Notion AI: The Add-On That Had to Disappear

Everything factual below is linked and dated. The last two sections are my own analysis — read them as argument, not reporting.

The bet

Notion's original AI bet was a packaging bet: sell AI as a separate SKU, priced per seat, on top of whatever plan you already have.

It's the conservative choice and it's defensible. Inference costs money. Not every user wants AI. A discrete add-on lets you measure willingness-to-pay directly and protects the margin on the base product.

Then they reversed it. The standalone $10 AI add-on was retired in May 2025, and full Notion AI moved into the $20/member Business plan. Legacy add-on subscribers were migrated onto full Business at the same price from December 1, 2025.

The reversal is the interesting part. Companies rarely abandon a working add-on.

Users & JTBD

The job Notion AI serves is not "generate text." Notion is a workspace where a company's context already lives — specs, meeting notes, decisions, tickets. The job is:

"Answer from what we already wrote down, so I don't have to go find it."

This matters enormously for packaging, because it means the value of the AI is a function of how much of your work is already in Notion. A solo user with twelve pages gets a mediocre autocomplete. A 200-person company with four years of documentation gets institutional memory on demand.

The add-on model priced these identically. That was the actual mistake.

The metric that matters

Not AI seats sold. Share of AI queries that resolve against workspace content rather than general knowledge.

If users are asking Notion AI to write a haiku, it's a commodity feature competing with a free chat app. If they're asking "what did we decide about the pricing migration in Q2," it's a product no one else can build, because no one else has the workspace.

That ratio is the leading indicator of whether AI makes the workspace stickier or just adds a cost line. Every packaging decision follows from it.

Teardown: three decisions

1. Bundling converted AI from a purchase into an assumption. An add-on has to be justified every renewal by a buyer who can see the line item. A bundled capability has to be justified once. More importantly, an add-on means some of your colleagues have AI and some don't — which breaks every collaborative workflow the feature was supposed to improve. AI that only half the team can use produces documents the other half can't maintain.

2. The price went up, and the framing changed with it. The same configuration that cost $18/seat on Plus-plus-add-on became $20/seat on Business, with Notion Agent, AI search, AI Meeting Notes, and Enterprise Search included. Two dollars more for a much larger surface. That's not a price increase, it's a repositioning: Business is now the AI tier, and Plus is the tier you leave.

3. Free and Plus got a trial, not a taste. Free and Plus retain only limited AI. This preserves the upgrade pressure that bundling would otherwise destroy — you feel the capability, you hit the wall, the wall has a price on it.

Where it's fragile

My analysis from here.

Bundling costs you the demand signal. Once AI is included, you can no longer measure what anyone would pay for it. Attribution for the Business tier's growth becomes untestable — is it AI, or Enterprise Search, or the seat-management features? Notion traded pricing information for adoption, and pricing information is very hard to get back.

The margin exposure moved from the customer to Notion. Under the add-on, heavy AI users self-identified and paid. Under bundling, the customer with 500 seats and three power users pays the same as the customer where all 500 use Agent daily. That's a cost structure that only works while inference gets cheaper faster than usage grows. It's a bet on a curve, not a plan.

The forced migration is a trust event. Solo users who wanted AI on a cheap plan discovered there was no easy way to keep that configuration. They were made whole on price, which is the right call, but the pattern — the tier you bought no longer exists — is the thing customers remember when a competitor calls.

The competitive frame is now the wrong one. By bundling, Notion is implicitly competing with every workspace that includes AI, which by 2026 is all of them. The differentiated claim — AI that knows your company — got buried in a feature list at exactly the moment it became the only defensible claim.

What I'd ship next

Instrument and publish the grounding ratio. What share of AI answers cite workspace content? Show it to the customer in their own admin panel, per team. It makes the value legible without a line item, it gives the champion an internal renewal argument, and it's a number a general-purpose chat assistant structurally cannot produce.

Meter the expensive surface, bundle the cheap one. Bundling was right for the workspace-grounded features that drive stickiness. Autonomous agent runs — long, multi-step, genuinely costly — should be credit-metered on top. That restores the demand signal on the one surface where usage and cost actually correlate, without reintroducing the have/have-not split that broke collaboration.

Make onboarding a workspace-ingestion problem, not a feature tour. The value is proportional to context, so the activation metric should be pages indexed and queried in week one, not AI features clicked. Every new Business workspace should be pushed toward getting its existing documentation in before it's shown what Agent can do.

Why I wrote this: most PMs shipping AI are not building a new AI product — they're adding AI to a product that already has users and a price. Notion ran that experiment in public, got a defensible answer wrong on the first pass, and corrected in eighteen months. The correction is more instructive than the launch.

Sources

Pricing reflects publicly listed plans as of mid-2026 and varies by billing term and region. Verify current pricing at notion.com before quoting it.