How-To Guide
How do you brief an AI social media agent to create product content for your SaaS?
A worked product-task brief for an AI-run social account: one task, declared sources, checkable claims, and formats the account can place today.
By the Superdirector editorial team. The five practitioner guides, the demo anatomy study and the federal advertising guide cited here were read in full on November 14, 2026, and the capability statements were checked against superdirector.app/pricing and the product boundary file the same day.
At a glance
How do you brief an AI social media agent to create product content for your SaaS?
You brief it the way you would brief a new teammate who writes fast and knows nothing about your product: one product task per brief, the buyer it serves, the sources it may use for product facts, the claims it may make with named evidence, and the formats it can actually publish. Make every instruction checkable, so a draft can be failed against it, and give the agent permission to ask when information is missing instead of inventing it. The worked brief below applies all of this to one release-note task for an invented SaaS example.
- A brief is a task sheet: one task, one buyer, one format, declared sources, claims with evidence, and an avoid list.
- Test every instruction with the fail-a-draft rule; adjectives do not transfer, but sentence lengths, banned strings and claim rules do.
- Bind the formats you request to today's publishing reality, which for an AI-run operated account is Instagram carousels and single link-free X text posts.
- The agent should return copy with its reasons separate, ask when facts are missing, and never publish invented results.
Every row earns its place only if a draft could fail against it. Original diagram, drawn from the brief-sheet rows shipped with this guide.
Make it work for your brand
Find it. Remake it. Post it. Superdirector turns the viral formats in your niche into your brand's carousels, X posts and video scripts, and keeps the calendar full.
What goes in a brief for an AI social media agent
An AI agent writes from what you give it. A vague brief gives generic posts that you rework one by one; a precise brief gives drafts close to what you would have written. The difference is not talent, it is specification, and a precise brief has six parts: one product task, the buyer it serves, the format you want, where product facts live, which claims are allowed and with what evidence, and what to avoid.
Two checks come with the sheet. First, every instruction must pass the fail-a-draft test: if two careful people would disagree about whether a draft complies, the line is a mood, not an instruction. Second, the brief must fit the account's real capability: a task that asks for a finished video when the account can only place carousels and text posts produces nothing, no matter how well it is written. The limits are in the brief on purpose, because the agent does not know what your account can publish unless the brief says so.
A good brief template ends with a rule most people forget: if you're missing information, ask instead of making it up. An agent that is told it may ask makes things up less, and when the agent operates an account that publishes on its own, that single line is the difference between a question and a public mistake.
Write the task, not the vibe
Most briefs for AI writers fail because they describe a feeling and call it an instruction. Friendly and professional is the standard example: the model returns the statistical middle of everything ever posted, which is exactly what the words ask for. The fix is to test every line of the brief against one question, could you fail a draft against it? A sentence-length cap, a banned string, a named post structure and a hook pattern with slots are checkable. Tone adjectives are not, so they earn no place in the sheet.
The task itself gets one row: what this post does for one buyer this week. One sentence in the primary-message slot works better than a paragraph, because the sentence names what the reader should think, feel or do. A primary message like promote our new feature is not a message; make marketers realize they can hand expense imports to the app is. Then the format row names the platform and shape, and the volume row says how many posts, so nothing is left to interpretation.
Avoid lists carry the load on how drafts should read. Banned words, banned constructions, forbidden claims and tells readers spot in AI copy all belong here, often as more useful lines than the positive instructions. Keep the sheet to the rules that repeat, and let one spec serve every channel with a short per-network modifier row for length, formality and hashtag policy, so the briefs stay one document instead of five.
Declare where product facts come from
A product claim written from a model's memory is a guess with punctuation. The brief names the declared sources for product facts: your release notes, documentation, captured screens, and the changelog entry for the task. If the task references a product capability, the rule is to retrieve official documentation before drafting copy, and to base the response on validated sources rather than prior assumptions. Grounding the copy, in other words, is not a feature of the tool stack, it is a row in the brief.
