A product can be enormous and shallow at the same time.
Adoption vs Tourism: Are Your Users Learning or Just Visiting?
A big weekly active number can hide the truth that most people tried your product and never came back.
ChatGPT has roughly 900 million weekly users, and for about 80% of them that adds up to fewer than 1,000 prompts in a year. One of the most widely distributed products ever built, and most of the people counted in its headline number barely use it.
That gap has a name once you look at it directly. Most of those 900 million did not adopt the product. They visited it. The distinction between adoption and tourism is the single most useful thing a beginner PM can learn to read in a usage chart, because nearly every dashboard you will ever inherit is designed to make tourism look like adoption.
What the Big Number Actually Counts
Weekly active users means a person opened the product at least once in seven days. That is the entire definition. It does not mean they returned. It does not mean they relied on it. It does not mean they would notice if it disappeared.
Benedict Evans published the underlying breakdown this month, drawing on OpenAI's own 2025 usage data. The top 5% of ChatGPT users sent between 1,000 and 5,000-plus messages across the year. The top 20% sent under 500. In the United States, daily chatbot use sits at about 15% of adults, even after roughly three years of mainstream availability.
Hold both facts at once. A product can be enormous and shallow at the same time. The reach is real. The habit, for most of the people counted, is not.
Why the Distinction Matters More Than the Total
The products that compound over years are not the ones people try. They are the ones people cannot get through a day without. Email. Maps. The calendar that wakes you up. Each of those earns a slot in someone's daily routine, and that slot is what makes the product durable.
A weekly active number tells you how many people walked through the door. It tells you nothing about how many came back the next morning. For 80% of ChatGPT users, the product sits in the same category as a gym membership bought in February: paid for, technically active, mostly unused.
A weekly active count measures who walked in. Only the daily number tells you who lives there.
This is why a beginner should be suspicious of any metric that goes up and to the right without a frequency breakdown beside it. Total users can grow while real usage stays flat, because a single distribution push (a launch, a press cycle, a default placement) floods the top of the funnel with people who try once and leave. This is the same trap that catches AI features shipped to inflate a launch metric: the pilot that everyone can build is not the product anyone keeps using.
How to Read the Chart Without Being Fooled
There is no framework here to memorize. There is a habit of asking the same three questions whenever someone shows you a usage number.
First, frequency. How often does an active user come back, daily, weekly, once and never again? A product with 900 million weekly users and 15% of its market using it daily is two different products wearing one number. Ask which one you are actually looking at.
Second, depth within the active group. The ChatGPT data is a clean example: the average prompt count is meaningless because the distribution is so skewed. A small cohort does almost everything, and a large cohort does almost nothing. An average usage figure hides the exact thing you need to see: whether your real users are a thin layer at the top or a broad base across the middle.
Third, what happens after the first visit. Day-30 retention tells you more about a product's future than any acquisition number. This is where a funnel earns its keep, not as a scorecard but as a way to locate which stage is actually broken, the way AARRR works best as a diagnostic rather than a metric. If people arrive in large volumes and most are gone a month later, you do not have an adoption story. You have a distribution win and a habit problem, and those require completely different work to fix.
What This Means for the Product You Are Building
Be honest about which number you are celebrating. It is tempting to report the weekly active figure because it is the largest and the most flattering. The discipline is to report the daily figure next to it, and to name the gap out loud before someone in the room asks the question for you.
When the gap is large, resist the reflex to treat it as a marketing problem. More people at the top of a leaky funnel does not fix retention; it just spends money to refill a bucket with a hole in it. The gap between weekly tourism and daily habit is almost always a product question: which moment in someone's day this is supposed to own, and whether it has actually earned that moment. Shipping the feature was never the hard part. As with most AI work, the bottleneck moved rather than disappeared, and here it moved to the question of whether anyone keeps the thing.
The honest framing is uncomfortable, which is why teams avoid it. Fifteen percent of US adults using something daily is tens of millions of real people, and that scale is genuine. But the same data says the large majority are not there, three years in, and no amount of weekly active growth changes what that means: the distribution exists, and the habit, for most people, does not.
That is the cost of reading the headline number and stopping. You feel like you are winning while the part that determines whether the product survives, whether anyone comes back tomorrow, goes unmeasured. The number that flatters you and the number that predicts your future are rarely the same one. Beginners learn to find the second.
Related Articles
Your AI feature has 2 million weekly users but only 4% open it on any given day. Do you celebrate the reach, fix the habit, or kill the feature?
Make the call in Reps and see whether your reasoning survives the daily-active number.
Make the call