Blog · 24 September 2026 · By HeardOf

Which AI assistant do people use? Five published figures, five units, dated

What is ChatGPT's market share, and share of what?

79.4% of AI chatbot referrals to websites worldwide in August 2026, on Statcounter's AI chatbot page as read on 24 September 2026, and a referral is a click from a chatbot to a site that runs Statcounter's tracking code. Its unit is clicks, not people: Statcounter's FAQ says its Global Stats rest on "more than 3 billion page views per month" to "our 1 million+ member sites", and its two 2026 press releases on this dataset, of 2 April and 7 May, both historical, count "AI chatbot referrals to websites", and the May release calls the statistic a "share of AI chatbot referrals to websites". A conversation that ends without a click is not in the denominator, and the same FAQ says of Bing chat that "you are less likely to click on a source website from bing chat than a regular search", so "that needs to be factored in when using our stats for your analysis."

The page's summary list reads, for August 2026, ChatGPT 79.4%, Google Gemini 10.9%, Perplexity 4.31%, Microsoft Copilot 2.79%, Claude 2.57%, DeepSeek 0.02%, identical as served and as rendered in headless Chrome. The bar chart on the same page, and its CSV, read 10.93, 4.34, 2.77 and 2.53 for the middle four, a second-decimal gap we record and do not explain. Every other figure in this post counts something else, which is why the table carries a unit column and no ranking across rows.

Which assistant is second, and does the window change the answer?

Google Gemini in the month of August 2026, at 10.9% against Perplexity's 4.31%; Perplexity over the twelve months from September 2025 to August 2026, at 8.7% against Gemini's 6.38%, from the same Statcounter chart with the period widened. The monthly series behind both runs Gemini from 2.19% in August 2025 to 10.93% in August 2026 (10.9% in the page's summary list) and Perplexity from 8.12% to 4.34% (4.31%); Microsoft Copilot from 5.17% to 2.77%; Claude from 0.89% to 2.53%, through 2.91% in March 2026, which Statcounter's release of 2 April 2026, historical, described as "more than doubling its share in a single month". That release is titled "Google Gemini overtakes Perplexity to become second largest source of AI chatbot referrals to websites" and dates the crossing to March 2026, 8.65% against 7.07%. It did not hold every month: in May 2026 the CSV has Perplexity back above Gemini, 7.67% against 7.04%, and Gemini ahead again from June.

Region changes the order again. Statcounter's United States chart for August 2026, and its CSV, read ChatGPT 72.26%, Google Gemini 11.71%, Microsoft Copilot 7.89%, Claude 4.37%, Perplexity 3.69%, DeepSeek 0.08%; the summary list on the same page reads 72.12%, 11.68%, 7.95%, 4.44%, 3.74% and 0.08%. Either way Copilot is third and Perplexity fifth, where worldwide they are fourth and third. Second place is a question with no answer until a window and a region are attached to it.

How many people use Gemini, AI Mode and AI Overviews?

Over 900 million, over 1 billion and over 2.5 billion monthly active users, in that order, in Sundar Pichai's I/O keynote post of 19 May 2026, historical, and the counts are Google's own. The post reads: "Last year at I/O, the Gemini app had 400 million monthly active users. Today, we've surpassed 900 million, more than doubling in a year", "AI Overviews now has over 2.5 billion monthly active users", and of AI Mode, "in just a year, it's already surpassed 1 billion monthly active users". No region is stated, and the figures count users as Google defines them, which the post does not say, so none of them sits on Statcounter's scale: a person who asks the Gemini app a question and never clicks a link can be inside Google's 900 million and cannot be in Statcounter's 10.9%.

Two of the three are Search features rather than a standalone assistant. The Reuters Institute's survey below sets them aside by footnote: when it talks about AI use for news it "generally" means "the intentional use of standalone AI chatbots for news discovery and not AI Mode or AI Overview answers to search queries".

How many people use ChatGPT?

34% of US adults said they had ever used it, in Pew Research Center's survey of 5,123 adults run from 24 February to 2 March 2025 and published on 25 June 2025, historical; the page as served shows no later date, though its structured data carries a modification date of 20 August 2026. The unit is a share of survey respondents, weighted to the US adult population, who said yes to ever having used one named product; no other assistant is asked about on that page. 58% of adults under 30 had used it, against 10% of those 65 and older; the rest of the age split is in the table.

