Media
4 items
Description
One click sets the response style of ChatGPT, Claude or any AI chat: ASD-STE100, BLUF, plain language. Ask two at once. Every AI assistant has the same voice. Long, hedged, wrapped in bullet points, opening by repeating your question back to you and closing with an offer to help further. So you type "be concise" at the end of your prompt. Again. And tomorrow, in a different tool, again. Stylespec turns that into a button. Start typing in any AI chat and a small "Set response style" pill appears next to the box. Click it, pick a style, and the instruction is added to your draft — visibly, where you can read and edit it before you hit send. Nothing is added behind your back. The library is not a list of moods. Alongside the everyday ones it carries the writing standards people are actually held to: ASD-STE100 Simplified Technical English, BLUF, plain language, and the house styles Google and Microsoft publish for their own documentation. WORKS ON EVERY ASSISTANT, NOT JUST ONE Stylespec has no list of sites it knows how to handle. It identifies a chat box by how it looks and behaves, so the same pill appears in the assistant you use today, in the self-hosted front-end your team runs, and in whichever one launches next month — with no update needed. Thirty-one assistant sites are enabled out of the box, and the current list is on the homepage. This is the part per-vendor settings cannot do. ChatGPT's Custom Instructions are excellent — inside ChatGPT. Claude has Styles. Gemini has Saved Info. Set up three tools, maintain three copies, watch them drift. Your Stylespec library follows you everywhere and outlives whichever assistant you prefer this year. ASK A SECOND ASSISTANT WITHOUT RETYPING ANYTHING Some questions are worth more than one answer. Getting a second one normally means selecting your prompt, copying it, opening a tab, pasting it and sending — every time. Open the Multiplexer in the picker, tick ChatGPT or Claude, and send the way you always do. Each one opens in a background tab to the right of the one you are in, already asking your question, with the style instruction attached. You never leave the tab you were in, and you read the answers whenever you are ready. Two things make this different from the comparison tools you may have tried. It starts where you already are. Those tools are a destination: you type into their sidebar or their grid instead of into your assistant, which means leaving the conversation you were already having. Stylespec works the other way round. It sends the draft you had already finished writing, from the site you were already on. The comparison is actually fair. Answers are only comparable if the prompt was identical, and the shape of a prompt drives the shape of an answer more than most people expect. Because the same style instruction goes to both assistants, you are comparing the models rather than comparing one model's default verbosity against another's. The details worth knowing before you rely on it: • ChatGPT and Claude are built in. Any other assistant that accepts a prompt in its address can be added in Settings, including an internal or self-hosted one • Answers arrive in ordinary browser tabs, on each assistant's own site. There is no split-screen pane and nothing is embedded • Nothing is collected, scored or summarised for you. Reading your answers back would mean reading page content, which is the one thing this extension is built not to do • It arms for one prompt and then switches itself off, so a long conversation never turns into a wall of tabs • Tabs open in the background and inactive. You keep your place 23 STYLE GUIDES, READY TO USE Clarity • Explain it like I'm 5 — everyday words and one concrete analogy • Plain language — plainlanguage.gov: short words, active voice • Feynman technique — first principles, and it flags what it hand-waves • Socratic tutor — questions instead of answers, one at a time • Hemingway — short declarative sentences, no adverbs • No LLM tells — no em dashes, no "delve", no wrap-up paragraph • Put the fries in the bag — answer first, zero preamble, then stop Technical • Simplified Technical English — ASD-STE100: one meaning per word, under 20 words per sentence • Google developer documentation style • Terse senior engineer — code first, prose only where it carries weight Professional • BLUF / executive summary — bottom line up front, three bullets, done • Microsoft Writing Style Guide • Academic — formal register, hedged claims, no invented citations • Legalese — defined terms, numbered clauses, shall and may • Financial analyst — numbers first, assumptions stated, units labelled Format • Actionable checklist — numbered, verifiable, with the exact commands • Bullets only • Table first — lead with the comparison, explain underneath Critique • Devil's advocate — strongest objections first, no encouraging closer For fun • Pirate, Shakespearean, Corporate LinkedIn, Gen Z — accurate underneath the accent Write your own too. Name it, give it an icon, type the instruction, and it sits alongside the built-ins. The preview pane shows exactly what will land in the chat box. STACK MODIFIERS ON TOP Any style can be adjusted without editing it: • Length — terse, medium or thorough • Formality — casual, neutral or formal • Audience — beginner, general or expert • Emoji — none or sparing • Language — 30 languages, with the right register That last one matters more than it sounds. Where a language separates formal from informal address, the register is set explicitly, so you get usted rather than tú, Sie rather than du, 존댓말 rather than 반말. Getting that wrong in an email is not a small thing. BUILT TO STAY OUT OF THE WAY • The pill appears on your first keystroke and vanishes when you are done • Keyboard: Command+Shift+Y on macOS, Alt+Shift+S elsewhere, rebindable • Apply a favourite straight from the toolbar popup • Light, dark, or match your system — because the AI site's theme and yours often disagree • Turn it off for any site in one click PRIVATE BY DESIGN, NOT BY PROMISE Stylespec cannot run on a site unless you have said so. Most extensions like this ask to "read and change all your data on all websites" the moment you install them. Stylespec does not, and cannot. It loads on the thirty-one assistant sites in its manifest and nowhere else. Anything beyond that list is one explicit click: the popup asks Chrome for permission to that single site, and that is all it ever gets. This is worth comparing against multi-assistant tools specifically. Sending one prompt to several assistants normally requires standing access to all of them at once. Stylespec needs no such thing: for assistants that take a prompt in their address it passes the prompt that way and injects nothing at all. Stylespec makes no network requests of its own. No analytics, no tracking, no remote code, no accounts. Your prompts are read by the page you typed them into and nowhere else. Your styles live in your browser's own sync storage. The one time your prompt leaves the page you typed it on is when you tick a Multiplexer target yourself, and then it goes to that assistant and nowhere else — the same as if you had pasted it in. This is not a policy that could quietly change next release. It is what the extension is structurally able to do, and you can verify it: the source is public, the manifest is 92 lines, and the whole thing is small enough to read in an evening. Open source: https://github.com/bilal62/stylespec FREQUENTLY ASKED Does this make the AI better? No. It makes the AI answer the way you want to read. The substance is the model's; the shape is yours. How is the Multiplexer different from the side-by-side comparison extensions? Those give you a dashboard with every model in it, and if that is what you want, one of them is the better tool. Stylespec is not a dashboard. It is the assistant you already had open, with a tick box that saves you the copying and pasting — and it is the only one that sends the same style instruction to both sides, so the answers are comparable in the first place. Can it send to Gemini or Perplexity as well? Add them in Settings. Anything that accepts a prompt in its address works, and the built-in list stays short on purpose: those addresses are undocumented and can change without warning, so each built-in target is a maintenance promise rather than a line on a feature list. Does it work with my company's internal assistant? Yes, for both halves. Open the popup on that site and hit "Turn it on here". Chrome asks for permission to that one site, and the same detection takes over. That click is deliberate: it is why Stylespec does not have to ask for access to every website you visit. Does a longer prompt cost more tokens? A style block is 40 to 70 words and reliably removes several hundred words of output. It nets out cheaper. Can I see what it adds? That is the point. The instruction goes into your draft in plain sight and you can edit or delete it before sending. Simplified Technical English and ASD-STE100 are specifications of ASD. Microsoft is a trademark of Microsoft Corporation. Google and Gemini are trademarks of Google LLC. ChatGPT is a trademark of OpenAI. Claude is a trademark of Anthropic. This extension is not affiliated with, endorsed by or authorised by any of them.