Interest in Speech Recognition: Wikipedia Page Views

People read about speech and typing on Wikipedia, but what they read has moved. Interest has shifted away from the technology and its older products, and toward the chatbots and assistants that people use by name. The dataset holds 39 data points ¹, each the page views by people (not known crawlers) of one article over the past year, together with the latest month and the same month a year earlier.

Three patterns stand out. A handful of assistant and chatbot articles dominate the table. The long-standing explainer articles on speech recognition, typing and dictation sit in a steady middle band, mostly below their earlier levels. Non-English articles on speech recognition are small and, for the oldest ones, much smaller than when tracking began.

Chatbots and assistants draw the most readers, and the newest draw the most

The top of the table is not about dictation. It is about products with a brand name that the news covers.

Showing 8 of 33 rows. See every row in the dataset.

The English article on Google Gemini had 1,093,862 page views ². Grok followed with 1,037,826 ³, and Microsoft Copilot with 644,119 ⁴. All three are articles that did not exist in their current form when tracking began: their histories start only a few years back.

Their curves are not alike. Gemini's latest month is close to its all-time high and far above the same month a year earlier, which is the steepest rise in the table. The year-earlier figure looks too small for a product that was already well known, so part of that jump may come from how the article was named or reached in search at the time. Grok and Copilot went the other way: their latest months are below the same month a year earlier, and both are well under their peaks. Copilot's peak came early, and Grok's came in a single spike.

The older voice assistants are mixed, and Siri is the one that is growing

Among the voice assistants that came before the chatbots, the picture differs by product.

Siri had 307,597 page views ⁶, the most of any assistant that is mainly voice. Its latest month is well above the same month a year earlier, though far below its first readings and its peak. Amazon Alexa had 191,749 ⁸. Its history shows the biggest collapse in the table: it peaked in its early years, at a level many times what it draws now. Alexa's latest month is a little below a year earlier, so the slide has slowed.

Google Assistant had 124,363 ¹⁰ and Cortana had 116,720 ¹¹. Cortana is the only article in the table whose latest month is level with a year earlier. Bixby had 65,260 ¹². The generic article on virtual assistants had 71,308 ¹³, well below the peak it reached earlier.

The general explainers hold a steady middle band, mostly below where they started

The articles that explain the subject, rather than a product, are close to each other in size.

• Speech recognition in English: 70,725 page views ¹⁴.
• Typing: 70,226 ¹⁵.
• Touch typing: 85,170 ¹⁶.
• Words per minute: 76,491 ¹⁷.
• Stenotype: 96,217 ¹⁸.

Speech recognition and typing are nearly tied. That is worth noting for anyone who assumes that talking to a computer has replaced typing in people's minds: as a topic to read about, they draw about the same. Speech recognition itself reads well below where it started, and its latest month is a little under a year earlier. Typing is almost flat from its first reading to its latest, though it has had higher months. Touch typing had its high well after tracking began, which suggests a one-off cause such as a link from a popular page, so treat that article's level with caution.

Speech synthesis, the opposite direction of the technology (text to voice), had 147,894 page views ⁹. It is the only general explainer whose latest month is above a year earlier and above its first reading. Readers seem more curious about machines that talk than about machines that listen, probably because of synthetic voices in the news.

The more specialized dictation articles are far smaller. Dictation machines had 17,091 ¹⁹, speech-to-text 1,560 ²⁰, and Windows Speech Recognition 4,951 ²¹. Nobody much reads about dictation as a named method. The searches go to products and to the broader term.

Dictation products have lost most of their readers

The registry products with articles show a consistent decline.

Dragon NaturallySpeaking had 39,540 page views ²², and Nuance Communications had 32,408 ²³. Both are below their peaks. Braina fell from its first reading to a small share of it, with 7,350 ²⁴ in the past year. Whisper, the open speech recognition model, had 54,254 ²⁵, with a latest month under its peak and under a year earlier.

An article for a product is read when someone is deciding whether to try it. A fall means fewer people looking up these names, and it matches the move to built-in dictation and to chatbots with a voice mode. It does not prove people use these tools less. For the market context see our market overview.

Typing injuries still draw readers, but fewer than before

Carpal tunnel syndrome had 191,983 page views ⁷, as many as Alexa. Repetitive strain injury had 50,361 ²⁶. Both have fallen sharply from their first readings, and the latest month of carpal tunnel is well below a year earlier. These articles are read for health reasons, not for dictation, so only part of the audience is people looking for a way to type less.

Outside English, speech recognition is a small topic, and the oldest articles have shrunk most

The six non-English articles are small, and the order follows the size of each language's Wikipedia more than its users' interest.

All of them draw fewer readers than the English article on speech recognition. Compared with their first readings, the Japanese, Spanish and French articles have dropped the most. Japanese and Spanish are now a small fraction of where they started. Their latest months are near a year earlier, though: Japanese, German and Portuguese are slightly up, and French, Spanish and Italian slightly down. The steep decline therefore looks like a long-run shift, not something still under way.

Where the data stops: it counts visits, not use

The figures cannot say how many people dictate, how often, or whether they are happy with it. A page view is a visit, not a unique reader, and it includes people who arrived by accident. A news event or a change in a search engine can move the count without any change in real interest. Wikipedia itself is also a different route to information now: answers shown directly in search results and in chatbots take readers who once clicked through, which could explain part of the decline of the general explainers without any change in what people want to know.

Other limits matter when you compare rows:

• Different windows. Each total covers the past year, but articles for Doubao (58,623 page views ³³), Alice AI and the chatbots have only a short history, so their long-run comparison is weaker.
• Names change. An article can be renamed, merged or split, which breaks the comparison with its past. The Gemini jump may partly be that.
• Language size. The Wikipedias differ in size, so the non-English numbers cannot be compared as shares of a language's users.
• Selection. The table covers articles chosen by us, in English and the main languages of our users. It is not a census of everything written about the subject.
• Crawlers are excluded. The counts exclude known automated agents, but unknown ones may remain.

The rows do not disagree with each other so much as answer different questions. The explainers show a stable but fading audience for the topic. The product articles show attention that follows the news. Neither measures adoption.

What this means if you are choosing how to write

Do not read Wikipedia interest as a vote on how to write. Typing is still read about as much as speech recognition, and the typing-injury articles show that many readers have a reason to look for another way to produce text. The shift toward named assistants suggests that people meet speech through products, often built in, and not by looking up the technology.

In practice:

1. Start with what is built in. If you already have an assistant or a built-in dictation tool, test it first. Our overview of speech recognition explains how these tools work and where they fall short.
2. Look at the speed and accuracy studies, not at page views. Popularity of an article tells you what is in the news, not what works for you.
3. Match the tool to the language. The non-English articles are small, and readers there have fewer explainers, so check support for your language directly.
4. If typing causes you pain, the health articles are a starting point, and a clinician is the right person to ask about your own case. Dictation is one option among several.

For the market side, with products and prices, see the market page.