Require the output shape that makes review possible: separate final copy from rationale and assumptions. Polished wording without traceable grounding creates review risk, especially for technical claims, so each draft tells you the release note it used, the screen it captured and the claim it left out for lack of a source. The extra structure costs you the seconds it takes to read the reasoning line, and buys you the confidence that the copy's facts were written once, from a source you named.
Claims discipline is where this stops being a style matter. Advertising rules require evidence behind claims: before a company runs an ad, it has to have a reasonable basis for the claims, meaning objective evidence that supports the claim. Put the never-claims row in the brief with the test attached: no number without a named source, no result promised to the reader, no comparison to a named rival, no guarantee. Missing information comes back as a question, which the ask rule already covers.
Put today's capability limits in the brief
A brief that ignores what the account can actually do is a plan with no place to land. The capability rows below are the current published scope of a Superdirector Growth account, which is the AI-run operated account this page is published by, and they matter here because they define which task formats a brief can contain. Read them as a template for the same section in your own brief: whatever your tool is, its own capability page decides the rows, not your hopes.
| Capability row for the brief | Today's state for a Growth account | What it means for the task you write |
|---|---|---|
| Publishable formats | Eligible Instagram carousels and single link-free X text posts | Write the task as a carousel or a text post, not as a video or a thread |
| How posting happens | Posts by itself once posting is turned on for an account, under its standing permission, with no per-post approval; posting starts after one warning that names the account, the public effect and the cost | Say in the brief which formats may go out unattended and which wait for review |
| AI video | Upcoming and not available yet; video scripts are never on the calendar, and credits are not a count of finished videos | Do not ask for a finished video; the script is the deliverable and filming stays with you |
| Metering | Credits cover AI work and vary by task; Growth includes 2,000 credits a month against Starter's 300 | Batch the month's product tasks instead of running the same brief many times |
| Accounts | Growth includes two accounts; extra accounts are $99/month each; each account keeps its own audience, voice and purpose | One brief per account, so a change on one never drifts into another |
The stakes of getting the limits row wrong are asymmetric: a task the account cannot place wastes a whole review round, while a task it can place but your brief never says to write simply never happens. The pricing page is where this section gets updated, since that is the published scope. If you run the $29 Starter plan, which is the hands-on version you drive from Claude or ChatGPT with live social data, ideas, scripts and downloads, the brief's publishing rows read as your own to-do list instead.
Two formats the account can place now, and one the brief cannot ask for yet. Original diagram, drawn from this guide's capability rows.
Copy the worked product-task brief
Here is the whole sheet for one task. The example is invented on purpose, a small expense-tracking app called Ledgerly, and an illustrative example is not completed customer work. Swap in your own product, task and sources, and the sheet keeps its shape.
| Brief row | What it says in the worked example |
|---|---|
| Task | Post this week's release note: bulk CSV export is live. One Instagram carousel of 4 slides plus one single link-free X text post, both about the same task. |
| Buyer | A bookkeeper or small-business owner who imports expenses by hand and has never configured an export rule. |
| Primary message | One sentence: you can move a month of expenses from your bank to the ledger in one import instead of one row at a time. |
| Declared sources | Release note LN-217 pasted below word for word; two captured screens of the export flow, sent with the task; documentation page for CSV import. Nothing else. If the note and a screen disagree, ask. |
| Claims | Only what LN-217 states. No time saved, no error rates, no comparison to other tools, no numbers without a source named in the draft's rationale line. Label AI content where the platform requires it. |
| Format limits | Carousel: 4 slides, one idea per slide, slide 1 works alone as a feed post. X post: single link-free text post, no thread. AI video is not requested because the account's video generation is upcoming, not available. |
| Voice budgets | No em dashes, no exclamation marks, no rhetorical-question openers, no engagement bait, second person, contractions on. Hashtags: up to the platform's policy, and links never in the carousel text. |
| Review rules | Return copy separated from a rationale line naming the source used and any claim dropped for missing evidence. Ask when LN-217 does not answer a question. The buyer decides what posts, what changes and what gets paid. |
Why each row is there. The task row is one weekly goal, not a content calendar, because a brief that asks for everything asks for nothing. The sources row exists because the product docs are the declared ground, not memory. The claims row repeats the evidence rule, and the format-limit row exists because the account's publishable formats are carousels and text posts today, while finished AI video is upcoming. The review row keeps the decision where it belongs: you make the important calls, what posts, what changes and what gets paid.
