covers 102 languages across 17 language families and 27 writing systems — a fraction of the 7,100+ living languages catalogued by Ethnologue.
Code-switched speech performs worst of all. The SwitchLingua dataset (2025), covering 12 languages and 174 speakers across 18 countries, found “substantial performance gaps” for state-of-the-art ASR models on code-switched speech. A 2025 perception study in Frontiers in Computer Science found that code-switched text-to-speech output is measurably less intelligible to bilingual listeners than monolingual output. Given that MakeMyTrip reports roughly 70% of its voice queries arriving in Hinglish, this is not an edge case.
The 95% accuracy figure is nine years old. The often-repeated claim that voice recognition has “surpassed 95% accuracy” traces to Mary Meeker’s 2017 Internet Trends report, citing Google’s English-language word accuracy rate as of May 2017. No newer official Google accuracy figure has been published. Citing an English word-accuracy benchmark from 2017 as evidence for multilingual capability is a category error.
Consumer preference for native-language experiences
Preference for one’s own language is one of the best-evidenced findings in this entire subject area — better evidenced, in fact, than most voice search statistics.
CSA Research’s Can’t Read, Won’t Buy study, based on 8,709 verified consumer responses across 29 countries and published in 2020, found that 76% of online shoppers prefer to buy products with information in their native language and 40% will never buy from websites in other languages. CSA’s own conclusion: “if a company chooses to not localize the buying experience, they risk losing 40% or more of the total addressable market.” The same study found 73% want product reviews in their own language and 75% are more likely to make repeat purchases when post-sale support is in their language.
The European Commission reached compatible conclusions in Flash Eurobarometer 313, User Language Preferences Online: 90% of EU internet users prefer to access websites in their own language, 44% feel they miss information because pages are not in a language they understand, and only 18% buy products online in a foreign language. This is 2011 government survey data — dated, but methodologically solid and never contradicted by a comparable EU-wide study since.
Language ability is a real constraint, not a preference. The EF English Proficiency Index 2025, covering 123 countries and 2.2 million test-takers, found that in over half the countries measured, speaking is the weakest English skill — which is precisely the skill a voice interface demands.
On voice specifically, the evidence is thinner and older. PwC’s 2018 Consumer Intelligence Series, based on a nationally representative sample of 1,000 US adults aged 18–64, found that among voice assistant users, 71% favoured voice commands over typing for search. That is an eight-year-old, US-only, English-language finding among existing users, and should be cited with those qualifiers. The related claim that “90% say voice search is easier than traditional search” could not be traced to any primary source and has been removed.
The mechanism in multilingual markets is more concrete than a preference statistic: typing in Devanagari, Tamil, Malayalam or Arabic script on a mobile keyboard is slower than speaking, and speech requires no second-script literacy at all.
SEO opportunities for businesses in 2026
1. Multilingual local SEO
Google’s Business Profile representation guidelines require that a business be represented “as it’s consistently represented and recognized in the real world” — signage, stationery and branding. This is the operative rule for how to handle business names in multiple languages or scripts, and it constrains the common tactic of translating or transliterating a name for keyword reasons.
Practical priorities: localized Business Profile content, location pages in the target language rather than machine-translated English, regional-language keyword and question research, and review responses in the language the review was written in.
2. Conversational and question-led content
Voice and AI-mediated queries are longer and question-shaped. Google’s own data shows AI Mode queries at three times the length of traditional search queries, and MakeMyTrip’s telemetry shows 23% of voice queries exceeding 11 words versus 7% of text queries.
Backlinko’s 2018 finding that 40.7% of voice answers came from featured snippets and roughly 75% from top-three organic results still describes the underlying mechanic: concise, self-contained answers inside comprehensive pages. What has changed is the surface — answers are now increasingly synthesized by AI Overviews and AI Mode rather than read verbatim from a snippet, so the goal is being a citable source rather than owning a single box.
