If you pay $20 a month for ChatGPT, Claude or Gemini, you probably assume everyone else does too. It feels like the default. It is not.
Card data from a panel of US consumers shows that in August 2026 only 4.5% of eligible consumers had an active paid personal subscription to ChatGPT, Gemini or Claude. That is up from 2.1% a year earlier, so it has more than doubled, but it still means more than nineteen people in twenty pay those three companies nothing at all.
Among the people who do pay, the median spend is $25 a month. And at the very top, the top 1% of AI payers spend an average of $903 a month on consumer cards.
Somewhere between $0 and $903 is you. Here is the real spread, what the big spenders are buying, and a simple test for whether you are paying too much, too little or exactly the right amount.
The numbers on what people pay
| Finding | Figure | Source |
|---|---|---|
| US consumers with a paid personal subscription to ChatGPT, Gemini or Claude | 4.5% in August 2026, up from 2.1% | YipitData card panel, via a16z |
| Median monthly spend among AI payers | $25 | same |
| Average monthly spend of the top 1% of payers | $903 | same |
| Payers of one AI product who also pay for a second | only 13% | same |
| Typical payer, self-reported | $20 to $49 a month | Menlo Ventures, 5,067 US adults, July 2026 |
| Payers spending $100+ a month | 14%, who make up 60% of all consumer AI spending | same |
| Households paying for AI services | about 3%, median $20 a month | Bank of America Institute, as reported |
Read the middle of that table and the picture is clear. For most people who pay at all, AI is one $20 subscription, and it stays one. The second, third and fourth tools that fill tech newsletters are bought by a small group.
Why one survey says 55% pay and another says 4.5%
There is a number that looks like it contradicts all of this, and it is worth understanding because it says something about your own spending.
Menlo Ventures' national survey found that 55% of US AI users pay for at least one AI product, about 90 million people. The card data says 4.5% of consumers pay ChatGPT, Gemini or Claude directly. Both are measured carefully. They are just counting different things.
| The 55% | The 4.5% | |
| What counts | any paid AI product | only personal subscriptions to three assistants |
| Who paid | you, a family member, or your employer | your own card |
| AI inside other subscriptions | counted | not counted |
| How it was measured | asked people | read their receipts |
The gap is mostly AI you pay for without noticing, and AI someone else pays for. Menlo found 48% of payers cover the cost entirely themselves, 34% have a family member or friend paying, and roughly 20% have an employer or school picking up the bill. Add the AI features now bundled into phone plans, office suites, photo editors and design tools, and a lot of people are paying for AI who would never say "I subscribe to ChatGPT".
The practical point: before you add another AI subscription, check what you already have. There is a decent chance the tool you are about to buy is sitting inside something you already pay for.
What the $903-a-month people are buying
The top spenders are the most interesting part of the data, because they are not who you would guess.
The a16z analysis found the top 1% over-index on automation and product-building tools such as n8n, fal, Manus and Nous Research's Hermes Agent, and on creative production tools such as Higgsfield, Figma and HeyGen. Their spending rose 80% over 18 months while the median barely moved. The top 10% together account for roughly half of all observed consumer AI spending.
That list tells you what $903 a month is. These are not people with expensive chat habits. They are people building automations for clients, producing video and design commercially, and running agents that do real work. For them, $903 is not a subscription, it is a cost of goods, the way a contractor pays for tools or a photographer for lenses. a16z itself notes that some of this personal-card spending supports professional work.
It matches the earnings side of the picture closely. When I looked at what 79 million Americans actually earn with AI, half of active side hustlers made under $500 a month, and the ones making more were the ones selling a defined service. The heavy spenders and the heavy earners are, to a large extent, the same people.
The Claude detail
One finding in the Menlo data is worth noting if you are choosing which assistant to pay for. 82% of Claude users pay, and Claude users are about twice as likely as ChatGPT or Gemini users to spend $100 or more a month. Agent users are even more committed: 92% of them pay.
That lines up with an earnings figure from a separate survey, where 45% of Claude users doing AI side work cleared $500 a month against 30% of ChatGPT users. As with that finding, the likeliest explanation is the type of user rather than the tool: people who pick a non-default assistant and pay for higher tiers tend to be doing heavier, more commercial work. But it does mean that among people using AI to make money, paying for more capability is normal rather than extravagant. If you are working out which to pick, ChatGPT vs Claude vs Gemini compares them task by task.
Should you pay, and how much?
