ChatGPT doesn't tell you how full your context window is. You're writing, the model is following, and then somewhere around message 40 of a long coding session, the quality drops and you wonder whether it's forgotten the system prompt or the early architecture decisions. It has. You hit the context limit.
This guide covers how to track your ChatGPT context window in real time, and why a browser extension is the right tool for the job.
Why ChatGPT doesn't show token usage
OpenAI made a deliberate UI decision to hide token counts from the interface. The reasoning is probably that it's intimidating and unfamiliar to non-technical users. For developers, this is a problem.
The context window isn't infinite. GPT-4o has a 128,000-token context window. That sounds large until you realize a moderate coding session — system prompt, some back-and-forth, a few code pastes — can consume 30,000 to 50,000 tokens in under an hour.
How token counting works in a Chrome extension
A browser extension can count tokens in two ways:
Client-side estimation — reading the text of the conversation from the DOM, counting characters, and estimating tokens at a ~4 characters-per-token ratio. This is an approximation, but it's fast and requires no API access.
API-based counting — making a call to OpenAI's tokenizer with the exact text. This is precise but requires an API key and adds latency.
TokenPulse uses client-side estimation with a ±8% accuracy note. For practical purposes — knowing whether you have 10% or 70% of context remaining — this is more than sufficient. The exact token count doesn't matter; knowing you're at 62% and closing in on the limit does.
What a good token counter should show you
A slim bar above the input box is the right UX pattern. You don't want a popup you have to click, you don't want a separate tab, you want a persistent indicator that's always in your peripheral vision as you type.
The bar should show:
- →What percentage of the context window you've used
- →Ideally, a rough token count alongside the percentage
- →A color signal when you're getting close (yellow at 75%, red at 90%)
TokenPulse injects exactly this bar above ChatGPT's input box — visible without any interaction, updating after every message.
Setting up token tracking on ChatGPT
1. Install TokenPulse from the Chrome Web Store 2. Open any ChatGPT conversation 3. The bar appears above the message input automatically
No API key. No configuration. The extension reads your existing ChatGPT session.
Practical strategies for managing ChatGPT context
Start a new chat for each distinct problem. The context window is per-conversation. If you're debugging three separate functions, use three conversations. Don't carry authentication code context into your database layer debugging session.
Summarize and restart when the bar hits 60%. At 60% of the context window, you still have enough room for Claude to give you a useful summary. Ask for a concise summary of the key decisions, constraints, and current state, then start a new chat with just that summary pasted in.
Put your system context last, not first. Most developers paste the full file or system at the top of a conversation and then ask questions. This is backwards — the model's attention is weighted toward recent tokens. Put the stable context at the top, yes, but front-load your most important question or constraint in the most recent message.
Use shorter code snippets. Instead of pasting an entire 300-line file, paste the specific function with 20 lines of surrounding context. You get better answers and burn a fraction of the tokens.
ChatGPT vs Claude context tracking
ChatGPT and Claude handle context differently:
Claude exposes internal rate limit data — exact 5-hour and 7-day utilization percentages — which TokenPulse can read directly. ChatGPT does not expose this data, so TokenPulse uses client-side estimation for ChatGPT context tracking.
For ChatGPT users, the token bar is the primary value. For Claude users, the popup gives you rate limits, reset countdowns, and utilization percentages on top of the bar.
Bottom line
If you're a developer using ChatGPT for coding work, you need a token counter. The alternative is flying blind until quality degrades and you have to restart a conversation. A browser extension that shows a live bar above the input box is the lowest-friction solution — install it once and forget it's there until the bar tells you to pay attention.