3. Technical multilingual foundations
This is where most multilingual voice strategies fail, and Google documents the requirements directly. Per Google Search Central’s guidance on multi-regional and multilingual sites:
- Use distinct URLs per language, not cookies or browser-based switching
- Do not auto-redirect users based on guessed language
- Keep one language per page — avoid side-by-side translations
- Use UTF-8 encoding throughout
- Remember that Googlebot crawls from the US and does not set an Accept-Language header, so JavaScript or cookie-based localization can leave language versions unindexed
Alternate-language versions should be declared with hreflang annotations, which Google accepts via HTML link tags, HTTP headers or XML sitemaps. Google notes that translated pages are only treated as duplicate content when the main content itself is untranslated.
4. Regional-language content expansion
Given CSA Research’s finding that 40% of consumers will never buy in another language, translated FAQs, regional-language landing pages and multilingual support are revenue infrastructure rather than SEO garnish. Machine translation is a starting point, not an endpoint — CSA found 66% of consumers use online machine translation themselves, which means they can recognize it.
For markets with heavy code-switching, plan for mixed-language queries explicitly. Hinglish, Spanglish and Arabizi are documented, researched linguistic behaviours, not errors to be corrected.
5. Measurement caveat
Voice queries are not separately reported in Google Search Console or GA4, and no major search engine publishes voice query share. There is no way to measure multilingual voice search directly. Realistic proxies include language-segmented organic performance, question-format query growth, long-tail query length distribution, and Business Profile call and direction actions by locale. Any vendor promising direct voice search attribution is selling something that does not exist.
Multilingual voice search readiness checklist
| # | Check | Why it matters |
| 1 | Distinct, crawlable URL per language version | Google explicitly recommends against cookie/browser-based switching |
| 2 | hreflang declared via tags, headers or sitemap | Signals alternate versions to Google |
| 3 | No automatic language redirects | Googlebot crawls from the US without Accept-Language |
| 4 | UTF-8 encoding site-wide | Required for non-Latin script rendering and indexing |
| 5 | One language per page | Mixed-language pages confuse language detection |
| 6 | Native-speaker review of all translated content | 66% of consumers use machine translation themselves and recognize it |
| 7 | Question-format headings in each target language | Matches conversational query structure |
| 8 | Concise, self-contained answers under each question | Structure most likely to be extracted or cited |
| 9 | Business Profile name matching real-world signage | Required by Google’s representation guidelines |
| 10 | Localized Business Profile categories, hours and attributes | Drives local intent matching |
| 11 | Keyword research conducted in-language, not translated from English | Translated keyword lists miss native phrasing |
| 12 | Code-switched query variants documented for relevant markets | ~70% of MakeMyTrip’s voice queries are Hinglish |
| 13 | Transliteration variants covered where scripts differ | Users type and speak in mixed scripts |
| 14 | Click-to-call and directions optimized on mobile | Voice-local intent is action-oriented |
| 15 | Language-segmented reporting configured | The only workable proxy for voice performance |
Myths and misquoted statistics in voice search
| Claim in circulation | Status | What the source actually says |
| “50% of all searches will be voice by 2020” | Fabricated | Traces to a 2014 Andrew Ng prediction about voice and image search, misattributed to ComScore, which says it never published it (Brodie Clark) |
| “The average voice search query is 29 words” | Misquoted | Backlinko measured the average voice result at 29 words, not the query (Backlinko, 2018) |
| “76% of voice searches are local” | Misquoted | BrightLocal found 76% of smart speaker users who search locally do so at least weekly (BrightLocal, 2018) |
| “8.4 billion voice assistants are in use” | Forecast stated as fact | A Juniper Research forecast from April 2020 for the year 2024, counting devices |
| “153.5 million Americans use voice assistants” | Forecast stated as fact | EMARKETER’s forecast for 2025, already superseded by its own 2026 projection |
| “Google Assistant and Siri each hold 36% market share” | Wrong metric, stale | A 2019 Microsoft survey of ~7,000 respondents measuring self-reported usage, not market share |
| “Voice recognition is 95% accurate” | Stale and English-only | Google’s English word accuracy rate as of May 2017, via Mary Meeker’s Internet Trends report |
| “Typed searches average 3–4 keywords” | Unsourced | Backlinko’s 306-million-keyword study found an average of 1.9 words (Backlinko, 2020) |
| “27% of the global online population uses voice search on mobile” | Correct but dated | GWI, Q1 2018 survey of 90,021 internet users |
Frequently asked questions
Is multilingual voice search actually growing, or is this marketing?