Here is a test you can run in two minutes, based on how the people in these surveys actually behave.
Test one: would losing access hurt? Menlo's own conclusion was that consumers become willing to pay once AI is useful enough that losing it would hurt. Payers use AI daily at nearly twice the rate of non-payers, 50% against 26%. If you open it a few times a month, the free tier is almost certainly enough. If you open it every working day, it is probably worth paying for.
Test two: one hour. A $20 subscription pays for itself if it saves you one hour a month and your time is worth $20 an hour. If you use it for work, it almost certainly clears that bar in the first week.
Test three: is it earning? The jump from $20 to $100 or $200 a month only makes sense when the extra capacity produces money, such as client work, content you sell, or automations you charge for. If you cannot point to income it produces, the higher tier is a luxury, and a fine one if you can afford it, but not an investment.
| Your situation | Sensible spend |
| Occasional use, a few times a month | $0, use the free tiers |
| Daily use for personal or office tasks | $20 a month, one tool |
| Daily use for work you are paid for | $20 to $40, one main tool plus one specialist at most |
| Paid client work, agents, content you sell | $100 to $200 on the tool that does the earning |
| Running automations or production for clients | several hundred, treated as a business cost |
What the $20 and $200 plans actually include is covered separately. The short version is that the expensive tiers mostly buy more usage and longer agent runs, not a smarter answer to an ordinary question.
The waste to look for
Two assistants doing the same job. Only 13% of payers pay for a second AI tool, and many of those are paying twice for overlapping chat assistants. Pick the one you open more and drop the other.
A top tier you do not fill. If you are on a $100 or $200 plan and rarely hit the limits of the $20 one, you are paying for capacity you do not use. Drop a tier for a month and see if you notice.
AI you already have. Check your phone plan, office suite and design apps before buying. The 55% versus 4.5% gap exists partly because so much AI now arrives bundled.
Prices that moved. API and plan prices have been falling. When both leading labs cut prices on the same day in September, the new price map changed what the sensible choice is. Re-check once a quarter.
What this means for you
If you pay nothing, you are in the large majority and you are not missing much if your use is occasional. Try a paid month only when you hit a free-tier limit in the middle of something that matters.
If you pay $20, you are the median payer. That is the right place for most people who use AI daily. The question is not whether to pay more, but whether you are using what you pay for.
If you are thinking of paying $100 or more, do it the way the top spenders do: because a specific piece of paid work needs it. The routes where higher spend turns into income are in the nine ways people make money with AI, and automation, the top category for big spenders, has its own guide in making money with n8n.
The honest take
The striking thing about these numbers is how far the real picture is from the one AI marketing paints. Most people pay nothing. Most payers pay $20 and never add a second tool. And the eye-watering spending at the top is concentrated in a small group who mostly treat it as a business expense because it is earning them money.
That is actually reassuring for an ordinary user. You do not need a stack of subscriptions to get real value from AI, and nobody serious is spending $900 a month on curiosity. The people at the top pay that much because their AI does billable work. If yours does not, $20 or nothing is not falling behind. It is exactly where the data says most people sensibly are.
So look at your bank statement this week: what are you paying for AI, and can you name what each line earned or saved you last month?
Frequently asked questions
How much does the average person spend on AI?
Most people spend nothing. Among those who pay, the median is about $25 a month based on US card data, and the typical payer reports spending $20 to $49 a month. About 14% of payers spend $100 or more.
How many people pay for ChatGPT, Claude or Gemini?
About 4.5% of eligible US consumers had a paid personal subscription to at least one of the three in August 2026, up from 2.1% a year earlier. Broader surveys that count any AI product, including ones paid by family or employers, put the share of AI users paying at around 55%.
Is a $20 AI subscription worth it?
It is worth it if you use AI daily or if it saves you more than an hour a month of time worth $20 an hour. If you use it occasionally, the free tiers are usually enough.
What do people who spend hundreds a month on AI buy?
The top 1% of AI payers spend about $903 a month on average, mostly on automation and product-building tools such as n8n, fal and Manus, and creative tools such as Higgsfield, Figma and HeyGen. Much of that supports paid professional work.
Do people pay for more than one AI tool?
Rarely. Only 13% of people who pay for one AI product also pay for another.
Sources: The Neuron on a16z and YipitData's consumer AI spending data; Menlo Ventures: The State of Consumer AI 2026; Moneywise on Bank of America Institute's AI subscription data.
