An intake form that stops bad jobs early
The nine required fields that prevent most rework, the default-answer trick that keeps intake fast, and the rule that nothing starts until the form is complete.
Run a post-mortem on your last ten pieces of rework and sort them by root cause. The pattern is boringly consistent: almost none of them are "the model produced a bad output." They are "we did not know where this was running," "nobody said the legal team had to see it," "the client meant the other product," and "we found out at delivery that it needed a disclosure label."
Every one of those was knowable at minute zero. The information existed. It just was not asked for, because the job arrived as a Slack message and a Slack message has no required fields.
An intake form is the cheapest quality control you will ever install. It costs the requester four minutes once, and it removes the class of failure that costs you a full second pass. Here is the version that works, and the enforcement rule that makes it stick.
The four things that are always missing
Before the field list, understand what you are actually buying. Nearly all preventable rework traces to one of four gaps.
Where it runs. Placement determines aspect ratio, duration, safe area, loudness, whether captions must be burned in, and whether a disclosure label is required. A brief without a placement is a brief without a spec.
Who decides. Not "the client." A name and a role. Multi-stakeholder feedback arriving through three channels is not one review round; it is three contradictory rounds arriving at once.
What cannot appear. Competitor products, unapproved claims, a discontinued SKU, a departed executive, a colour that belongs to another brand line. Negative constraints are almost never volunteered and almost always exist.
What rights are needed. Whether a real person's face or voice appears, whether music is licensed for paid use, how long the asset runs, and in which territories. This is the field that turns a small problem into an expensive one when it is skipped.
The nine required fields
Nine is the ceiling. Past that, requesters start guessing to get through the form, and a guessed answer is worse than a blank because it looks like information.
| # | Field | Why it prevents rework | What a bad answer looks like |
|---|---|---|---|
| 1 | Objective, in one sentence | Separates "we need a video" from "we need trial signups" | "Awareness" |
| 2 | Placement and format | Sets aspect ratio, duration, captions, safe area | "Social" |
| 3 | Audience | Drives casting, tone, language, reading level | "Everyone" |
| 4 | The single approver | Ends contradictory feedback rounds | "The marketing team" |
| 5 | Hard date, and what it is tied to | Distinguishes a real deadline from a preference | "ASAP" |
| 6 | Must include | Product, offer, legal line, logo lockup, CTA wording | Blank |
| 7 | Must not include | Competitors, unapproved claims, dead SKUs, people | Blank |
| 8 | Rights and usage | Paid vs organic, territories, run length, whose likeness and voice | "Standard" |
| 9 | Reference and anti-reference | One example that works, one that does not | A mood board with 40 images |
Field 9 does more work than its position suggests. A single anti-reference — "not this, it reads too corporate" — resolves more taste ambiguity than a page of adjectives, because taste is easier to bound than to describe.
Two fields deserve extra strictness.
Field 7 must not be optional. Make it a required text box with no "n/a" shortcut. If the requester genuinely has no exclusions, they type "none," and that word is now on the record. That single line has settled more disputes than any other part of the form.
Field 8 needs sub-questions, not free text. Ask three yes/no questions: does a real identifiable person appear, is this running as paid media, and is it reaching an EU audience. Each yes routes the job differently. The third one matters because Article 50 of the EU AI Act became binding on 2 August 2026 and requires realistic synthetic content to be labelled clearly enough that a viewer notices — what Article 50 means for sponsored content has the thresholds. Discovering that requirement at delivery costs a re-export and a re-approval; discovering it at intake costs a checkbox.
The default-answer trick
The objection to required fields is always the same: it slows requesters down and they route around it. The fix is not fewer fields. It is pre-filled defaults that are wrong in a visible way.
For every field, publish a house default. Placement defaults to 9:16, 15 seconds, burned-in captions. Rights defaults to organic-only, 90 days, no identifiable person. Approver defaults to the requester themselves.