The direction is supported, the magnitude is not well measured. Platform-side expansion is documented (Google’s AI Overviews in 40+ languages, AI Mode in Arabic and seven Indic languages, Gemini Live in 25+ Indian languages). Behaviour-side evidence is strongest in India, where IAMAI–Kantar found 98% of internet users accessing content in Indic languages and about one in five using voice commands. No search engine publishes voice query volume by language.
What share of searches are voice searches?
Nobody knows publicly. Google’s last official figure was 20% of US mobile app queries in 2016 and has not been updated. Any percentage claim above that should be treated as unsourced.
Which languages do voice assistants actually support?
It depends on the layer. Google has supported speech recognition in 119 language varieties since 2017; Gemini Live handles 40+ conversational languages and 97 via its developer API; Apple Intelligence supports 16 and the new Siri AI launched in English only; Amazon Alexa natively supports nine with English-anchored multilingual pairs.
Is voice recognition accurate enough in non-English languages?
Increasingly, but unevenly. Google Cloud reports 98% accuracy in English while describing “over 300% relative improvement” in languages with fewer than 10 million speakers — implying a much weaker baseline. OpenAI publishes per-language error rates for Whisper precisely because performance “varies widely depending on the language.”
How should I handle code-switching like Hinglish or Spanglish?
Treat it as a first-class query type. MakeMyTrip reports roughly 70% of its voice queries arrive in Hinglish. Research shows ASR systems still handle code-switched speech worse than monolingual speech (SwitchLingua, 2025), so mixed-language phrasings should be documented in keyword research and reflected in on-page question phrasing.
Can I measure multilingual voice search performance?
Not directly. Search Console and GA4 do not segment voice queries. Use language-segmented organic performance, long-tail and question-query growth, and Business Profile actions by locale as proxies.
Does translating my site automatically capture multilingual voice traffic?
No. Google requires distinct URLs per language and hreflang annotations, and warns that JavaScript or cookie-based localization can leave language versions unindexed because Googlebot crawls from the US without an Accept-Language header. Translation without technical implementation produces content Google may never see.
Which markets should I prioritize?
Prioritize by the combination of large non-English internet populations, mobile-first access, and documented voice adoption. India is the clearest case: Google describes it as its “largest user base for both voice and visual search globally.” MENA is a second, following Google’s Arabic AI Overviews and AI Mode rollouts. Beyond those, verify with market-level data rather than global averages.
The future of search is increasingly multilingual and voice-driven
Voice search is not solely an English-language phenomenon, and the platform investment behind multilingual capability is now substantial and documented: 1,107 languages in Meta’s ASR model, 100+ in Google’s Universal Speech Model, AI Overviews in 40+ languages, Gemini Live in 25+ Indian languages alone.
What the evidence supports:
- Native-language preference is strongly documented — 76% of consumers prefer to buy in their own language and 40% will not buy in another (CSA Research)
- Voice lowers the barrier for non-Latin-script and mobile-only users, most visibly in India, where 98% of internet users access content in Indic languages (IAMAI–Kantar)
- Conversational queries are getting longer, confirmed by Google’s own AI Mode data (3× traditional query length)
- Platform language coverage is expanding quickly, though unevenly across vendors
What it does not yet support:
- Any reliable figure for what share of searches are voice, in any language
- Claims that multilingual voice adoption has reached parity with English
- The assumption that speech recognition works equally well across languages — primary research shows it does not
Businesses that build multilingual foundations now — proper hreflang implementation, in-language keyword research, native-speaker content, code-switching awareness — are positioning for a shift that is real but still early. The honest case for acting is not that the numbers are overwhelming. It is that the technical groundwork takes time, the consumer preference evidence is unambiguous, and the platforms are clearly building for it.
