Google AI Pro Just Changed Your Usage Limits: What Actually Happened (And What To Do)

Open your inbox. If you pay Google $20 a month for AI Pro, there is probably an email sitting there from the Google One Team with a subject line that sounds boring on purpose: “Updates to your Google AI Pro plan.”

Read it twice. The thing it is telling you, in the calmest possible language, is that the rules of your subscription changed on May 20, 2026. Not the price. The way you are allowed to use it.

Here is the thing. This is not a UI tweak, and it is not just a Gemini app story. It is the moment flat-rate consumer AI quietly stopped being flat-rate. Let’s break down what actually changed, what did not, and the four things worth doing this week before your usage runs out at the worst possible time.

What Google Actually Changed

For most of the last year, your AI Pro plan worked on a simple promise: a fixed number of prompts per day. Up to 100 Gemini Pro prompts daily, regardless of whether you asked it to fix a typo or write a 40-page research report. Simple to understand. Easy to plan around.1

That model is gone. Three things replaced it.

1. Compute-based usage limits in the Gemini app. Google now meters you on how much computing your requests actually burn, not how many you send. The limit factors in the complexity of your prompt, the features you use, and the length of your chat. A quick question costs you a little. A long, tool-heavy, multi-step conversation costs you a lot more.2

2. A rolling 5-hour window, capped weekly. Your usage allowance refreshes every five hours, and that cycle keeps running until you hit a separate weekly ceiling. As an AI Pro subscriber, your limit sits at 4x higher than what free users get.2, 3

3. AI credits are now a separate purchase. The same compute-based model is rolling out to other Google products, starting with Flow (the AI filmmaking tool) and Antigravity (the agentic coding platform). Here is the part that stung subscribers: the 1,000 AI credits that used to come bundled into your base plan every month are no longer included. If you want more, you buy credits.3, 4

Google’s framing is that the new limit structure should let you “maintain the same experience you are used to.” A lot of users on Reddit and X did not read it that way. Several reported a single heavy prompt eating a visible chunk of their quota, and the change landed days after Google showed off a wall of flashy new Gemini features at I/O 2026. The timing felt, to put it gently, awkward.4, 5

What Did Not Change

Before the panic sets in, the email is also clear about what you keep. This matters, because the headlines have mostly skipped it.

Model access. You still get Google’s latest Gemini Flash and Pro models, plus Thinking capabilities for the harder problems. No model has been pulled out from under you.6
Features. Deep Research, video generation, and the rest of the AI Pro feature set are still there. The capability list is intact.6
Storage. You keep your 5 TB of shareable storage across Google Photos, Drive, and Gmail. That alone is worth a chunk of the subscription for a lot of people.6

So this is not a downgrade in what the plan can do. It is a change in how much of it you can do before you hit a wall. Whether that is better or worse depends entirely on how you use it, which is exactly why the next section matters.

Compute-Based Limits, Explained Like A Human

The old system was a parking meter. One coin, one hour, no matter what car you drove.

The new system is a taxi meter. A short ride across town is cheap. A long ride with three stops and the engine idling while you shop is expensive. Same meter, very different bills.

What this really means: the people who lose out are heavy users running long, complex, agentic sessions, the kind where Gemini spins up sub-tasks and chews through tens of thousands of tokens from one instruction. The people who barely notice are everyone asking normal questions, drafting emails, summarising documents, and running the occasional research report.7

And Google is not alone here. GitHub overhauled Copilot onto a token-based AI credits model less than a month before this. The whole industry is hitting the same wall: agentic AI features are powerful enough that flat-rate pricing simply does not survive contact with how people actually use them.7

Four Things To Do This Week

1. Find Your Usage Dashboard (2 Minutes)

You cannot manage what you cannot see. Google added a live usage tracker, and most people do not know it exists.

Open the Gemini app on web or mobile.
Go to Settings.
Tap Usage limits (also shown as “Usage limit” in some versions).
Note where you stand right now, then check it again after a heavy session.