Now the form takes ninety seconds, because the requester is confirming rather than composing. And critically, the requester who accepts "organic only, 90 days" has stated it. When the asset then runs as a paid ad in month five, that is a new deliverable with a new brief, not a rework, and the conversation is about scope rather than about blame. The revision policy that stops scope creep covers where that boundary sits and how to price the other side of it.
Keep the defaults in the same document as the form so they are visible while being accepted. A default the requester never reads is a fabrication with your name on it.
The rule: nothing starts until it is complete
The form is worthless without one enforcement rule, stated once and applied without exceptions:
A job with an incomplete intake is not in the queue. It has not been declined, deprioritised, or lost. It is simply not started, and its date has not begun.
Three implementation details make this survivable rather than obnoxious.
1. The clock starts at completion, not submission. This is the part that changes behaviour. If the deadline is measured from a complete form, the requester's own date depends on their four minutes. You are not blocking them; they are.
2. Incomplete jobs get one specific question back, not a rejection. "This is ready to start as soon as I know who the single approver is." One question, answerable in a reply, no form to reopen.
3. There is exactly one bypass, and it is named. A genuine emergency route that a named person can invoke, which explicitly waives the delivery estimate and the revision cap. Not a secret. Written down next to the rule. Bypasses that exist informally get used constantly; bypasses that exist formally get used about twice a year.
If your approval stage is already the constraint, intake discipline is worth more than extra production capacity, because incomplete briefs generate exactly the second passes that saturate review. Approval workflows that don't stall covers the downstream half of the same problem.
Wiring intake into production
An intake form that lives in a spreadsheet nobody opens during production is a form that gets filled in and ignored. Three connections make it load-bearing.
Fields 6 and 7 become the QC checklist. Do not write a separate one. The must-include and must-not-include lines are literally the checks that run before delivery, which means the reviewer is checking against what was asked for rather than against their own memory. Quality control for AI-generated marketing assets has the gate structure those checks slot into.
Fields 2 and 3 become the prompt frame. Placement and audience are the two brief fields that translate most directly into generation parameters, and they are the two most commonly reconstructed from memory halfway through a job.
Standing constraints live outside the form. Anything true of every job for a client — colours, fonts, logo rules, banned phrasing, tone — belongs in a brand kit rather than being retyped nine times. You can set a brand kit once and teach the agent a lasting preference for the rules that never change, which keeps the intake form focused on what varies job to job. That distinction — standing rules in the kit, per-job facts in the form — is what stops intake from bloating past nine fields.
Then keep the completed form attached to the output. Asset naming and version discipline is where that link lives, and it is what lets you answer "why is it like this?" six weeks later without a thread archaeology session. For the full path from a completed brief through production and handoff, the content team handoff workflow picks up where intake ends.
FAQ
Won't clients refuse to fill this in?
Some will push back on the first job and almost none on the second, because the first job is where they experience a delivery that matched what they meant. Frame it in their terms when you introduce it: the form is what lets you commit to a date, and a date is the thing they actually want. If a client genuinely will not complete it, fill it in yourself from the kickoff call and send it for confirmation — the confirmation is the part that matters, not who typed it.
How is this different from a creative brief?
A brief describes the work. Intake decides whether the work can start. They overlap, but the intake form is deliberately shorter, mostly structured rather than prose, and it has a pass/fail state. A brief with a beautiful strategy section and no named approver is a brief that fails intake, and that is the correct outcome.
What if the answers change mid-job?
They will, and the form is what makes that visible. A changed placement or a changed approver is a scope event, not a note — it should trigger a re-estimate rather than being absorbed silently. Log the change against the original answer so the rework is attributable to a brief change rather than appearing in your metrics as a production defect.
Does this work for internal teams, not just client work?
It works better internally, because you control both sides of it. The main adjustment is that field 4 gets harder rather than easier — internal requests routinely arrive with three stakeholders and no named decider, and forcing that name is most of the value. Internal intake also tends to need a tenth field for budget owner, since credits and media spend often come from different places.