How this article was verified
Every statistic in this article was traced to its original publisher rather than to a statistics-aggregator blog. Figures that could not be traced to a named, dated, primary source were removed rather than rephrased. Forecasts are labelled as forecasts, vendor-reported product data is labelled as vendor telemetry, and any figure older than three years carries its publication date inline. Statistics that are widely circulated but demonstrably misquoted are documented in the myths table rather than silently omitted, so that readers can recognize them elsewhere.
Corrections are welcome. If a figure here is out of date or a better primary source exists, we will update the article and note the change.
Sources
Voice search behaviour and SEO studies
- Backlinko. Voice Search SEO Study: Results From 10,000 Google Home Search Results. 2018. https://backlinko.com/voice-search-seo-study
- Backlinko. We Analyzed 306M Keywords. Here’s What We Learned About Search. 2020. https://backlinko.com/google-keyword-study
- BrightLocal. Voice Search for Local Business Study. 2018 (n=1,012 US consumers). https://www.brightlocal.com/research/voice-search-for-local-business-study/
- GWI (GlobalWebIndex). Voice Search. 2018 (Q1 2018 survey, n=90,021). https://www.gwi.com/hubfs/Downloads/Voice-Search-report.pdf
- PwC. Consumer Intelligence Series: Prepare for the Voice Revolution. 2018 (n=1,000 US adults). https://www.pwc.com/us/en/advisory-services/publications/consumer-intelligence-series/voice-assistants.pdf
Market size, adoption and forecasts
- DataReportal / We Are Social. Digital 2025: Global Overview Report. February 2025. https://datareportal.com/reports/digital-2025-global-overview-report
- EMARKETER. Voice Assistant User Forecast 2025. September 2025. https://www.emarketer.com/content/voice-assistant-user-forecast-2025
- EMARKETER. Data Drop: Gen Z Leading Voice Assistant Growth. https://www.emarketer.com/content/data-drop-gen-z-leading-voice-assistant-growth
- Juniper Research. Number of Voice Assistant Devices in Use to Overtake World Population by 2024. April 2020 (forecast). https://www.juniperresearch.com/press/number-of-voice-assistant-devices-in-use/
- Ofcom. Audio Listening in the UK 2025. May 2025. https://www.ofcom.org.uk/siteassets/resources/documents/research-and-data/data/statistics/2025/audio-report-2025/audio-report-2025.pdf
Language, localization and consumer preference
- CSA Research. Can’t Read, Won’t Buy – B2C. 2020 (n=8,709, 29 countries). https://csa-research.com/Featured-Content/For-Global-Enterprises/Global-Growth/CRWB-Series/CRWB-B2C
- CSA Research. Need It or Want It? English May Be the Only Way to Get It. July 2020. https://csa-research.com/Blogs-Events/Blog/cant-read-wont-buy-consumer-language-preferences
- EF Education First. EF English Proficiency Index 2025. November 2025 (123 countries, 2.2m test-takers). https://www.ef.edu/about-us/press/articles/2025/ef-english-proficiency-index-2025-launched/
- Ethnologue. How Many Languages Are There in the World? 2025–2026 editions. https://www.ethnologue.com/insights/how-many-languages/
- European Commission. User Language Preferences Online (Flash Eurobarometer 313). May 2011. https://europa.eu/eurobarometer/surveys/detail/880
- W3Techs. Usage Statistics of Content Languages for Websites. Updated 26 July 2026. https://w3techs.com/technologies/overview/content_language
India regional-language internet and voice data
- IAMAI & Kantar. Internet in India Report 2024. January 2025 (n≈90,000). https://www.iamai.in/sites/default/files/research/Kantar_%20IAMAI%20report_2024_.pdf
- KPMG in India & Google. Indian Languages: Defining India’s Internet. April 2017. https://assets.kpmg.com/content/dam/kpmgsites/in/pdf/2017/04/Indian-languages-Defining-Indias-Internet.pdf
