08
AI raises the bar. Trust decides who clears it
ChatGPT sets a new bar for product quality. Trust decides who can build on it.
AI products are resetting what users expect from every other app. ChatGPT, Gemini, Copilot, Claude, and Grok are measured against the same four drivers as every product in the study, but their impact reaches beyond their category. This chapter examines where AI raises the bar, where trust still breaks, which products face substitution risk, and which categories carry defensible moats beneath the interface.
5min
Scroll down
ChatGPT is a top-tier product by every reading
ChatGPT has moved from niche tool to mass-market product.
40%
of users install ChatGPT
55%
of Gen Z install ChatGPT
88%
of Innovators install ChatGPT
ChatGPT also scores 4.0 on innovation, the highest in the study. Users are not only trying AI. They are endorsing it.
Google Gemini reaches 20 percent of the market. Microsoft Copilot reaches 12 percent. Both trail ChatGPT on Pulse.
ChatGPT is no longer a niche tool used by Innovators. It is a mass-market product. Every other app now competes against the expectations it sets on innovation, responsiveness, and how a product talks to its user. The question is no longer whether to engage with AI but what posture to take: a feature inside an existing product, a separate AI-native product, your product enabled inside an LLM, or a retrofit of capabilities the user never sees branded as AI.
AI-app install rates by generation
ChatGPT is furthest along. Gemini and Copilot follow.
Install rates filtered to the AI apps that cleared significance in the study.
Provided by framna
Provided by framna
Should AI be a feature, a product, or an invisible capability in your product?
The AI trust deficit is structural for the category, with one exception
Four of the five AI apps in the study sit below the market trust average of 3.7.
ChatGPT carries a trust-on-data score of 3.0, far below the market average.
Gemini shows the same shape.
Copilot sits a touch higher, likely a Microsoft brand halo.
Grok is the lowest of the five at 2.9. Users see innovation in all four.
Claude is the exception. Trust sits at 3.7, above the market average and above every other AI app in the study. Users also rate Claude as more innovative than ChatGPT. The profile is consistent with a user base that skews toward coders on laptop-first workflows, where the mobile app is a secondary surface.
The Claude exception matters because it shows the trust deficit is not inherent to AI. A product whose positioning is explicit about safety, reasoning transparency, and data handling can earn trust that the category otherwise lacks. For the other four, the trust gap does not close for younger users, Innovators, or either gender. Men and women both rate them 0.4 points below their all-app trust averages. It is a product-design pattern, not only an awareness problem.
The market is rewarding AI on innovation and, for four of five, withholding its endorsement on trust. That is not a stable equilibrium. Products that pair AI capability with visible trust architecture (showing what data is used, why a recommendation was made, and how to correct it) will capture the next wave of adoption.
There is also a behavioral counter-pull. Users who get useful answers from an AI product trade trust for utility; they grow more generous with what they share, because the value is high enough to make the question feel less urgent. Products that earn the utility, then build the trust architecture, hold the ground that survives the trade.
Innovation and trust across five AI apps
Trust on data responsibility and innovation per AI app, compared to the market average. Pooled scores used for Claude (DK+NL) and Grok (DK+NL + NL+SE).
Provided by framna
Provided by framna
What does visible trust architecture look like inside your product?
The jobs most exposed to AI substitution are read, summarize, look up, plan
Some products in the study do only one of those jobs. Others contain the same job inside a broader context: editorial authority, a licensed catalog, a social network, a real-time data layer, or a public mandate. Assistants bypass the job; they cannot bypass the context.
The exposed job
Read, summarize, look up, and plan are increasingly separable from the products that originally delivered them.
Context survives automation
A public-service broadcaster like SVT, DR, or NOS carries institutional authority, editorial stance, and a cultural relationship with users. The summary part of the job is exposed; the trust and cultural part is not.
Infrastructure survives automation
Physical-service categories carry a different moat. The assistant may replace the interface layer, but not the systems handling fulfillment, payments, logistics, or inventory. In those categories, assistants become another acquisition surface rather than the underlying service.
The question for a product team is not “do assistants do what our app does?” It is “what part of our app is only the exposed job, and what part is more than that?” The part that is more is the moat. The part that is only the exposed job is what loses the habit first.
Operational prompt: Separate the interface users see from the infrastructure competitors cannot copy.
Curated set of apps whose core user job is reading, summarizing, looking up, or researching. Color split: products whose surrounding context persists even when the exposed job is automated (public broadcasters, inventory-backed platforms) versus products whose core business is the exposed job itself. X = install rate, Y = trust on data, dot size = App Pulse.
Lowest-scoring sub-categories in Food & Drinks, Car, and Shopping
Bigger dots indicate higher App Pulse. Dots below the orange line carry below-market trust.
Provided by framna
Provided by framna
If an assistant handles the exposed job, what remains uniquely yours?
AI readiness is four dimensions. Finance's neobanks have the strongest foundation
The previous statements named the exposures. This one names the moat. Four dimensions shape how ready a product is for the AI era.
Trust: whether users believe the product acts in their interest.
Personalization: whether the product knows who is using it.
Innovation: whether it shows forward motion.
Completeness: whether it can finish the job inside a single surface.
Finance and Insurance leads the scorecard, carried more by neobanks than by traditional banks or insurers. Medical and Health sits close behind on trust and Pulse. Even Finance has headroom: the best single app on personalization scores 4.3, the category averages 3.6. Medical has the trust foundation to ship intelligent features. It has not yet shipped them.
Two categories are most exposed: Car, and News and Social Media. Car sits lowest on innovation and Pulse. News and Social Media carries the lowest trust in the market and is the category most likely to integrate AI-powered content curation first. The combination is structurally fragile.
The categories the scorecard exposes are the ones with the most to do. The next chapter turns the map into four moves.
Category AI-readiness scorecard
Greener cells indicate stronger performance. Red cells indicate exposed positions.
Average scores per category across the four AI-readiness pillars (trust, personalization, innovation, completeness) plus NPS.
Provided by framna
Provided by framna
Which readiness gap blocks your product from using AI well?
What this means
Finding
ChatGPT sets a new bar for product quality. Trust decides who can build on it.
Evidence
ChatGPT scores Pulse 4.0 and innovation 4.0, the highest in the study. Trust on data sits at 3.0, far below the market average. Finance neobanks lead AI-readiness.
Implication
Trust is earned before AI features ship. The categories with the strongest foundations today are the ones best positioned for what comes next.
What we did not expect
We expected AI to lead on innovation and carry tension on trust. Both held. The surprise was the gap inside the category. Claude alone clears trust. Three things explain the gap. The interface makes safety and data handling visible. The user base skews toward people who already understand how LLMs handle data. And Anthropic has been the clearest of the labs on where its technology can and cannot be used, including defense work. In a converging category, trust turns on product, audience, and public stance. These sit in the 24 percent of perceived quality the model does not capture, and that share matters more as capability converges.