AI Assistant

Learn what the AI chat agent can do and how to get the most out of it.

Overview

The SonetHub AI assistant is available from the Chat page. It's a conversational agent that can take real actions on your behalf — not just answer questions, but actually create posts, schedule content, query analytics, manage your inbox, and more.

What It Can Do

Here's a summary of what you can ask the assistant:

Content Creation

  • Create content — "Write an Instagram post about our summer sale"
  • Adapt for platforms — "Turn this into a LinkedIn post and a tweet"
  • Create polls — "Create a Twitter poll asking which feature users want next"
  • Create threads — "Write a 5-tweet thread about our product launch"
  • Publish stories & reels — "Publish this image as an Instagram story"
  • Publish documents — "Post this PDF as a LinkedIn carousel"
  • Publish articles — "Share this link on LinkedIn with a summary"
  • Generate images — "Create an image of a sunset over a city skyline" (costs media credits; a draft preview costs 1 credit)
  • Generate videos — "Create a 5-second video of ocean waves for Instagram Reels" (Creator+, costs media credits). Briefs with multiple beats expand into a multi-shot chain automatically — the assistant reports the shot count and total credit cost in its reply. Pass auto_chain: false if you want to refuse expansion. Single-shot durations: 5 or 10 seconds. Multi-shot total lengths: 12, 24, 36, or 48 seconds (the brief expander splits these into 2-6 shots). Video generation runs in the background; ask the assistant to check the status when ready.
  • Regenerate one shot — "Redo shot 2 of that reel" — re-renders just that shot of a completed multi-shot video at its original tier and re-stitches the chain. Charges only that shot's credits.
  • Extend a video — "Add 6 more seconds to that reel" — appends an extension shot anchored on the last frame. Single-shot videos are promoted to a 2-shot chain on first extension. Standard-tier videos can't be extended; regenerate them on High or Premium first.
  • Generate music — "Generate upbeat background music for my 18-second reel" (costs media credits). Music duration is flexible (any 5-60s) — ask for the exact length of your video so the audio fits with no silence at the end. Cost ≈ 1 credit per 5 seconds.
  • Generate voiceover — "Voice this line: 'Done. Your launch now looks intentional.'" (Creator+, costs media credits). The text is spoken verbatim — write the exact words you want said. Pick a voice (20 presets, or paste a custom/cloned voice ID); use a different voice per line to give each person their own voice, and reuse the same one to keep a character consistent across clips. Cost ≈ 1 credit per 100 characters.
  • Lip-sync a voice onto a video — "Lip-sync that voiceover onto the talking-head clip" (Growth+, costs media credits). Re-syncs a speaker's mouth in an existing video to a new audio track, or animates a still photo into a talking video. Pair with the brand voice so the same voice speaks across every clip. Cost ≈ 4 credits per second.
  • Assemble scenes into one video — "Stitch those three clips into one video" — joins finished clips in order into a single video (clips must share resolution/aspect/frame rate). Add a music bed or voiceover afterwards.
  • Analyze images — "Analyze this image's visual style and save it as my brand style" (costs 5 credits as a chat message, no separate media credits)
  • Subject preservation — "Generate a photo of me at a tech conference using this headshot as reference" (uses your reference image to keep the same subject in a new scene; 5 credits at high quality, 10 at ultra)
  • Style transfer — "Generate an image in the style of my brand reference" (copies the visual look or edits the reference; 10 credits)
  • Search for images — "Find a stock photo of a team meeting"

Publishing & Scheduling

  • Publish immediately — "Publish this to all my accounts"
  • Schedule posts — "Schedule this for tomorrow at 9 AM"
  • Edit published posts — "Update the text on my latest LinkedIn post"
  • Search locations — "Find the location tag for New York for my Threads post"

Engagement & Discovery

  • Monitor mentions — "Show me my recent mentions" or "Any unread Instagram comments?" (Creator+)
  • Manage comments — "Show me the comments on my latest post" or "Reply to that comment"
  • Like/hide comments — "Like the top comment on my TikTok" or "Hide that spam comment" (TikTok Business)
  • Manage inbox — "Show me my important DMs" or "Reply to the latest message"
  • Search social content — "Search Threads for posts about AI marketing"
  • Look up profiles — "Look up @username on Threads"
  • TikTok trends — "What's trending on TikTok right now?", "Show me popular TikTok sounds", "Suggest keywords for fitness content" (TikTok Business)
  • Repurpose content — "Repurpose this article for Instagram and LinkedIn: https://example.com/post" — extracts the article and generates multiple distinct posts from different angles

Engagement triage

The assistant can read and manage your engagement queue — the priority slice of comments and mentions that need a human.

