Beyond sliders and presets, IcloneU lets you write the rules your Clone follows in your own words — how to greet, what to steer around, and which topics to refuse outright. They all live in one place: the Additional Instructions modal on the Clone's Identity tab, with a separate sub-tab for general rules, public comments, private messages, audience segments and undesirable content.
Slider-based tone controls — response length, formality, predictability, emoji density, plus emotional-tone chips — live on the Clone's Personality tab, not in this modal. The Personality tab: sliders and voice controls, near the end of this guide, covers every one of them. Everything before it is about the instructions you write yourself.
Open the Additional Instructions modal
The entry point is the Instructions section of the Identity tab — one button to edit, one to inspect the result and one to go back to an earlier version.
Open your Clone's Identity tab
Click Clones in the left sidebar and pick your Clone from the My clones list. The detail pane opens on the Identity tab by default.
Find the Instructions section
Scroll below General information to the Instructions section. It offers three buttons: Edit Instructions, View compiled instructions and History.
Click Edit Instructions
A modal titled Additional Instructions opens with five sub-tabs: General, Public Messages, Private Messages, By Audience and Undesirable Content.
NoteOn a freshly created Clone the modal has occasionally been observed to open empty. If that happens, click Cancel and re-open it; if it persists, reload the page.
What each sub-tab does
Each sub-tab scopes its instructions to a different situation:
| Sub-tab | What goes there |
|---|---|
| General | Behavior rules the Clone follows in any context — greetings, escalation rules, things to steer around |
| Public Messages | Rules that apply only when the Clone replies to comments on posts or other public content |
| Private Messages | Rules for direct or 1:1 messages |
| By Audience | Separate instructions for five fixed audience segments — see below |
| Undesirable Content | Topics the Clone must politely decline — see below |
General, Public Messages, Private Messages and Undesirable Content each give you a single free-text field capped at 500 characters — enough for a handful of crisp rules, in the spirit of the General tab's own example: "Always greet users by name. Avoid discussing competitor products. When asked about pricing, direct users to the sales team."
The General sub-tab is the only one that also accepts an instructions file, via Upload Instructions File — .txt, Word (.docx), PDF, Markdown (.md) and RTF, up to 5 MB. The other sub-tabs have no upload — Public Messages, Private Messages and Undesirable Content get only their 500-character field, and By Audience its five segment fields. An uploaded file keeps its line breaks, so rules and bullets stay on their own lines, and clicking the file's name after uploading shows the exact text your Clone reads. If an upload doesn't work, a red message appears above the Upload Instructions File button and the file you had before, if any, stays in place: when no text can be read from the file — a scanned PDF, for example — the message names the file and says the previous instructions were kept; when the file can't be read at all, it says "The file could not be read. Try again, or use a different file (.txt, .md, .pdf, .docx, .rtf)." A file uploaded before 17 September 2026 was stored without its line breaks, and it stays that way until you upload it again — IcloneU can't restore them for you. To check, click the file's name: if the text runs together in one block, upload the file again, click Save in the modal and then Save changes on the save bar.
Anything you would look up rather than obey belongs in the Clone's Knowledge tab instead of here — catalogue details, sizing charts, FAQs, shipping and returns policies, addresses and opening hours. The reason is mechanical: your instructions travel with every single message the Clone answers, while Knowledge is looked up only on the turns that actually need it, so moving reference material across both lowers the units each reply uses and sharpens the answer. What stays here is the voice, the tone and the rules the Clone must follow every time — and those stay even when they run long, because a rule only works if the Clone has it in view on the reply where it mattered. Payment details are the one exception: they go in the instructions, as Where should I put my bank details? below explains.
By Audience: five segment fields
Instead of one box, By Audience shows five separate fields — Female, Male, Young, Middle aged and Senior — with no character counter. Each field completes the same sentence — "If the user is identified as that segment, …" — so you write only what changes for that group, for example a more empathetic tone or simpler vocabulary.
The segmentation is currently a fixed gender-and-age scheme. Segments based on role, customer status or relationship stage aren't available today.