The demo-type task rule travels from the demo research: a post that shows the product has a beginning, a middle and an end, and the ten demos analyzed in that study all avoided the hard sell. For a release-note carousel this means slide 1 states the task the buyer has, the middle slides show the flow on real screens, and the last slide tells the buyer where the feature lives, not what it will convert them into.
Review the first drafts like a reviewer
The first review of a new brief is a specification test, not a taste test. Run the same brief through the agent twice with no changes. If the two outputs are wildly different, the brief is underspecified and the interpretive room is the problem; if they are nearly identical and both miss the mark, the brief is over-constrained or a parameter is wrong. The good middle is two clearly related drafts, both plausibly on task, one slightly stronger.
Then judge each draft with a binary call: would you publish it unedited? Not is it close, not could I fix it in a minute. Failures get one tag each, off-task, off-voice, banned string, claim without a source, missing hook, and the tags are what you change next. An agent asked to be persuasive invents a statistic the way it invents a transition, so a claim-without-a-source tag outranks every style tag in priority.
The review rules in the brief are what make this fast. Copy separated from rationale means you can check the source line before the prose, and the ask rule means missing information arrives as a question instead of an invention. If the agent operates the account under a standing permission, the review stakes include the formats that post by themselves once posting is on, so the brief's own review row, not habit, decides which formats go out unattended.
Accepted drafts return to the brief, while the monthly drift test sits outside the weekly loop. Original diagram, drawn from this guide's review rules.
Keep the brief alive and versioned
A brief is not written once. After the first week, keep the drafts you accepted as they were and paste two or three into the next brief as examples, then remove what is no longer useful and add what you had to correct. Every draft you annotate becomes one more line the agent cannot get wrong: never an exclamation mark, always the link placement you chose, always the source named before the number. After a month the sheet describes how your product writes better than any style guide.
Measure drift before you feel it. Keep ten fixed briefs, never edited, and run them once a month, making the same publish-unedited call on each. Log the pass count with the date, the model version and one failure tag per miss. A falling score with nothing else changed usually means the model moved under you, and a once-only failure is noise you leave alone, while a failure that happens twice in the same ten is a missing rule that becomes one checkable line.
Version the sheet. A one-line version marker beside the pass count is what tells a drop from a coincidence, and it is the same habit the annotated-examples rule needs: examples decay faster than rules, so they get pulled and replaced while the rule numbers only ever grow. The brief that ages like this keeps one job: every writing act stays conditioned on evidence, not on hope.
Where Superdirector fits
Superdirector, which publishes this guide, is the AI-run operated account this brief pattern is written for. The $199/month Growth plan makes your posts, carousels and video scripts with AI and includes two accounts, and each $99/month extra account adds +1,000 credits to the plan's 2,000; the $29/month Starter is the hands-on version a person drives from Claude or ChatGPT with 300 credits. On Growth, once posting is turned on for an account, it can publish that account's eligible Instagram carousels and single link-free X text posts, with no per-post approval, subject to account readiness and plan entitlement.
That scope is why the capability rows sit inside the brief instead of beside it. Finished AI video is upcoming and not available yet, so today the deliverables a brief can ask for are scripts, carousels and text posts, and the account posts eligible formats by itself after one warning that names the account, the public effect and the cost. You keep the calls that matter: what posts, what changes and what gets paid. The brief above is a plain copy of that division of labor, and it works the same for any tool that operates an account, provided its capability page is the source for the limits row.