  • "Show me what needs attention" — reads the priority queue (read_mentions with the needs-attention filter): external mentions plus important, negative, question, or automation-flagged comments that are still open.
  • "Filter to open questions on YouTube" — filters by platform (now including YouTube and TikTok Business), category, and triage status.
  • "Mark this handled" / "dismiss this" / "flag as important" / "reopen it" — changes triage state (triage_mention). Dismissing clears an item from the queue so it trends toward zero.
  • "Reply privately to this commenter" — sends a private reply or DM where the platform supports it (reply_comment), matching the page's "reply privately."
  • "How's my engagement trending this week?" — engagement analytics (query_analytics engagement overview): volume, sentiment split, reply rate, category mix, daily trend, and the posts driving the most comment engagement.
  • "Email me when someone leaves a negative comment" — configures engagement alerts and the daily-digest rollup (update_notification_settings), including natural-language criteria for what makes a comment important.

These actions respect your workspace role: triage and like/hide need edit access; replying needs publish access.

Meta Ads (Growth+)

The assistant manages Meta ads end to end — with hard safety rules: everything launches paused pending your approval, and it never raises a budget without asking.

  • Set up from zero — "What's my ads setup?" audits your Business Manager, ad account, payment method and account standing live, and walks you through every step
  • Boost a post — "Boost my last reel with €20 for 3 days in Spain" — drafts a paused boost for your approval
  • Create a campaign — "Run a traffic campaign for my web shop, €150 over two weeks" — synthesizes a brief from your brand context, researches a custom audience, and explains its reasoning on the approval card
  • Research audiences — "What Meta interests fit a vegan restaurant in Madrid?"
  • Manage campaigns — "Pause the summer boost", "approve the pending campaign", "list my campaigns"
  • Report performance — "How are my ads doing this month?" — spend, impressions, clicks, CTR/CPC/CPM per campaign and rolled up
  • Auto-boost rules — "Boost any post that gets 100 likes in its first 24 hours" (see automations)

Analytics & Management

  • Check analytics — "How did my posts perform last week?"
  • Get insights — "Give me analytics insights" — returns benchmark comparisons, competitor analysis, posting time recommendations, content topic performance, and anomaly detection. Insights are pre-computed from your data (no extra AI cost).
  • Content topics — "What topics drive the most engagement?", "How do my travel posts compare to food posts?"
  • Benchmark check — "Is my engagement rate good for my follower count?" — compares against industry benchmarks specific to your platform and follower tier.
  • Search the web — "What are the trending topics in AI this week?"
  • Set up automations — "Create a daily posting schedule for LinkedIn" (Pro: up to 5, Business: unlimited)
  • Manage workspace — "Invite jane@example.com as an editor"
  • Check usage — "How many Agent messages and media credits do I have left?"
  • Manage WhatsApp — "Check my WhatsApp status" or "Disable WhatsApp access" (Pro+)
  • Submit feedback — "I'd like to request a feature for bulk scheduling"

Brand Intelligence

  • Teach your brand — "Here's my website and brand guidelines, learn my brand"
  • Analyze brand materials — Upload logos, PDFs, or reference images to extract visual style, voice, and preferences
  • Brand intelligence is extracted and saved as memories that personalize all future content generation
  • You can also upload files via Settings > AI > Teach AI Your Brand
  • Each analysis takes up to 20 inputs: every file, every URL and the pasted text count as one. Files can be images, PDFs, CSVs or text files, up to 5 MB each
  • The analysis runs in the background and shows each item as it finishes; large batches can take a few minutes

Brand Voice Memory

On all plans, the AI remembers details about your brand — tone, audience, key messages, style preferences. When you tell it things like "we always use a friendly, casual tone" or "our audience is B2B SaaS founders," it saves these as memories and applies them to future content generation.

Memories are extracted only from your own conversations with the AI assistant — not from your DMs, analytics, or any third-party data. You can view, edit, and delete individual memories at any time from Settings > Brand Voice or by asking the assistant "what do you remember about me?" or "forget that our tone is formal."

Visual Style Memory

The AI can learn and remember your brand's visual identity. Upload a reference image and ask the assistant to analyze it — it will extract colors, mood, composition, and style. Save the analysis as your brand's visual style, and all future image/video generations will incorporate those preferences automatically.

