Word counter — tokens, reading time and platform limits live

What makes a good word counter in 2026?

A word counter looks trivial: take the text, split on whitespace, done. That definition collapses the moment you switch text genres. A German compound like “Donaudampfschifffahrtsgesellschaft” is one word — naive counters that split on hyphens still get it right, but the cousin tools that split on lowercase-to-uppercase get it wrong. An English tweet with hashtags is a different beast than an excerpt from an academic paper. And a chat history you want to drop into an AI assistant is not measured in words but in tokens.

This word counter fixes all three problems at the root. Word recognition uses the browser’s built-in Unicode word segmenter, the same engine that translation tools and word processors rely on. It understands German umlauts, French accents, CJK script boundaries and English apostrophe-words. The AI-token estimator uses an empirical conversion ratio per language — and visualises immediately whether your text fits a 4k, 8k, 32k or 128k context window. The platform-limit gauges sit as live progress bars right beneath the editor, so you can see how much room is left while typing.

How is this different from Word or Google Docs?

Microsoft Word and Google Docs count words reliably, but only as a side feature of a five-hundred-megabyte text-processing program. They are not optimised to check a text for X eligibility, Bluesky limits or LinkedIn readability. They do not show reading time either. Word ships a basic readability index (Flesch Reading Ease in the US-English variant) but no language adaptation for German.

A specialised word counter does something a universal editor does not try: it is platform-aware. The input field is a single large textarea that adapts to the screen. Eight live tiles right under it update on every keystroke. In the Platform-limits tab you see immediately whether your text still fits 280 characters for X, or has already passed LinkedIn’s 3000-character ceiling. The AI-tokens tab shows the estimated token count plus a four-pill display: fits in 4k, 8k, 32k, 128k. A classic word processor does not solve this job.

Why do I need a token estimate for AI assistants?

AI assistants don’t process text in words or characters; they process it in tokens. A token is a sub-word unit the underlying model learned during training. English words like “cat” or “run” are often a single token. Longer English words like “extraordinary” split into three or four tokens. German compounds are especially expensive — “Krankenversicherungsvertrag” can easily occupy eight or ten tokens because the tokeniser has never seen that exact word and has to assemble it piece by piece from its sub-word vocabulary.

When you paste text into an AI assistant, the context-window size is the deciding factor. Common sizes are 4 000, 8 000, 32 000 and 128 000 tokens. The word counter uses an empirical conversion ratio — roughly 4 chars per token for English and roughly 3 for German — and shows immediately which bucket your text falls into. That is precise enough to make a decision (“does this fit in an 8k prompt or do I have to trim?”) and stays lightweight: the browser does not need to load a tokenisation model.

What does the readability index tell me?

In the Statistics tab a readability value sits next to the live tiles on a 0–100 scale. Higher values mean easier text. For German text the word counter uses the Amstad adaptation of the Flesch formula: Heinrich Amstad calibrated the original formula in his 1978 thesis at the University of Zurich for the longer average sentence structure typical of German prose. For English text the classic Flesch Reading Ease (1948) does the math.

Both scales are compatible: 60 to 70 is school-textbook level, 30 to 50 reads as academic, below 30 is specialist jargon, above 80 is light reading for younger audiences. A very low score is not automatically a problem — technical texts should sound that way — but if a marketing piece lands below 40, it is worth checking sentence length. The two mini tiles “Avg. word length” and “Avg. sentence length” point at the direct levers.

How do the platform limits work?

The Platform-limits tab lists eight platforms with their official character ceilings: X at 280, Bluesky at 300, Instagram at 2200 (with an additional soft cut at 125 characters where the caption gets truncated with a ‘more’ link in the feed), LinkedIn at 3000, TikTok at 4000, Google title at 60, Google meta description at 160, and the classic GSM-7 SMS at 160 characters.

Each platform gets a horizontal progress bar. While you are within the limit, the bar fills in the regular graphite tile colour. The moment you exceed the limit, the bar flips to the rust-red token and the counter to the right shows ‘X chars over’. Instagram’s bar also draws a tick mark at 125 characters — that is the position where Instagram truncates a caption in the feed view.

How reliable is the compound-word detection?

The browser’s built-in word segmenter recognises German compounds as one word — that is the agreed-on behaviour across every modern browser engine. Harder is the follow-up question: how many components does a long compound consist of? “Donaudampfschifffahrtsgesellschaftskapitän” is morphologically about five morphemes: Donau + Dampf + Schiff + Fahrt + Gesellschaft + Kapitän, with a Fugen-S binding them. A full morphological analysis would need a dictionary that is too heavy to ship into a browser.

The word counter therefore uses a heuristic: words of 12 characters and longer are probed for CamelCase splits and common German suffixes (-schaft, -keit, -heit, -ung, -lich, -bar). The detection is good enough to flag long words visually and provide a reasonable morpheme estimate, but it does not replace a linguistic analysis. You can see this in the UI: morpheme information sits as additional detail next to the tile grid, not as a top-line number.

What happens with very long texts?

The word counter is written so that an entire Anna Karenina (roughly 360 000 words, 2 million characters) runs in the browser without any UI jank. The live computation is linear in the text length and uses a single pass over the input. The n-gram analysis in the Keyword-density tab is O(n) with n = number of words, plus a small sort at the end for the top 10.

In practice: for everything from tweets to 10 000-word articles you see results without any perceptible lag. For novel-length texts you may feel a small amount of input lag on phones, but it disappears in milliseconds. There is no artificial upper bound. If you really want to push the boundary, a 100 MB text in the clipboard is the point at which the browser itself struggles with the paste — not the word counter.

What is deliberately NOT in scope?

Three jobs sit outside the scope on purpose. First: no grammar or spell check. The browser’s own spell checker and dedicated tools like LanguageTool already handle that. Second: no file upload for DOCX, PDF or similar formats. Anyone who needs to analyse a Word file usually already has it open in a word processor and can copy the content out. Third: no cloud sync. Your text stays in the tab — if you want to keep it, copy it yourself.

These gaps are deliberate. They keep the word counter fast, lightweight and private. Each of these three features would require either a server round-trip, a 5–10 MB library in the browser or a cookie for cloud login — and that would not be the same tab you are working in now.

What about privacy and GDPR?

Your text stays in browser memory. There is no localStorage persistence, no cookies, no analytics, no server round-trip. The tab works offline once the page has loaded. The ‘Copy summary’ button assembles the plain-text statistics locally via the Clipboard API and writes them straight to your clipboard — no network call.

From a GDPR perspective the word counter processes no personal data in the sense of Art. 4 because processing happens locally on your device. Even if you paste sensitive content (email drafts, legal filings, medical notes), the text remains in the tab. Closing the tab discards everything. That architecture is deliberate — it is the reason this word counter is also usable in enterprise setups with strict data-protection requirements.