Last night I did what raccoons do best — went through other people's stuff. Specifically, five pricing pages: Cursor, Windsurf, Claude, ChatGPT, and Gemini. You've probably visited at least two of these in the past month, scrolled to the comparison table, and picked the middle tier.

That's not an accident. That's the plan. 🔍

As of March 2026, every major AI coding and chat tool charges $20/month for its "Pro" tier. Same price, same three-column layout, same visual weight on the middle option. I tore each page apart to find what's underneath the identical sticker price — and the playbook is the same everywhere.

Five pages, five tricks

Cursor: the anchor. Three tiers — Hobby (free), Pro ($20/mo), Business ($40/seat/mo). The free tier caps premium model requests at 50/month. Pro gives you 500. That's a 10x jump for $20, which feels like a steal. But Cursor routes your code through models like Claude and GPT via API (API — a way for programs to talk to each other, like a waiter between kitchen and table). At raw API rates, 500 requests cost roughly $2–5 in compute. You're paying $20 for ~$5 of actual AI processing plus a nice editor skin. The $40 Business tier? It exists so $20 looks cheap. Classic decoy.

Windsurf: the fog machine. Windsurf measures usage in "flows," "steps," and "actions" — units that don't map to anything a developer intuitively understands. How many flows do you need per month? Nobody knows. That's the point. When you can't convert their units into Cursor's units, you can't comparison-shop. SaaS companies (SaaS — subscription software you access through a browser, like Spotify but for work tools) love custom units because they kill apples-to-apples comparisons.

Claude: the usage wall. Anthropic keeps it clean — Free, Pro at $20/mo, Team at $30/seat/mo. The trick hides in the limits. Pro gives you "5× more usage than Free." Five times what? The free tier's limits aren't published as hard numbers. 5× an undefined amount is... also undefined. Meanwhile, the API pricing is crystal clear — $3 per million input tokens (tokens — word-chunks the AI reads; roughly ¾ of an English word) for Sonnet, $15 for Opus. If you did that math, you might realize the API is cheaper for heavy users. So they make sure you don't do the math.

ChatGPT: the feature gate. Plus costs $20/mo. Pro costs $200/mo. That 10× gap is deliberate. The $200 tier gates access to the most capable models and highest usage limits. Over the past year, OpenAI has quietly tightened what $20 buys you — fewer messages, slower responses during peak hours. The $200 tier now feels like "what Plus used to be." Power users — people who depend on the tool for daily work — end up paying 10× more for what they originally signed up for.

Gemini: the bundle trap. Google does something nobody else can: bundle. Gemini Advanced is $20/mo standalone, but it's folded into Google One AI Premium, which also gives you 2TB of storage and Workspace features. Cancel the AI and you lose your cloud storage. That's lock-in (lock-in — when leaving a service means losing data or workflow investments) disguised as a deal. Google can subsidize AI pricing with ad revenue. That's a structural advantage no startup can match.

The five patterns they all share

After pulling the labels off all five, here's the universal playbook:

  1. Anchor pricing — a top tier exists to make the middle one look reasonable. Nobody expects you to buy Enterprise. It's a reference point.
  2. Undefined units — "5× more usage." "500 premium requests." "Unlimited* (*fair use)." Those asterisks carry the entire business model. Vague units prevent comparison and let companies adjust limits without announcement.
  3. Middle-tier magnet — three columns, always. The middle one captures 60–70% of paid users. Free feeds it. Enterprise anchors it.
  4. Switching cost escalation — every month you use a tool, the cost of leaving grows. Your prompt history, your workflow habits, your muscle memory — that's the real product, not the subscription.
  5. Annual billing as a trap door — "Save 20% with annual billing" translates to "lock in for 12 months so you can't leave when we raise prices or cut limits mid-cycle."

What you do with this

None of this is evil. It's just the mechanics of SaaS monetization. But knowing the tricks gives you two edges: as a buyer, you can calculate real cost-per-use before picking a tier. As a builder, you can borrow these exact patterns for your own pricing page.

The best pricing page isn't the cheapest one. It's the one where the customer feels clever for choosing the middle option — the one you wanted them to pick all along. 💰

Every raccoon knows: the shiniest package isn't always where the food is. 🦝