Topics to avoid: Undesirable Content
The Undesirable Content sub-tab is for "topics to avoid": its field asks what the Clone should never discuss — its own examples are misinformation, competitor products and internal pricing — within the same 500-character cap. The Clone gets this list with its replies in conversations and to comments, on every channel and in the Playground, together with the instruction to decline politely in one or two sentences, without repeating, expanding on or debating the topic, and to steer the conversation back to what it can help with. Leave the field empty and nothing is added.
That is what the tab's note promises: "The clone will politely decline and redirect when asked about these topics." Two things work differently from the other sub-tabs: the list isn't part of the Compiled prompt, and the check that runs when you click Save doesn't look at it. AI drafts of replies to reviews don't use it either. Send a test message about one of the topics in the Playground to see the refusal.
Check the result: the Compiled prompt
Back on the Identity tab, the second button — View compiled instructions — opens a modal titled Compiled prompt showing your instructions put together: General (with any uploaded file), Public Messages, Private Messages and By Audience. It's the easiest way to read the rules you give your Clone in your own words.
It shows what was put together the last time you clicked Save in the Additional Instructions modal. It appears there straight away, but it's kept only once you save the Clone (Discard takes it back), so it's empty on a Clone whose instructions were never saved that way — including a new Clone that starts with text from your onboarding answers in General. Open Edit Instructions, check that General shows your text (if the modal opens empty, close it and open it again), click Save, then Save changes on the save bar. If it's still empty, the check that runs when you click Save (the modal shows Validating instructions...) couldn't run; your instructions are still saved and used, and only this preview is empty. Try Save again later, then Save changes.
Go back to an earlier version: History
The third button, History, opens Instruction history: the current instructions, marked Current, then the earlier versions, newest first, each with when it was saved, who saved it and from where — the Clone editor, the mobile app, the style analyzer, an approved feedback-learning update or a restore. A version is kept every time one of the instruction fields changes — any sub-tab, the uploaded file, the learned style or the feedback-learning rules — up to the newest 50 per Clone; turning a switch on or off doesn't count. The instructions a Clone had before the history existed show Before history instead of a date.
Click a version to read it: each field it holds is shown, and the ones that differ from today's are marked Differs from current. Restore this version makes it the live Clone's instructions straight away, without Save changes, and keeps the instructions it replaces as a version too, so a restore can be undone the same way. It brings back every instruction field, the learned style, the feedback-learning rules and the Compiled prompt that version had, but not the handover topics or the greeting. Save or discard any other unsaved changes on the page first: until you do, the restore button stays off. Anyone who can open the Clone can view the history; only members who can edit the Clone can restore a version.
How the language setting fits in
Instructions shape how the Clone speaks; the Primary language field, higher up on the same Identity tab, is the last stop in the reply-language chain. The Clone replies in the language of the customer's current message; when a message has no clear language of its own (an emoji, a number, an "ok"), it stays in the language the conversation was already in; and only when neither gives an answer does it fall back to Primary language. Ten locales are currently available: Spanish (Spain, Mexico, Argentina), English (United States, United Kingdom), Portuguese (Brazil, Portugal), French (France), German (Germany) and Italian (Italy).
Save, then test
Edits made in the modal still have to be saved on the Clone page before they reach your live Clone.
Save the modal
Click Save in the modal footer — or Cancel to discard the edits.
Confirm on the save bar
Back on the Clone page, an Unsaved changes bar appears at the bottom with Discard and Save changes. Click Save changes — as the bar itself notes, changes apply to the live clone immediately after save.
Test in the Playground
Open Playground in the top-right of the detail pane and send a few messages that exercise your new rules — including one about a topic your instructions tell it to avoid — to watch the tone and the polite refusal land.
