Page Token Inspector
See your page through AI's eyes. Tokenize any URL or pasted text with the GPT-4o tokenizer and find out where your message lands.
Why Page Token Inspector matters
AI models do not read your page like a human. They tokenize it, and long URLs, heavy code blocks, and generic boilerplate eat the budget before the model reaches your substance. This tool shows exactly where your tokens go, which words fall inside the opening slice, and how your page compares against the context windows of the models that would have to hold it.
How it works
- 01
Give it a URL or paste text
A URL is fetched and reduced to its main content, the way a model would receive it after boilerplate is stripped. Pasting raw text skips the fetch entirely, which is useful for grading a draft before you publish it.
- 02
We tokenize with the real encoder
o200k_base, the GPT-4o family tokenizer, run locally through js-tiktoken. These are the actual token boundaries, not a characters-divided-by-four estimate.
- 03
You see the split
Total tokens, a per-section breakdown, the exact text inside the first 512 tokens, and how the whole page measures against four real context windows.
What it checks
- Total token count
- What your page costs a model to process, computed with the GPT-4o encoder rather than approximated from word count.
- The first 512 tokens
- The opening slice, rendered token by token so you can see precisely where it ends. If your central claim sits after it, you find out here.
- Per-section breakdown
- Title, H1, H2, paragraphs, lists, and alt text tokenized separately, so you can see which parts of the page are consuming the budget.
- Context-window comparison
- Your page measured against GPT-3.5 Turbo at 16k, GPT-4o at 128k, Claude at 200k, and Gemini 2.5 Flash at 1M, with the percentage each would spend on you.