AI chat

What the AI chat does

This is a free, multi-turn AI writing assistant. You type a message, it answers, and the reply streams in word by word so you can start reading before it finishes. It is built for the everyday writing work that does not fit a single button: drafting something from scratch, rewriting a paragraph that is not landing, brainstorming angles, tidying rough notes into prose, or just thinking out loud with a responsive second pair of eyes. The point is the conversation. You are not filling in a form and getting one output; you are working with a tool that keeps the thread and lets you steer.

Being plain about what it is matters. It is an assistant and a drafting aid, not a search engine and not an authoritative source. It writes well, which is exactly why it is easy to over-trust. Used as a fast first pass that you then edit and verify, it saves real time. Used as a publish button, it will eventually embarrass you.

How it works

You send a message and the assistant replies. The whole conversation stays in context, so each new message builds on what came before: you can ask for a draft, then ask for a shorter version, then ask it to make the third paragraph warmer, all without repeating yourself. Anything you paste, a draft, a brief, a list of notes, is treated as material to work on rather than as instructions to the tool. When you switch to a new topic, start a fresh chat so the old context does not muddy the new request.

The reply streams in, which is worth using to your advantage. You do not have to wait for the whole answer before you react. If the opening sentence is already heading the wrong way, stop it, adjust your message, and send again, which saves time and keeps the conversation tight. When an answer is close but not quite right, the fastest path is usually a short follow-up that names the one thing to fix, rather than rewriting your original request from scratch. Small, specific nudges between turns are how you get from a decent first reply to one you can actually use.

What it is good at

The sweet spot is open-ended text work. In practice that means first drafts, rewrites in a different tone, outlines you can rearrange, shortening or expanding a passage, turning bullet points into flowing prose, explaining why a sentence reads awkwardly and offering three fixes, and breaking the inertia of a blank page. Because it holds the conversation, it is also good at iteration: the second and third tries are usually better than the first because you have told it what you did not like.

What it is not

It does not browse the web. It has no live data, no awareness of this week's news, no current prices, and no knowledge of anything after its training. It is not a search engine, and it is not an authoritative source. Every reply is AI-generated and should be checked before you act on it. If you ask for a current fact, it will tell you it cannot look that up rather than inventing one, and it is told not to fabricate statistics, quotes, studies, or named sources. That instruction reduces the problem; it does not remove your responsibility to verify.

Multi-turn, done well

The conversation is the feature, so use it. Give feedback between turns rather than starting over: say what you liked, what missed, and what to change next. Paste a draft and ask for one specific edit at a time, which gives you cleaner control than asking for five changes at once. When the reply drifts, pull it back with a short correction. And when you move to an unrelated task, start a new chat. A long thread that has wandered across three topics gives worse answers than a fresh one focused on the task in front of you.

Writing a prompt that gets a useful reply

The clearer the request, the better the reply. State the goal, the audience, the length, and the tone, and give an example of what good looks like when you can. One clear instruction beats five vague ones. Compare these:

• Weak: “Write something about onboarding.” The model has to guess the format, the reader, and the length, and it tends to guess generic.
• Strong: “Draft a 150-word welcome email for new hires at a small design studio. Warm but not gushing. Mention that their laptop is ready and that their buddy will message them on day one.” Now it has a shape to fill.

You do not have to get the prompt perfect. That is the advantage of a conversation: send a rough request, see what comes back, and refine. The first reply is a starting point you react to, not a verdict.

A worked example

Say you have three scrappy notes for a team update: shipped the new export feature, two bugs still open, and a demo is booked for Friday. You send: “Turn these notes into a short, plain status update for a non-technical manager,” and paste the notes. The assistant comes back with a tidy paragraph. You read it and decide it is a touch formal, so you reply: “A bit more relaxed, and lead with the demo.” It rewrites with the demo up front and a friendlier tone. Finally you ask: “Now give me a one-line version for Slack.” Three turns, and you have both the paragraph and the one-liner, each shaped by feedback you gave along the way. That is the workflow the tool is built for.

Honest limits and using it responsibly

A few rules keep this useful and safe. Verify facts and figures, because the model can state something wrong with full confidence. Never paste anything confidential you would not want processed by a third-party model: passwords, private client data, financial details, or anything under an agreement you have signed. Treat the output as a draft you own and edit, adding the one thing the model cannot supply, which is your own knowledge and judgement. The tool removes the slow, mechanical part of writing; the thinking and the accountability stay with you.

How the chat compares to the focused tools

Phrasit also has single-purpose tools, and often one of those is the faster, more predictable choice. If you want to reword a paragraph, the paraphraser gives you a mode and a strength and a side-by-side diff. If you want a summary, the summarizer lets you pick the shape and length. If you want a post from a brief, the blog writer is built for that exact shape. Those tools are predictable because they do one thing. The chat wins when the task is fuzzy, when you want to talk it through, or when you will need a few rounds to land it. A good habit is to start in the chat when you are figuring out what you want, then move to a focused tool once you know the single transform you need.

Privacy and cost

The chat is free with no signup. To keep it that way there are fair-use limits per visitor and an overall daily budget, which protect the tool from abuse and keep it available for everyone. There is also a cap on how long a single conversation can run, so when a chat gets long, start a new one, which tends to sharpen the replies anyway. On privacy, the simple rule is to share the minimum the task needs and to keep anything sensitive out of the box. The reply is generated by a third-party model, so paste only what you are comfortable having processed off your device.