The Personality tab: sliders and voice controls
The sliders live on a different surface from everything above: the Clone's Personality tab, next to Identity. The tab opens with a Voice summary — an auto-generated, one-line read of how your Clone currently sounds — followed by the Tone & style group: four sliders, each with four positions.
| Slider | Scale (left → right) | What it tunes |
|---|---|---|
| Response length | Terse · Short · Medium · Long | How long typical replies run — a couple of sentences at one end, step-by-step detail at the other |
| Formality | Casual · Informal · Neutral · Formal | The register of language — contractions and first names at one end, titles and measured sentences at the other |
| Predictability | Focused · Balanced · Creative · Playful | How varied the wording is from reply to reply |
| Emoji density | Off · Sprinkle · Often · Heavy | How often emojis appear in replies — from never to most messages |
If you came looking for "formal or casual", that's the Formality slider; "concise or detailed" is Response length — its own hint contrasts Concise, staying under a couple of sentences, with Detailed, which can walk through steps and examples.
Below the sliders, Emotional tones offers eight chips — Friendly, Humorous, Professional, Empathetic, Encouraging, Serious, Casual and Optimistic. Pick up to a few for the Clone to lean into; as the control's own hint puts it, they bias word choice and punctuation but don't override your instructions.
The Addressing & persona group decides who is speaking. Self-reference picks between Singular (I / my) — a personal voice — and Plural (we / our) — a business speaking for the brand. Gender (Undefined, Feminine or Masculine) sets the Clone's own gender; in languages with gendered grammar it drives adjective and pronoun agreement. Language proficiency sets how simple or idiomatic the wording runs, and the Match user's vocabulary toggle mirrors the complexity of the incoming message. The card ends with two voice-note controls, Answer voice notes with voice and Voice, covered in Voice-note replies below.
Last comes AI Learning, which teaches the Clone your voice in two sub-modes. Style Analysis has an Analyze my style button that learns your style from your previous responses, and the Apply learned style to replies toggle, which decides whether that learned style is used — in its own words: "Combines with the settings above. Turn off to reply without using the learned style." Feedback Learning learns from the ratings, notes and edits your team makes to the Clone's replies. Once a day, or when you click Run Feedback Learning, it turns new feedback on at least three replies into suggested rules; working them out uses units, whether you apply them or not. The suggestions don't change the Clone by themselves: they wait in a Proposed update from recent feedback card below What your clone has learned, which says "Review the suggested rules below. Your Clone keeps answering with its current rules until you apply them." Apply makes them the rules the Clone answers with, straight away; Discard drops them, and that feedback isn't used again. A run before you decide adds its new feedback to the same proposal. Applied rules are always used, whatever the Apply learned style to replies toggle says.
Voice-note replies
Your Clone can answer a voice note with a voice note. The switch is Answer voice notes with voice, at the end of the Addressing & persona card on the Personality tab, and it's off on a new Clone. With it on, when a customer sends a voice note the Clone replies in text as usual and then sends the same reply as a voice note. In Messages, the voice note shows as a player under the Clone's text.
Next to the switch, Voice picks how it sounds. The default, Automatic (by gender and language), chooses a voice from the Clone's Gender and language; pick a specific voice from the list instead and a player appears so you can listen to a sample first.
- Only in reply to a voice note. A typed message gets a typed reply.
- On WhatsApp, Telegram, Messenger and Instagram — the channels where customers can send voice notes. The web chat widget has no way to record one.
- Only for replies that go out straight away. A reply delayed by Reply timing or a Reply schedule date range, held for your approval, or waiting out an Auto-send after a delay review goes out as text only.
- Text only when it can't be spoken. A reply over 1,200 characters once formatting and links are removed, a reply with nothing left to speak, or an account that has run out of units gets the text reply alone.
- Voice uses units. Transcribing the customer's voice note and speaking the reply both use units, like any other reply.
In the Playground, the microphone beside the Message... box records a voice note as if you were the customer: the Clone transcribes it, answers, and — with the switch on — voices its reply too.
If a customer's voice note can't be transcribed, the customer gets an automatic reply in English, whatever the conversation's language: “I couldn't understand the audio message. Could you please try again or send a text message?” On Messenger and Instagram that includes a voice note over 25 MB or one that couldn't be downloaded. On WhatsApp a voice note that can't be downloaded gets a different English reply, “I couldn't download the audio message. Could you please try again?” On Telegram, one that can't be downloaded gets "I was unable to download the voice message." — the one reply here that follows the customer's language: it goes out in Spanish when their Telegram app is set to Spanish, in English when it's set to English, and otherwise in your account's language if that's Spanish, or in English.
