AI Studio is four generators on one screen: captions, hashtags scored by competition, CTAs and content ideas. You pick a brand, type a topic, hit Generate, and copy what survives your read. Output quality tracks input specificity almost exactly. A vague topic gets you filler. A topic with a product, an audience and a reason gets you something you can post.
What each generator takes and returns
The four modes take different inputs. Captions and hashtags need a topic. CTAs do not — they take a goal instead. Ideas need only a brand and a platform.
| Generator | You give it | You get back |
|---|---|---|
| Captions | Brand, topic, platform, optional tone | 4 caption variants, ready to post |
| Hashtags | Brand, topic, platform | 20 tags, each with a low/medium/high competition bucket and a 0-100 score |
| CTAs | Brand, a goal in plain words | 5 calls to action, each under 12 words |
| Ideas | Brand, platform | 8 ideas, each with a title, an angle and a content type |
Every result has a copy button. Nothing publishes from this screen. AI Studio is the drafting surface — you copy into the composer, attach media, and schedule from there. If you want posts drafted and slotted without you sitting in Studio, that is autonomous mode, which is a different workflow with an approval gate in front of it.
Pick the brand first
The brand dropdown defaults to "No brand context". Leave it there and you get competent, anonymous social copy — the kind you then rewrite for twenty minutes. Pick a brand and the generation is conditioned on that brand's stored record before the model writes a word: voice, audience, reading level, sentence length, banned words, emoji policy, hashtag ceiling, CTA style, claims policy.
This is the single largest quality lever in the product, and it is a dropdown people skip. If your captions keep coming back sounding like everyone else's captions, the fix is usually not a better prompt — it is a filled-in Brand Brain. Banned words in particular do real work: add "revolutionary", "unleash" and "game-changing" once and you stop deleting them by hand forever.
Brands are isolated. An agency running eleven client brands gets eleven separate Brand Brains and eleven separate sets of connected accounts. Nothing carries across.
What makes a good topic input
The topic field accepts up to 500 characters. Most people use fifteen. A topic is not a subject line — it is a compressed brief. The model already knows your voice from the Brand Brain, so do not spend the field on tone. Spend it on facts it cannot know.
Weak topics, and what they produce:
- "summer collection" — captions about sunshine that could belong to any label in India
- "new launch" — the model invents what launched
- "engagement post" — a format request with no subject, so you get a poll about coffee
- "skincare tips" — five tips you have read on forty other accounts
Stronger versions of the same topics:
- "Summer cotton kurta set, ₹1,899, launched today, for women 25-40 in Delhi and Bengaluru who work in offices with unreliable air conditioning"
- "Restocking the oil-free sunscreen after nine weeks out of stock — apologising for the wait, one bottle per customer for the first week"
- "Asking followers which of three sneaker colourways to produce next: bone, olive, or all-black. Votes close Friday"
- "Why our moisturiser has no fragrance, explained by the formulator, for customers with reactive skin"
The pattern: a product, a price or a constraint, an audience, and a reason the post exists this week. Four facts. That is the difference between output you edit and output you post.
Tone is a separate optional field on captions, capped at 60 characters. Use it for a one-off deviation — "apologetic", "deadpan", "urgent" — not for your usual voice. Your usual voice belongs in the Brand Brain, where it applies to every generation including the autonomous ones.
Reading the hashtag competition score
Hashtags come back as 20 tags, each in a bucket and each with a 0-100 score for reach potential. Be clear about what that number is: it is the model's estimate based on what it knows about the tag, not a live count pulled from Instagram. It is a ranking to sort by, not a measurement to report. Analytics readback — actual performance data flowing back into the product — is on the roadmap, not shipped.
Read the buckets like this:
| Bucket | What it means | What to do with it |
|---|---|---|
| Low | Small tag, few posts competing | Take most of your set from here — this is where a small account is actually visible |
| Medium | Real audience, survivable competition | Two or three. These are your reach attempts |
| High | Huge tag, your post is buried in seconds | One at most, and only if it genuinely describes the post |
A workable mix for an account under roughly 10,000 followers is one high, three medium, and the rest low — weighted toward tags that describe the specific thing in the post rather than the category it belongs to. #SunscreenForOilySkin will find fewer people than #Skincare, and more of them will care.
