How widespread is GenAI use?

TL;DR: very very roughly, triangulating from multiple datasets suggests around 25% of adults under 70 in the UK are somewhat regularly using GenAI, and around 25% do not use it at all, leaving 50% as very occasional toe-dippers.

When we do AI-related workshops, we try to be sure everyone is on roughly the same page in terms of their understanding the tech. Often it feels like there is some anxiety around people’s personal experience — perhaps they feel like they should be seasoned ChatGPT pros, where in fact they’ve just tried it briefly.

Which made me wonder — how many people are actually using GenAI? If you are running workshops asking participants to think about applying GenAI in their organisation, how many of them have experienced how GenAI first hand?

As ever, definitions are a problem. Some of surveys of AI uptake ask questions about AI including smart thermostats, voice assistant, route planning — these are important too, but I’m interested specifically is user’s understanding of GenAI. A further complication is that many users don’t know what they are interacting with —a customer service chatbot could be a sophisticated LLM or basic decision-tree based chatbot. ONS data (more info below) suggests that about 40% of users believe they can never or hardly ever tell if they are using an AI based system.

ONS data (for the UK)

In the UK, the ONS publishes a monthly survey of ‘public opinion & social trends’. In July 2024, they reported the following rates of use for AI in the last 12 months (under 70s have very roughly similar rates):

Under 70s

• 14% of people have used health or therapy related chatbots (might not actually be AI)
• 25% of people have used customer service chatbots (might not actually be AI)
• 23% of people have used multimodal AI for hobbies or recreation (again there is some scope for confusion here)
• 35% have used chatbots to ‘research or write documents’ (seems quite likely to be referring to an LLM)
• 26% ‘have not knowingly used AI in the last 12 months’

Over 70s
Roughly, half the numbers of those for under 70s. Which, to me, highlights an incredibly high rate of adoption for a new technology.

Stanford Deliberative Democracy Lab

I’ve been interested in deliberative democracy for a long time, and this research (note: sponsored by Facebook) is a very detailed investigation into what users across Brazil, Germany, Spain and the US think about AI after they have had time and resources to reflect and consider through in-person deliberative sessions — group conversations and access to experts. For the purposes of understanding the spread of GenAI, the research is useful because it is carefully demographically balanced for age, gender and eduction, and the very high level of engagement with respondents might ensure a more considered responses.

They find that 60% have used ‘AI’ chatbots before — which is remarkably consistent across countries, and between 15–25% use it daily (seems high). Again — hard to know how much of this is an occasional run in with a customer service bot, and how much is using ChatGPT style services.

How many people are using ChatGPT or Gemini?

Could we work out GenAI uptake through usage stats? I can’t find official figures — they may not be published — but reports seem to suggest that both ChatGPT & Gemini have 100–200 million regular users — which would fit with about 11% of the world with access to the Western Internet using them.

What about other GenAI products?

• Users may experience GenAI in Google Workspaces or MS Copilot without really even noticing — this seems hard to quantify.
• Users may also experience GenAI from customer service chatbots, but again, it’s hard to tell how sophisticated such chatbots are.
• Character.ai is one of the most successful GenAI products where users are exposed directly to generative AI. It appears to have around 20 million users — so not really moving the dial.
• Anthropic has around 4 million users per month.
• It seems likely most direct exposure to LLMs comes from Google & Microsoft

As rough estimate, in total…

In a professional setting with a range of working age participants, and taking a rough average of the figures above, it seems reasonable to assume that around 25% of participants will use LLM-based chatbots at regularly, perhaps 50% occasionally, and 25% hardly at all or never. Which is to say — you should probably assume that quite a lot of participants in a workshop really don’t know much about LLMs.

Caveat: this is very back of the envelope stuff, and I may well have missed products that expose users to GenAI.