How is bulk text classification different from a keyword filter?

A keyword filter matches literal words or phrases and breaks the moment someone phrases something differently than expected. A classifier reads the actual meaning of each item against your category descriptions, so 'I want a refund' and 'can I get my money back' both land in the same billing category even though they share no keywords.

Can I use this for languages other than English?

The tool itself works with whatever text you paste in — it doesn't restrict input language. Accuracy for non-English text depends on the underlying model's own language coverage, which isn't something this tool controls or can guarantee.

What happens if my categories overlap?

The model will pick whichever category your description makes the strongest case for, and the confidence score will usually be lower on genuinely ambiguous items. If you notice a lot of low-confidence results, it's often a sign two of your category descriptions are describing similar things — tightening the descriptions to be more distinct usually fixes it.

Is there a paid tier for higher volume?

Yes — Pro removes the daily request cap for this and the other Jev-powered tools on the site.

How to Sort Customer Feedback and Support Tickets Into Categories Automatically

The problem with sorting text by hand#

Every support inbox, feedback form, and comment section eventually turns into a pile of freeform text that someone needs to sort. A hundred support tickets need routing to billing, technical, or sales. A week of app store reviews needs grouping by theme before anyone can act on them. The usual options are manual triage (slow, and it doesn't scale past a few dozen items) or writing keyword-matching rules (fast, but brittle — 'cancel' shows up in both 'I want to cancel my subscription' and 'please don't cancel my order'). Neither holds up well once the volume gets real.

What actually makes this fast: a model built only for classification#

The Bulk Text Classifier is powered by Jev, a model from TypeSafe AI that is deliberately narrow in scope — it doesn't write text, summarize, or hold a conversation. It answers one kind of question extremely well: given a piece of text, which of a fixed set of categories does it belong to, and how confident is that decision? Because it isn't generating free-form language, a single classification typically completes in under 500 milliseconds. That's roughly the difference between asking someone to write a paragraph versus asking them to point at one of five labeled boxes — the second question is just inherently faster to answer, and Jev is built around answering exactly that kind of question.

How the tool actually works, step by step#

Paste your list of items into the text box, one per line — up to 50 per request. Then define your own categories: a short label and a one-line description of what belongs in each one. The description does more work than the label — 'technical: bugs, outages, integrations' will sort more accurately than the word 'technical' on its own, because it gives the model something concrete to match against rather than a single ambiguous word. Click Classify, and the tool sends each item to Jev individually (with several running at once to keep things fast), then shows you a table: each item, its assigned category, and a confidence percentage. Anything that fails — a rare temporary API hiccup — shows up clearly marked as failed rather than silently guessing.

Reading the confidence score correctly#

The confidence percentage isn't a generic accuracy estimate bolted on afterward — it's Jev's own calibrated assessment of how sure it is about that specific choice, produced as part of the same decision. A 95% confidence classification and a 55% confidence classification were reached by the same process, but one of them is telling you plainly that it's a close call. Treat anything under roughly 70% as worth a human glance rather than an automatic decision, especially for anything customer-facing.

Where this is genuinely useful — and where it isn't#

This tool is a good fit for exactly one kind of job: taking a batch of independent text items and sorting each one into a category you define. Support tickets into departments. Feedback into themes. Product descriptions into internal taxonomy buckets. It is not a summarizer, it will not tell you what your feedback means in aggregate beyond the category counts, and it won't write anything for you — for that, the site's Text Summarizer or Sentiment Analyzer are separate tools built for those separate jobs. Using the right narrow tool for a narrow job is usually faster and more reliable than reaching for a general-purpose one.

The free plan, honestly#

5 requests per day, up to 50 items each — so up to 250 items classified per day at no cost, shared with the newer Bulk Feedback Triage tool (both run on the same underlying limit, tracked per IP address per day, not per tool). That's a meaningfully more generous limit than the site's LLM-based AI tools get, because a Jev classification costs a small fraction of what a full conversational AI call costs — the free tier reflects the real cost difference rather than an arbitrary number.