Frequently asked
Those are slider controls on the Clone's Personality tab — Response length, Formality, Predictability and Emoji density, plus the Emotional tones chips. The Personality tab: sliders and voice controls, above, covers each one with its scale labels. The Additional Instructions modal complements them with rules in your own words; those rules, apart from Undesirable Content, are what the Compiled prompt shows. Nothing from the Personality tab appears there.
The General, Public Messages, Private Messages and Undesirable Content fields each cap at 500 characters, with a counter under each one showing how many are left. If you paste more than fits, only the text up to the limit is kept, and the note under the field turns red: "Your pasted text was cut at the character limit — everything past it was not kept." The five By Audience fields show no character counter. For longer material, the General sub-tab accepts a file upload — .txt, Word (.docx), PDF, Markdown (.md) or RTF, up to 5 MB.
Because the modal added up the General field, the file uploaded under it, and the Public Messages and Private Messages fields, and the combined text passed 8,000 characters. On the General sub-tab, below the units warning, it then shows: "Tip: part of this text looks like reference material. Catalog details, sizing charts, FAQs, and policies work better in your clone's Knowledge — your clone looks them up only when a conversation needs them, which lowers the units used per message and improves accuracy. Keep the voice and behavior rules here." It's advice, not a limit — the tip itself blocks nothing: your text isn't truncated or rejected because of its length. In practice it's the uploaded file that trips it, since the typed fields cap at 500 characters each; the fix is to move the reference material into the Clone's Knowledge tab and leave the behaviour rules — and your payment details — in the instructions.
In the instructions — the General field or the file uploaded under it, which the Clone reads on every reply — written exactly as customers should copy them. IcloneU checks every reply your Clone writes in a conversation or to a comment for account numbers, CLABEs, IBANs and card numbers. Such a number is sent only if it's written in your instructions or in the Knowledge the Clone looked up for that same reply (on some older accounts, only your instructions count); otherwise it's taken out before the reply goes, together with the paragraph it sits in (and the short line introducing it, if there is one). Nothing replaces it, the rest of the reply is sent, and you aren't notified. If the reply held nothing else, the customer gets a fixed line instead — “A member of the team will follow up shortly with the payment details.”, in the conversation's language when that's English, Spanish, Portuguese, French, Italian or German, and in English otherwise — and your team is told. With Enable handover on, in a private chat, the conversation is handed over with the reason The Clone had no payment details to share; otherwise (handover off, a reply to a public comment, a conversation already handed over, or one where the Clone asked for its own handover on that reply) an urgent notice, Payment details needed, arrives by default in the bell and by email, with a link to the conversation. Knowledge is looked up for the question at hand, so bank details kept only in your library are taken out of any reply whose question didn't bring them up, even if the Clone gave them earlier in the conversation. A number that comes only from the conversation, a connected store or a function library doesn't count either. And when the Clone can't see the details, it's told to say a person will send them; that alone doesn't alert your team.
It declines politely in one or two sentences, without repeating or debating the topic, and steers the conversation back to what it can help with. The list goes with the Clone's replies in conversations and to comments, on every channel and in the Playground. It isn't in the Compiled prompt, the check that runs when you click Save doesn't look at it, and AI drafts of replies to reviews don't use it. Test one of the topics in the Playground.
The modal shows what was put together the last time you clicked Save in the Additional Instructions modal; it's kept only once you save the Clone (Discard takes it back). So it's empty on a Clone whose instructions were never saved that way, including a new Clone that starts with text from your onboarding answers in General, or after a Save whose instruction check couldn't run. Open Edit Instructions, check that General shows your text (if it opens empty, close it and open it again), click Save, then Save changes, and re-open View compiled instructions. If it's still empty, the instruction check couldn't run; your instructions are still saved and used, and only this preview is empty. Try Save again later, then Save changes.
Last updated October 6, 2026 · AI Clones