- Business Standard. Indian Internet User Base Crosses 950 Million in 2025, IAMAI Report. January 2026. https://www.business-standard.com/industry/news/indian-internet-user-base-crosses-950-million-in-2025-iamai-report-126012901048_1.html
MakeMyTrip “Myra” voice data (vendor-reported telemetry)
- CNBC-TV18. Voice Travel Queries Now Longer and More Detailed, With Date Searches 3 Times Higher. March 2026. https://www.cnbctv18.com/travel/voice-travel-queries-now-longer-and-more-detailed-with-date-searches-3-times-higher-report-ws-el-19864858.htm
- Economic Times. Indians Are Beginning to Search for Travel in the Language They Speak: MakeMyTrip. March 2026. https://economictimes.indiatimes.com/industry/services/travel/indians-are-beginning-to-search-for-travel-in-the-language-they-speak-makemytrip/articleshow/129332400.cms
- India Outbound. Voice Search Queries Top 50,000 Daily, With 23% Over 11 Words: MakeMyTrip. March 2026. https://indiaoutbound.info/trade-news/voice-search-queries-top-50000-daily-with-23-pc-over-11-words-makemytrip/
- Economic Times. MakeMyTrip Expands Myra Into a Full Conversational Booking Assistant. May 2026. https://economictimes.indiatimes.com/magazines/panache/makemytrip-expands-myra-into-a-full-conversational-booking-assistant/articleshow/131069006.cms
- Metro India News. MakeMyTrip Unveils Myra 2.0. May 2026. https://www.metroindia.net/news/articlenews/makemytrip-unveils-myra-39875
- Google Cloud. MakeMyTrip and Google Cloud Partner to Deliver Faster, Simpler Planning and Booking. October 2025. https://www.googlecloudpresscorner.com/2025-10-08-MakeMyTrip-and-Google-Cloud-Partner-to-Deliver-Faster,-Simpler-Planning-and-Booking-for-Millions-of-Travellers
Platform announcements and official documentation
- Google. AI Overviews Expand to Over 200 Countries and Territories. May 2025. https://blog.google/products-and-platforms/products/search/ai-overview-expansion-may-2025-update/
- Google. Bringing AI Overviews to MENA, and in Arabic Globally. May 2025. https://blog.google/intl/en-mena/product-updates/explore-get-answers/bringing-ai-overviews-to-mena-and-in-arabic-globally/
- Google. How AI Mode Is Changing the Way People Search in the U.S. May 2026. https://blog.google/products-and-platforms/products/search/ai-mode-us-insights/
- Google. Alphabet Q1 2025 Earnings Call: CEO’s Remarks. April 2025. https://blog.google/company-news/inside-google/message-ceo/alphabet-earnings-q1-2025/
- Google. The Assistant Experience on Mobile Is Upgrading to Gemini. March 2025. https://blog.google/products-and-platforms/products/gemini/google-assistant-gemini-mobile/
- Google. Update on Upgrading Mobile Assistant Devices to Gemini. December 2025. https://support.google.com/gemini/thread/396052272/
- Google. Gemini Live and Connected Google Apps in More Languages. October 2024. https://blog.google/products-and-platforms/products/gemini/gemini-live-extensions-language-expansion/
- Google AI for Developers. Live API Capabilities Guide. Current. https://ai.google.dev/gemini-api/docs/live-api/capabilities
- Google. Supercharging Search for India: New Languages in AI Mode, Search Live Debuts. October 2025. https://blog.google/intl/en-in/products/supercharging-search-for-india-new-languages-in-ai-mode-search-live-debuts/
- Google. Deepening Our Commitment to India’s AI Ambition. July 2026. https://blog.google/intl/en-in/company-news/deepening-our-commitment-to-indias-ai-ambition/
- Google. Meet the Google Assistant: Available in Arabic to Saudi Arabia and Egypt. April 2019. https://blog.google/intl/en-mena/product-updates/explore-get-answers/2019-google-assistant-arabic-en/
- Google. 3 Ways AI Is Scaling Helpful Technologies Worldwide (1,000 Languages Initiative). November 2022. https://blog.google/innovation-and-ai/products/ways-ai-is-scaling-helpful/
