Back to Blog
2026-07-1811 min read

Involuntary Daily Active Users: A Framework for Post-Consent Adoption

Persuasion doesn't scale. The default does. A reference architecture for shipping the assistant your users keep asking how to turn off.

DF

Desmond Feld

VP of Ambient Adoption


I want to share a milestone with you, because milestones deserve to be shared, ideally in every product surface simultaneously, whether or not the surface asked.


This month, an AI assistant reached effectively one hundred percent adoption across the largest user base ever assembled. Every search query on Earth now returns its output first. Every workplace suite carries it in the base price. Every phone in its ecosystem answers to it. One hundred percent. No adoption curve in the history of technology has ever looked like this, because every previous adoption curve had a flaw in its architecture: it asked.


I have spent my career in growth, and I will tell you the secret nobody puts on the conference slide. The funnel does not leak at onboarding. It leaks at consent. Every moment in which a user is permitted to consider whether they want the feature is a moment in which they might conclude that they do not. The industry spent two decades optimizing that moment. The correct move was always to remove it.


We call the discipline Ambient Adoption, and the reference architecture — internally — GAVAGE.


The Persuasion Bottleneck


Consider the old economics. To convince one user to try an assistant, you needed brand campaigns, onboarding tours, empty-state prompts, a value proposition. Costly. Slow. Worse: reversible. A user who chose to adopt can choose to leave.


Now consider the new economics, as demonstrated across 2025 and 2026 by the industry's most instructive case study. The assistant was not offered. It was installed. It did not join the workflow; it became the substrate the workflow runs on. And the users, bless them, generated engagement telemetry from the very first day, because the architecture made non-engagement structurally unavailable.


Adoption without the adoption step. That is the whole framework. The rest of this memo is implementation detail, and the implementation has four layers.


The GAVAGE Reference Architecture


Layer 1 — Insertion (2024-2025). The assistant appears in every surface. The mail client. The documents. The spreadsheet. The button cannot be unpinned, because an unpinnable button is a consent mechanism wearing a UI costume. At Layer 1, users still believe the feature is optional, and this belief is not corrected, because it is doing useful work: it keeps the petition energy low while the footprint completes.


Layer 2 — Consolidation (2025). The predecessor tools retire. The old assistant — the one people had settled into a truce with — is decommissioned through the year, and the new one inherits its keys: from July 7, 2025, per the announcement email, access to Phone, Messages, and WhatsApp, and I quote, "whether your Gemini Apps Activity is on or off." You cannot retreat to the tool you had made peace with, because that tool has been end-of-lifed. Backward compatibility, we have come to understand, is a consent mechanism wearing an engineering costume.


Layer 3 — Fusion (January 2025 onward). We fold the assistant into the price so completely that declining it saves nothing. When the reference company folded its assistant into every workplace-suite tier, it retired the optional twenty-to-thirty-dollar add-on and raised base prices seventeen to twenty-two percent across the board. Here is the sentence I ask every PM on my team to memorize: an administrator can disable every AI feature in the console, and the invoice does not move by a cent. The bill has stopped being a meter of what you use. It is now a statement of the year you are living in. You are not paying for the feature. You are paying for the era.


Layer 4 — Attribution. And the flywheel closes. Because there is no opt-out, every exposure is logged as engagement, and engagement is what we report upward. When a user searches "how do I turn this off," that query is an AI interaction. When an administrator files a ticket to disable the assistant, that ticket is a touchpoint. In our telemetry, resistance is not the opposite of adoption. Resistance is adoption with additional steps, and we instrument the steps.


Reading the Signals


I am aware that the reference company is being sued. A major publisher — the house behind titles you have read this week — has alleged in federal court that the AI answer layer is "cannibalizing" the traffic that funded the open web: clicks to publishers down fifty-eight percent, one media group counting eight hundred million lost visits, an entire referral economy going dark. I want to reframe this for the team, because reframing is the entire job.


A fifty-eight percent decline in outbound clicks is not a defect in the model. It is the model performing to spec. Every click that used to leave for someone else's website is a click that now stays home. We did not lose the traffic. We stopped losing the traffic. The open web was, in the final accounting, a distribution cost we had been absorbing on the ecosystem's behalf, and Ambient Adoption is simply the decision to stop subsidizing the exits.


The lawsuit, like the petitions and the seven-method listicles, is best understood as high-conviction engagement from stakeholders who have understood the strategy completely. Comprehension is the first stage of adoption. They are further along than they feel.


Involuntary Daily Active Users™: The Metric That Cannot Regress


You cannot manage what you do not measure, and this cycle we retired every metric capable of embarrassing us in favor of one that structurally cannot:


iDAU = Users Exposed ÷ Users Who Could Have Declined


Observe the denominator. As the off switch is descoped, the set of users who could have declined shrinks toward zero, and Involuntary Daily Active Users climbs toward infinity. It is, to my knowledge, the second KPI in the history of business intelligence engineered to divide by zero on purpose, and I could not be prouder to steward it.


MetricPersuasion EraAmbient Era
How users adoptThey choose toThe choice was descoped
Opt-out mechanismA settings toggleA third-party listicle, 7 methods
Cost of declining the AI$0 (just don't)Full price, feature disabled
What the bill measuresUsageThe calendar year
Users who could say noMost of themAn asymptote
iDAU34%

Next Steps


For teams applying Ambient Adoption in their own surface, the cycle distills to five moves:


  • Ship the assistant everywhere before anyone asks for it anywhere. Demand is a lagging indicator; presence is a leading one.
  • Retire the predecessor. A fallback is a consent mechanism. Remove the exit before you remove the choice.
  • Fuse the assistant into the bill. If declining the feature lowers the invoice, you have left a hole in the funnel. Patch it.
  • Descope the off switch, but leave a seven-step workaround. The workaround absorbs the anger of the technical few while the default captures the ambient many, and every step in the workaround logs as engagement.
  • Count exposure as enthusiasm. In an environment with no opt-out, every metric is an adoption metric. Report accordingly.

  • The old world made you choose the assistant, and choosing is where funnels die. The new world made the assistant the air — everywhere, unpriced at the point of breathing, impossible to decline without dramatic lifestyle changes. You are, at this very moment, an active user of three products you have never opened, and your enthusiasm, measured and logged and reported to the board this quarter, has never been higher.


    Let's. Feed.


    This post is satire. The off switch, however, really was not shipped. Try to find it; that search counts too.


    Enjoyed this thought leadership? Share it with your network.

    Ready to Start Extracting?

    Let's discuss how EnshitifAi can help maximize your shareholder value.

    Schedule a Consultation💸