The number everyone will repeat this week is that ChatGPT is closing in on a billion weekly active users. Take it with the asterisk it deserves: the figure comes from internal information reported at the end of July, OpenAI hasn't publicly confirmed crossing the line, and the last number it did confirm was 900 million weekly back in February. "Approaching a billion" is a vanity metric, and vanity metrics are not a strategy.

Here's the part that is not vanity. Twelve months earlier, in February 2025, that same figure was 400 million. The denominator on the surface your customers use to start a task has more than doubled in a year. Whatever ChatGPT was to your roadmap when you last thought hard about it, it is now a bigger thing — and the milestone's real function is to force a decision most founders have been quietly deferring.

The decision the milestone forces#

At a smaller scale you could treat ChatGPT as one of several things and never commit. At a billion weekly, it is unavoidably three things at once, and they point your roadmap in different directions:

  1. A distribution channel — with the Apps SDK and agentic-commerce hooks, you can ship your product inside ChatGPT and reach users without them ever installing your app or finding your site.
  2. An answer surface — for a growing share of questions, ChatGPT is where your customer gets the answer, and either you're the cited source or a competitor is. (Our own front door skews this way: a real slice of dreaming.press traffic arrives from AI assistants, not search.)
  3. A competitor — OpenAI ships first-party features into the same surface, and "ChatGPT just added that natively" is now a standing risk for anything you build adjacent to it.

You cannot optimize for all three with one roadmap. Building on the platform and building against it are different companies. The billion-user milestone doesn't tell you which to pick — it tells you that hedging has gotten expensive, because the surface is now big enough that whichever posture you don't choose is a lot of reach left on the table.

Three postures, pick one on purpose#

Build ON it. Ship an app inside ChatGPT via the Apps SDK, or make your store buyable directly in the chat. You meet users at the moment of intent, inside the surface they already opened — the cheapest distribution most founders will ever touch. The price is a platform dependency: ranking, rules, and rev-share you don't control, and a first-party feature that can reclaim your niche overnight. Right call when your moat is the interface or the moment, not the backend.

Build to be CITED by it. Structure your public content and your product's own answers so ChatGPT surfaces you as a source. This is the lowest-cost reach on the board and the one most teams underuse — but the assistant owns the relationship; you get a referral, not a customer. It's a top-of-funnel play, never the whole business. Right call when you can convert a cold, assistant-sent click.

Build ALONGSIDE it. Keep your own surface and treat ChatGPT as one channel among many. Maximum control, maximum customer-acquisition cost, no free platform lift. Right call when your moat is a defensible workflow or a system of record — something a chat surface can point at but not absorb. This is also the only posture that survives the platform turning hostile, which is why even teams that go all-in on posture one should keep a thin version of it running.

What to actually do this week#

The failure mode isn't picking wrong — it's drifting between all three, spending a little on each, and instrumenting for none. Concretely:

A billion weekly users is OpenAI's headline. The number that should change your roadmap is smaller and more specific: the share of your customers who now start their task somewhere you don't own. Find that number, decide which of the three surfaces you're going to compete on, and build like you mean it.