AI answers are too long and padded
Ask a simple question and get a wall of text: a restated question, a bulleted overview, caveats and a summary of what it just said. You skim for the answer every time.
Open the fixLast verified: Claude Sonnet 5.5 ยท October 2026
You read the draft and you know within a line. The em dash, the 'it's not X, it's Y' turn, the tidy list of three, the opener that announces what is coming. None of it is wrong. All of it says a machine wrote this. This page gives you a prompt that builds your own anti-tell block, built around one detail that matters: an AI follows a ban far better when you tell it what to do instead.
Paste this into a fresh Claude or ChatGPT chat. It interviews you about the habits that give AI writing away, then hands back an anti-tell block you paste at the top of any writing chat.
You're going to build my ANTI-TELL BLOCK: a reusable instruction block I'll paste into any AI chat that writes for me, so the output stops carrying the habits that make it read as machine-written. Interview me first, one question at a time. Where I don't know, propose a default and move on: 1. What do I write with this AI (emails, posts, proposals, reports) and who reads it? 2. Which tells annoy me most? Offer these for me to react to: the em dash, 'it's not X, it's Y' constructions, lists of three, opening lines that announce the point, closing lines that summarise it, rhetorical questions, stacked adjectives, and words like 'delve' and 'landscape'. 3. What punctuation do I want instead? Show me the same sentence three ways (full stop, comma, colon) so I can pick. 4. Paste two or three sentences I wrote that I'm happy with, and one AI sentence that grates, so you can see the gap. 5. Which tells are fine in my context, so you don't over-ban? Then output my ANTI-TELL BLOCK as one copy-paste block containing: (a) one line on who I write for, (b) punctuation rules, each with its replacement (a full stop or a colon where the dash would go), and the obvious substitutes (semicolon, en dash, spaced hyphen) named as banned too, (c) pattern rules, each with the plain alternative (state the point once instead of 'not X but Y'; make the statement instead of asking a rhetorical question), described in words rather than by quoting the banned phrasing, (d) a banned-phrase list, (e) a closing self-audit: before showing me any draft, scan it for every banned item and rewrite the sentence, then append 'Tell check: pass' or list what changed. Keep it under 350 words. After the block, name the three tells the AI is most likely to swap in once the obvious ones are banned, and how the block catches each.
Telling an AI what not to do tends to get you the next-worst habit. These four steps keep the block working.
Here is the block the generator hands back, for our running example company, the fictional consultancy Fenwick & Cole.
CONTEXT: I write client proposals and board memos for senior retail clients who are short of time. PUNCTUATION: Never use the em dash. Where a dash would go, use a full stop or a colon. Do not swap in a semicolon, an en dash or a spaced hyphen. Use hyphens only inside compound words and number ranges (3-5). PATTERNS: State each point once, as a plain claim. Never frame a point by denying one thing and asserting another, in any wording. Use three items only when there really are three things. Make the statement instead of asking a rhetorical question. Begin with the point, never with a sentence announcing what comes next. Stop when the content stops, with no closing line that restates it. BANNED PHRASES: delve, landscape, in today's fast-paced world, it's worth noting, at the end of the day, game-changer, unlock, seamless, robust. SELF-AUDIT: before showing any draft, scan it against every rule above and rewrite any sentence that breaks one. Then append one line: 'Tell check: pass', or list what you changed and why.
The model reproduces patterns from the polished writing it learned from, where the em dash is common, and it reaches for it as an easy way to join two thoughts. It is a habit of the output, not a setting you can switch off. What you control is the instruction you give it.
Ban the character and give the replacement in the same instruction: use a full stop, a comma or a colon instead. Store that in your project or custom instructions so it applies to every chat, and add a line telling the AI to scan its draft and fix any it finds before it shows you. The generator prompt on this page builds that block.
An instruction typed into a single chat fades as the conversation grows, and a bare ban with no replacement leaves the AI looking for a substitute. Put the rule in stored instructions, pair it with the alternative, and add the self-audit line. For long drafts, check the output anyway.
No. 'Not X but Y' constructions, lists of three, announce-then-say openers and neat summarising closers are just as common, and banning only the dash tends to move the habit somewhere else. The block this page builds covers punctuation and sentence patterns together.
Ask a simple question and get a wall of text: a restated question, a bulleted overview, caveats and a summary of what it just said. You skim for the answer every time.
Open the fixAI output arrives with curly quotes, arrows, long dashes and non-breaking spaces. Pasted into code, a CSV, a CMS or an email tool, they become broken strings, odd characters or failed imports.
Open the fixThe AI writes competent but generic prose: wrong register, hedging, buzzwords you would never use, sometimes American spelling. Someone rewrites every draft to sound like the company, which cancels out the time saved.
Open the fixThe anti-tell block you just generated is the kind of rule an AI Brain stores as always-on context, tuned on your real work and kept current as new problems appear. Paste it and this chat stops carrying those habits; installed, every chat starts that way, because the rule lives in the system and not in your memory of what to fix.
What else does AI get wrong that you fix by hand? Browse the other fixes
Free founder tools
Each tool gives you a first-pass answer in your browser. No email gate and nothing stored.