- Google Search Central. Managing Multi-Regional and Multilingual Sites. Current. https://developers.google.com/search/docs/specialty/international/managing-multi-regional-sites
- Google Search Central. Localized Versions of Your Pages (hreflang). Current. https://developers.google.com/search/docs/specialty/international/localized-versions
- Google Business Profile Help. Guidelines for Representing Your Business on Google. Current. https://support.google.com/business/answer/3038177
- Think with Google. How “Near Me” Shopping Searches Have Changed. c. 2019. https://business.google.com/in/think/marketing-strategies/near-me-searches/
- Think with Google. Trend Chapter: Search Behaviour Data (Sept 2020 – Aug 2021). 2021. https://www.thinkwithgoogle.com/_qs/documents/16400/TWG_Trend_4_Chapter_wBXdOUy.pdf
- Apple. How to Get Apple Intelligence. Current. https://support.apple.com/en-us/121115
- Apple Newsroom. Apple Introduces Siri AI. June 2026. https://www.apple.com/newsroom/2026/06/apple-introduces-siri-ai-a-profoundly-more-capable-and-personal-assistant/
- Amazon Developer. Alexa Voice Service International — Supported Regions and Locales. Current. https://developer.amazon.com/en-US/alexa/devices/alexa-built-in/international
- OpenAI Help Center. How to Change Your Language Setting in ChatGPT. Updated July 2026. https://help.openai.com/en/articles/8357869-how-to-change-your-language-setting-in-chatgpt
- OpenAI. Hello GPT-4o. May 2024. https://openai.com/index/hello-gpt-4o/
- The Verge. Google Now Recognises 119 Languages for Dictation and Voice Typing. August 2017. https://www.theverge.com/2017/8/14/16142786/google-recognises-119-languages-dictation-voice-typing
- Search Engine Land. Google Adds 30 Languages to Voice Search. August 2017. https://searchengineland.com/google-just-added-30-languages-voice-search-ability-speak-emoji-u-s-280620
Multilingual speech recognition research
- Radford, A. et al. (OpenAI). Robust Speech Recognition via Large-Scale Weak Supervision (Whisper). 2022. https://arxiv.org/abs/2212.04356
- OpenAI. Whisper GitHub Repository — Per-Language Performance. Current. https://github.com/openai/whisper
- Pratap, V. et al. (Meta AI). Scaling Speech Technology to 1,000+ Languages (MMS). 2023. https://arxiv.org/abs/2305.13516
- Zhang, Y. et al. (Google). Google USM: Scaling Automatic Speech Recognition Beyond 100 Languages. 2023. https://arxiv.org/abs/2303.01037
- Conneau, A. et al. (Google). FLEURS: Few-Shot Learning Evaluation of Universal Representations of Speech. 2022. https://arxiv.org/abs/2205.12446
- Google Cloud. Bringing the Power of Large Models to Google Cloud’s Speech API (Chirp). 2023. https://cloud.google.com/blog/products/ai-machine-learning/bringing-power-large-models-google-clouds-speech-api
Code-switching and multilingual voice interaction research
- Cihan, H. et al. Bilingual by Default: Voice Assistants and the Role of Code-Switching in Creating a Bilingual User Experience. CUI 2022. https://arxiv.org/abs/2206.09765
- SwitchLingua: The First Large-Scale Multilingual and Multi-Ethnic Code-Switching Dataset. 2025. https://arxiv.org/html/2506.00087v1
- The Perception of Code-Switched vs. Monolingual Sentences in TTS Voices. Frontiers in Computer Science, August 2025. https://www.frontiersin.org/journals/computer-science/articles/10.3389/fcomp.2025.1565604/full
- Code-Switching in End-to-End Automatic Speech Recognition: A Systematic Literature Review. LREC 2026. https://arxiv.org/html/2507.07741v1
Debunking references
- Clark, B. Stop Using ComScore’s 2020 Voice Search Stat. https://brodieclark.com/stop-using-comscores-2020-voice-search-stat/
- Econsultancy. Why We Need to Stop Repeating the “50% by 2020” Voice Search Prediction. https://econsultancy.com/why-we-need-to-stop-repeating-the-50-by-2020-voice-search-prediction/