OpenAI's own count we could not read from OpenAI. Wikipedia's ChatGPT article, read on 24 September 2026, says the product "reached 900 million weekly active users" in February 2026, historical, and cites a Newsweek article of 28 February 2026 for it. Newsweek's article of 28 February 2026, historical, quotes an OpenAI statement giving "more than 900M weekly active users"; OpenAI's own pages answered with a bot challenge, so we have the figure only as Newsweek quotes it, and it is not in the table. Weekly active users would be a further unit in any case: neither a share of clicks, nor a monthly count, nor a share of respondents.

How many people use AI chatbots for news?

10% of respondents across 48 markets used one in the last week, up from 7% a year earlier, in the Reuters Institute Digital News Report 2026, whose AI chapter is dated 16 June 2026, historical. YouGov ran the online questionnaire from the middle of January to the end of February 2026; a chart note in the executive summary gives the total sample as 97,520 and about 2,000 per market, and the methodology page says the results are best read as representative of each market's online population. The unit is people, for one purpose, in one week: the report's executive summary says the figure relates "specifically to the usage of dedicated, standalone AI chatbots such as ChatGPT, Perplexity, and Google Gemini", and the pages we read give no split by brand. 16% of under-35s use one for news weekly, 1% of all respondents call AI their main source of news, the UK is lowest at 4%, South Korea reached 14% and Greece 12%, and "the USA, the UK, France, and Germany reported no increase in AI chatbot use for news over the year".

The chapter relays one figure for use beyond news: the institute's Generative AI and News Report 2025, historical, "found that weekly use of generative AI tools across six markets grew from 18% to 34% between 2024 and 2025". That is a different survey, six markets rather than 48, and we did not open it; it is here as the chapter states it.

Can these figures be ranked against each other?

No, because they are five units from five sources: a share of clicks to websites, a rank of websites by visits, a vendor's count of monthly active users, a share of respondents using any chatbot for news in a week, and a share of respondents who ever used one product. Similarweb's is the rank: its most-visited-websites page for August 2026, last updated 1 September 2026, lists chatgpt.com fifth in the world, gemini.google.com sixteenth and claude.ai thirty-fifth, with google.com first, and the page, as served and as rendered, does not say how the positions are computed. It ranks websites by visits as Similarweb estimates them, so a Gemini session inside Android or Workspace that Google counts is not necessarily in it.

Two figures with the same name still need their unit read. Wikipedia's article relays Statcounter's August figure as ChatGPT holding "an estimated 79% of the market share", and the word referral, which is the whole unit, is gone by then. A ChatGPT market share figure needs that word beside it. The same reading applies to conversion figures: does AI traffic convert better compiles the published ones with the unit each counted.

Did these figures decide which engines HeardOf checks?

No: our engine list is not derived from these figures. It is stated, with its reasons and its limits, in the four engines we ask, as updated 23 September 2026, from our own systems, and nothing in this compilation was used to choose it. What the five sources add for a buyer is the unit column: when a vendor says an assistant has a given share, these are five pages that number can have come from, and each one says what it counted.

How was this read?

Every page was fetched on 24 September 2026 between 22:47 and 23:00 UTC with curl and a Chrome user-agent, and read in full. Statcounter's page was also rendered in Chrome 153 headless at 22:49 UTC, and its summary list read the same six figures; its August bar chart, rendered at 23:36 UTC, draws the values its CSV gives. The twelve-month series came from the chart's own CSV endpoint at 22:49 UTC, the monthly and United States series at 22:51 UTC. Similarweb's ranking page was rendered in headless Chromium at 23:23 UTC by our reviewer, not by the writer, and the render carries the same rows and the same absence of a method statement as the served page. Google's post and Pew's page are dated from their structured data, the Reuters chapter from its byline. Historical means older than 90 days on the reading day: Google's post is 128 days old, the Reuters chapter 100, Pew's page 456 and the end of its fieldwork 571, Statcounter's two releases 175 and 140; Statcounter's page and Similarweb's ranking are current.

What we could not do: OpenAI's own pages answered with a bot challenge, so its weekly-user figure is here only as Newsweek's article of 28 February 2026 quotes OpenAI's statement, and Newsweek answered our first request, at 22:50 UTC, with a refusal and the second, at 23:35 UTC, with the article; Similarweb's page for chatgpt.com answered with a 202 and an empty body, and the writer's own headless renders of Similarweb's pages did not complete, so every Similarweb figure here is from the ranking page as served and as our reviewer rendered it; and no source on this list, on the pages we read, gives use by brand and by person on one denominator. Nothing addressed to AI agents was read on any page; Similarweb's pages carry a "Summarize with AI" button whose prompt is built by script and was not in what we fetched. Everything this post says about the five sources carries a link and the day it was read, and you can open any of it; what it says about our own system comes from our own post and cannot be checked from here.

Does AI traffic convert better? The published conversion figures, with what each one counted