Two habits worth forming. First, cut. Twenty tags is what the generator returns, not what you should post. Set a hashtag ceiling in your Brand Brain and the ceiling is applied to generated posts automatically. Second, look at the tags that are obviously wrong for you — a tag pulled from the wrong country or the wrong price tier tells you the topic input was too thin, and it is faster to fix the topic than to hand-curate the list.
Hashtags do not behave the same way everywhere. Instagram and TikTok reward them; LinkedIn tolerates two or three; on X they mostly cost you characters. The per-network limits and dimensions are listed on the platforms page.
Pick a CTA by the action you actually want
The CTA generator has no topic field. It has a goal field — 120 characters, plain English — and it returns five CTAs, each under 12 words. The default goal is "saves and shares", and most people never change it, which is why most people's CTAs do not do what they need.
Decide the action first, then type it in:
| What you want | Type a goal like | What good output looks like |
|---|---|---|
| Saves | "saves — the post is a reference people return to" | Names the future moment: the CTA tells them when they will need this again |
| Shares | "shares to stories with a named friend" | Gives them a person to send it to, not just a nudge to share |
| Link clicks | "clicks to the product page from the bio link" | One destination, one reason, no second ask |
| DMs | "DMs — reply with a keyword to get the size chart" | A specific keyword, so the conversation starts itself |
| Comments | "comments answering a question with a short answer" | A question with a two-word answer, not an essay prompt |
| Follows | "follows from people who found this post through a hashtag" | States what the next post will be |
Three rules that matter more than the wording. One CTA per post — two asks split the action and you get neither. Match the CTA to the format: a carousel that teaches something should ask for a save, not a click, because people who are learning do not leave. And keep CTA style in the Brand Brain rather than picking it fresh each time, so autonomous posts carry the same ask your manual ones do.
Ideas fill a calendar, not a post
The ideas generator returns eight items, each with a title, an angle and a content type — post, carousel, reel. It is the only mode that takes no topic at all, which makes it the right tool for the Monday morning problem of having nothing planned and the wrong tool for writing anything.
Use it as a sorting exercise. Generate eight, keep the two or three you would actually defend in a meeting, then take each survivor back to the captions tab as a properly specified topic. The angle line is the useful part — it is the sentence you expand into your topic brief. Ideas you can only describe as "engagement post" were never ideas.
What this costs, and what it does not do
One Generate click is one generation against your monthly quota, whether it returns four captions or twenty hashtags. Free gives you 30 generations a month, Starter 500, Pro and above unlimited — the full breakdown is on the pricing page. At 30, you are budgeting; at 500, you can afford to regenerate a caption three times before you like it.
You choose the model behind all of this per workspace: OpenAI, Claude, Gemini, Groq, Mistral, DeepSeek or HuggingFace. If no key is configured, a deterministic mock provider answers instead, so the screen never hard-fails on a missing key — but mock output is placeholder text, not content. If your captions look strangely repetitive, check which provider is actually selected before you blame the prompt.
What AI Studio does not do, plainly: it does not read your analytics, so it cannot tell you which of yesterday's captions worked. It does not detect trends, do SEO, render video, or track competitors — those five agents are labelled SOON in the product because they are not built. It does not moderate comments or manage your inbox. And the monthly spend cap is stored and displayed but not enforced by the worker; the daily post cap is the one that actually stops things.
A ten-minute weekly routine
- 01Open Ideas, pick your brand and platform, generate eight. Keep three. (1 generation)
- 02Take idea one to Captions. Write a topic with four facts: product, price or constraint, audience, reason it is posting this week. Generate four variants. (1 generation)
- 03Read the four out loud. Keep the one that sounds like a person at your company. Edit the first line — it is the only line most people see.
- 04Switch to Hashtags with the same topic. Generate twenty, keep one high, three medium, six low, weighted toward the specific over the categorical. (1 generation)
- 05Switch to CTAs. Type the action you want in plain words — not "engagement" — and generate five. Take one. (1 generation)
- 06Repeat steps 2-5 for ideas two and three. Nine generations for three finished posts, which fits inside the free plan twice over.
- 07Anything you rewrote by hand in the same way twice, put into the Brand Brain instead. That is the edit you never make again.