In early 2025, ChatGPT began displaying product cards with images, prices, reviews, and buy links when it detected shopping intent in a prompt. This ChatGPT Shopping feature is one of the most talked-about new surfaces in product discovery.
We tracked ~2 million unique prompts repeatedly over several months from September 2025 through January 2026. We aimed to measure not just whether shopping triggers, but how consistently it triggers over time.
This is the first in a series. Here, we cover what we found. A follow-up analysis will break down what brands should actually do about it.
Two findings define the landscape:
- The distribution is all-or-nothing. 79% of prompts never trigger shopping across any run over 9 months. Just ~6% trigger reliably.
- When a prompt does trigger, it tends to stick. If a prompt triggered shopping on a given run, there's an ~83% chance it triggers again the next day. Prompts lock into a state and stay there.
How often does shopping actually trigger?
Across millions of prompt submissions, the shopping feature activated on ~9% of them.
But 8.82% is an average, and averages hide the real story. When we segmented by prompt type, the gap was immediate.
A note on terminology: throughout this piece, we distinguish between two prompt types. Open-ended prompts describe a need without naming a brand - things like "best business laptops with strong battery life" or "lightweight running shoes for flat feet." Brand-direct prompts name a specific brand — "Nike running shoes" or "Dell XPS 15 review."
Prompt Type
Open-ended prompts, "best business laptops with strong battery life," "lightweight running shoes for flat feet," "which comfort socks don't stretch out after a few washes?" trigger shopping at 12.1%. That's roughly 4× the rate of brand-direct queries.
And it's not that brand-direct prompts lack purchase intent. When a brand-direct prompt describes a product need, shopping will still trigger. But when brand queries reference services, they basically never trigger shopping, and they drag the brand-direct average down to 3.1%.
The implication for brands: ChatGPT Shopping is a discovery surface, not a search engine. You don't win by being the brand someone searches for. You win by showing up in the consideration set when users are still in discovery mode: describing what they need but haven't decided what to buy yet.
Now the harder question: can you count on it?
Everything above is a cross-sectional snapshot. Millions of prompts, one point in time. But brands build strategies across quarters and years. So we ran ~2 million prompts repeatedly over several months to answer two questions: does shopping show up for your prompts at all, and if it does, does it keep showing up?
The answers are very different.
Most prompts never enter the game.
The consistency distribution across ~2 million prompts is striking.
79% of prompts never triggered shopping across any run over the entire 9-month study. At the other extreme, only 0.7% triggered every single time. The "reliable" tier, prompts that trigger at 80%+ consistency, captures about 6% of the total.
There's a massive cluster at zero and a small cluster at the top, with almost nothing in between.
So what do the prompts in that top 6% have in common? From what we can see directionally, they tend to describe a specific, shippable product need with enough detail that ChatGPT can match it to a real SKU. We'll be digging deeper into the anatomy of high-trigger prompts in our follow-up analysis.
But the prompts that do trigger are surprisingly sticky.
Here's where the data gets interesting. If a prompt triggered shopping on a given run, there's an ~83% chance it triggers again the next day. The reverse is equally sticky: if a prompt didn't trigger, there's only a ~1.7% chance it starts triggering the next day.
Prompts lock into a state and stay there. That's actually the mechanism behind the all-or-nothing distribution. The 79% at "never" and the ~6% at "always/high" aren't random outcomes. They're self-reinforcing defaults. Once a prompt lands on one side, the system keeps it there.
What this means for brands
Classify before you optimize.
Pull the prompts your customers would actually type. Test them repeatedly over weeks. ~80% of prompts never trigger shopping.
Your branded queries matter less than you think → unless they describe a product.
Brand-direct prompts trigger at 3.1% overall, versus 12.1% for open-ended. Brand queries about tangible, purchasable products still fire shopping consistently. Brand queries about services, evaluations, or anything you can't ship in a box → essentially never.
Short-term persistence is real, but it decays.
If your prompts trigger today, there's an ~83% chance they trigger tomorrow. Model updates may reset your visibility overnight. The brands that win here detect changes fast.
Methodology
How we measured this
- We analyzed 260 million prompt-level results across 13,000 product categories from September 2025 through January 2026.
- For the consistency analysis, we tracked ~2 million unique prompts with at least 10 runs each over the same period, measuring how often each prompt triggered shopping across repeated runs.