This single screen tells you whether the change affects you at all. For a lot of light and medium users, it will turn out to be a non-event.3

2. Run A “What Eats My Quota” Test

Spend one normal working day watching the meter. After each session, glance at the usage screen and notice what moved the needle. You are looking for the expensive habits:

Very long, never-reset chat threads (chat length is now a cost).
Heavy feature use like video generation or deep, multi-step research.
Asking the most powerful model to do work a faster model handles fine.

Within a day you will know your personal burn rate. That is worth more than any blog estimate, because compute cost depends on your behaviour, not the average user’s.2

3. Switch To Cheaper Habits Where It Costs You Nothing

Once you know what is expensive, the fixes are easy and mostly painless:

Start fresh chats for new topics. A 200-message thread carries its whole history into every new reply, and you pay for that length. Open a new chat when the subject changes.
Match the model to the job. Use Flash for everyday speed and save the heaviest Thinking model for problems that genuinely need it.
Batch your research. One well-scoped Deep Research request beats five vague ones that each spin up the full machinery.
Be specific. Tighter prompts get the answer in fewer turns, and fewer turns means less compute.

4. Decide If You Actually Need AI Credits

The bundled 1,000 credits are gone, so the honest question is whether you ever used them. Most Gemini app users never touched Flow or Antigravity. If that is you, the credit removal changes nothing.

If you do create AI video or run agentic coding workflows, you can now top up by buying AI credits directly. Check the Google Help Center for current credit pricing in your region before you buy, since rates and availability vary by country.3

The Bigger Picture: AI Pro vs The New Ultra Tiers

This change did not arrive alone. At I/O 2026, Google reshuffled the whole subscription ladder. A new $100/month AI Ultra tier appeared, aimed at developers and heavy creators, and the top Ultra plan dropped from $250 to $200.4, 5

Here is what that signals. Google is splitting its audience. AI Pro at $20 is being positioned for normal, everyday use. The serious compute, the 5x limits, the priority access to agentic tools, all of that is being nudged toward the Ultra tiers. The compute meter on AI Pro is part of that sorting.8

For most people, that is fine. You are probably not the user Google is trying to push upmarket. But if you have started living inside agentic AI for hours a day, this is your early warning that $20 may not be your plan for much longer.

The Honest Limitations

A few things worth keeping in perspective before you draw conclusions:

Google has not published the exact number of tokens or credits an AI Pro prompt consumes. They tell you it is 4x the free tier and leave the rest opaque, which is precisely the predictability complaint subscribers raised.3, 4
The limits can change without notice, including due to capacity constraints. Whatever you measure this week is a snapshot, not a guarantee.9
Reports of credits “burning fast” are real but uneven. Some heavy users hit walls quickly, while plenty of others say they have noticed nothing at all. Your mileage genuinely will vary.5

None of this changes the direction of travel. It just means the next few months of consumer AI pricing will be messier than any single announcement makes them sound.

The Bottom Line

Your Google AI Pro plan still has the models, the features, and the 5 TB of storage you signed up for. What changed is the meter underneath it. You are now billed on compute, not on prompt count, the bundled credits are gone, and the same system is spreading across Google’s AI products.

If you are a normal user, open the usage screen, confirm you are nowhere near the ceiling, and get on with your life. If you are a heavy user, this is the nudge to either tighten your habits or start eyeing the Ultra tiers. Either way, the era of “unlimited-feeling” $20 AI is quietly over, and Google just sent you the memo.

If you pay for any AI subscription, send this to the person you know who runs 50 prompts a day and assumes the tap never runs dry. They need to read it before their meter does the talking.

Sources and further reading

PCWorld, Google just made big changes to Gemini usage limits.
Google, Everything new in our Google AI subscriptions, fresh from I/O 2026.
Android Authority, Google quietly nerfed its AI Pro plan, and here’s what you get now.
Business Standard, Google reshuffles Gemini AI plans with new Ultra tier, revises usage limits.
Android Central, The Google AI Pro plan just got a quiet downgrade, here is the new deal.
Google One, Google AI Pro & Ultra plan details.
PCWorld, on the shift from request-based to compute-based limits.
Google, on the new $100 AI Ultra tier and 5x limits.
Gemini Apps Help, Gemini Apps limits & upgrades for Google AI subscribers.