Examples:

  • "Analyze this image and save it as my brand style"
  • "My brand uses coral and teal with minimalist photography"
  • "Generate an Instagram post image using my brand style"

Reference Image Library

Beyond brand style, you can teach the AI named subjects — people, products, recurring scenes — so it can compose them into images and videos on request. Upload an image, give it a name (e.g. "Peter") and a one-sentence visual description, and from then on:

  • Ask "Generate an image of Peter at his desk" and the AI will compose Peter into the new scene.
  • Ask "Peter and Rachel having coffee in a cafe" and it will pull both named references into one image — multi-subject composition works out of the box.
  • Mention "use my AutumnPalette as a style reference" and it will transfer that visual look to the new image.

The library lives under Settings → AI → Reference images and you can also manage entries by asking the assistant directly ("save this image as Peter — he's a man in his 30s with brown hair and glasses"). The AI only uses references you explicitly mention — generating an unrelated logo won't accidentally feature Peter.

Saved Designs

When the AI designs a graphic you like (an ad, a quote card, a carousel), ask it to keep it: "save this design as Auction promo". Later, ask for "a new post in the Auction promo style" — in chat or in an automation — and the AI reuses that design's colours, fonts, spacing, logo treatment and format, changing the text and imagery for the new content and rearranging elements only when the new content needs it. Manage saved designs under Settings → AI → Saved designs.

Motion Videos

Ask for a Reel, Story or short video made of text, numbers, steps or an interface mock-up ("make the myth vs fact Reel") and the AI designs and renders it itself: it writes the scenes, animates them and produces an MP4, usually in 1 to 3 minutes. Typical uses are explainers, FAQs, data points, quotes, announcements and animated versions of a carousel. These videos use no media credits.

Before showing you the result, the AI checks it: every scene stays on screen long enough to read, nothing is cut off or hidden under the app's buttons and caption, text contrast, and fonts. Videos are up to 60 seconds. They are silent unless you ask for sound: the AI can lay a track from your library, generated music or a generated voice under the video (generated audio uses credits). Ask for changes ("make the second scene slower", "change the photo") and the AI edits the same video, checking the affected frames before rendering again. Screens shown in these videos are designed mock-ups with example data, not recordings of a real product, and footage of people or places still comes from AI video generation or your own uploads.

Narrated videos

Ask for a voice that explains the video ("make it a narrated Reel") and the AI writes a script with one line per scene, records it as a single take, and times every scene to the moment its line is spoken, so voice and picture stay together. If you also ask for music, it plays underneath at a low, steady volume so the voice stays clear.

  • Cost: the voice costs about 1 credit per 100 characters, typically 3 to 5 credits for a Reel. The render itself is free. Changing the spoken words means recording again; changing only the picture does not.
  • Choosing the voice: the accent comes from the voice, not from the text. Ask the AI for a few samples, pick one by listening, and it will remember it for the brand's next videos.
  • Listen before publishing: the AI checks the timing but cannot hear the recording, so it will always ask you to listen.
  • The on-screen text stays, written to work on its own for people watching without sound.

Content Strategy and Planning

The assistant keeps one content strategy per workspace (goal, pillars and their mix, cadence per platform, key dates) and plans against it. Ask for a week or a month of content and it proposes dated planned posts for your approval before writing any of them. See Scheduling & Automation for the Plan panel on the Calendar.

Design Quality Checks

Every design the AI renders is checked automatically before you see it: text cut off or too close to the edges, overlapping text, text contrast against the actual pixels behind it, more than two typefaces, and fonts that failed to load. The AI fixes what the check reports, then reviews the image itself. Character limits and hashtag caps (for example, at most 5 hashtags on Instagram and TikTok, and X's weighted count) are checked when a post is saved or published, and you get a clear message naming what to change.

Long Tasks

Designing a carousel or a video can take a few minutes. While the assistant works, the chat shows what it is doing (thinking, preparing a tool, running it) and for how long.

  • You can reload or close the page. The assistant keeps working on the server. When you come back, the chat shows the steps it has completed so far, what it is doing right now and how long it has been at it.
  • Stop ends the task. Closing the page does not. Work the assistant had already finished and saved stays in the conversation, followed by a note that you stopped it.
  • You can send the next message once the current task has finished or you have stopped it.
  • Batches of planned posts are written in the background. The chat shows a card with each post's progress, and when the last one is done the assistant comes back on its own to review the drafts and tell you how they turned out (this counts as one message).

Tips for Best Results

  • Be specific — "Write a LinkedIn post about our Q1 results highlighting 40% revenue growth" works better than "write a post"
  • Iterate — Ask the AI to refine: "Make it shorter," "Add more emojis," "Make the hook stronger"
  • Use context — Mention your platform, audience, and goals for more targeted content
  • Confirm before publishing — The AI always asks for confirmation before publishing or scheduling, so you can review and adjust