Did a request fail with “Request declined by platform policy.” or did a
media buy or creative stop at Platform review? That is the platform safety
policy, not your Brief Acceptance policy. See
Request declined by platform policy.
- Open the portable AI Business Rules page from Apostra, Claude, ChatGPT, or another MCP Apps host to create or edit the full policy inline.
- Tell Apostra the exact rules you want. Apostra shows you the complete markdown for confirmation, then writes it through the same typed configuration contract.
VALIDATION_ERROR. To stop applying rules,
turn a policy off instead (see Turn a policy off). If a
save cannot complete, the
Page keeps the draft. The Page shows the server’s safe reason and recommended
action when available. When the save never reached the server, the Page says
so; otherwise it uses generic guidance. Every save attempt
has a correlation ID that remains visible for support even if the host loses
the server response.
You don’t pick from a fixed menu of categories. You describe your rules, and the
agent reads what you wrote. Every save creates an immutable version and can
make it live at once, so version history never contains a half-applied update.
Edit and restore versions
Brief Acceptance and Creative Policy are two policies, and each keeps its own version history. Saving new text creates a new version of that policy without overwriting the old one, and makes it live unless you save it as a draft. Each policy has at most one draft. You can make any earlier version live again, and a policy can be turned off, with no version live. The AI Business Rules page shows each policy’s live version, its draft and its history, and performs these operations directly; it does not turn a button click into a new chat prompt or depend on a particular host model. Version numbers carry over from before the two policies were kept apart: version 7 of Brief Acceptance is the text that was live as version 7 before. Later numbers can differ between the two policies, and Creative Policy can skip a number where an older version had no Creative Policy.Save each policy from an agent
save_business_rules changes each policy on its own under policies, with
briefAcceptance, creativePolicy or both. For each policy, send one of:
text, with optionalnotessaying why. The new version is live at once. Withpublish: falseit is saved as the policy’s draft instead, and nothing that applies to buyers changes.publishVersion, naming an existing version to make live. Naming the current draft publishes it and clears the draft. Naming an earlier version makes it live again and leaves an open draft open.turnOff: true, alone, to make no version of the policy live (see Turn a policy off).
publishVersion, need expectedDraftVersion: the
policy’s current draft version, or null when it has none. If another
administrator changed the draft in the meantime, the save is refused with
CONFLICT and reason draft_changed, and nothing is written, so neither of you
overwrites the other’s draft. Notes are saved with each version, so text
identical to the live version (or, with publish: false, to the current draft)
saves a new version when notes differ from that version’s. With the same
notes, or none, it writes nothing.
Every change in one call happens together or not at all. The answer’s
policyChanges says, for each policy you named, what happened (published,
draft_saved, draft_published, reactivated, turned_off or unchanged),
the version written or made live, and the version that was live before.
Publishing new text while a draft is open leaves the draft open, and the answer
says so.
A call can also set the approval settings and the listing’s Advertising Policies.
These are saved after the policies. If they cannot be saved once the policies
were, the call answers with an error that says the policy changes are already
saved, and carries policySaved: true; retry with only the settings. If the
call moved creative approval to manual, the settings are saved first, and if the
policies then cannot be saved, the error says the settings are already in place
and carries settingsUpdated: true; retry with only the policies.
The older policy input, which takes the whole document, still works. Each
section saves its own policy, a section identical to the live text writes
nothing, and an empty section turns that policy off. Its createdVersion and replacedVersion name the Brief Acceptance
version, and are left out when the call wrote no Brief Acceptance version.
Turn a policy off
You can turn Brief Acceptance, Creative Policy, or both off. A policy that is off has no live version, so nothing checks briefs, buys or creatives against it. Your version history and any draft stay as they are. Turning rules off is a decision, not missing setup: your storefront still counts as set up, readiness stays complete, and both approval settings stay as you set them. While a policy is off:
Briefs that need Brief Acceptance rules, such as a sandbox brief or a brief on a
plan that charges for brief responses, still count Brief Acceptance that is off
as meeting the requirement.
From an agent, send
turnOff: true for each policy in save_business_rules
policies, with nothing else for that policy. The answer says what changes.
Turning Creative Policy off while creative approval is automatic, or in a call
that sets it to automatic, needs acknowledgeNoHumanReview: true, because from
then on no person sees new creatives. Nothing is written when:
- the caller is not an account administrator (
ACCOUNT_ADMIN_REQUIRED); - your storefront has never had live rules and the call makes none live
(
VALIDATION_ERROR, reasonno_rules_to_turn_off): there is nothing to turn off, and setup still needs rules. A call that publishes one policy and turns the other off is accepted, and turning off a policy that was never published answersunchanged; turnOffcomes with any other field for the same policy, or isfalse(VALIDATION_ERROR; to turn a policy back on, usepublishVersion);- creative approval is automatic and the acknowledgement is missing.
unchanged.
Turning back on. Make a version live with publishVersion, usually the
version that was live when you turned it off, or publish new text. Saving a draft
leaves the policy off.
Reading a policy that is off. In get with kind: "business_rules", a policy
that is off has no live version and a turnedOff object: when it was turned
off, who did it, and lastLiveVersion, the version that was live. Who did it is
a person’s name, or the kind of credential for an agent, API key or service; it
is never an email address, and is left empty when there is no name safe to show.
turnedOff is null while a version is live, and for a policy that has never
had one. A policy that was already off before turn-offs were recorded has a
turnedOff with no time, no name and no version.
Test a brief against your rules
Before you rely on a rule, you can see how it treats a real buyer brief. On the AI Business Rules page, Test a brief lists the briefs your storefront received, taken from your Sessions: your latest 25 discovery briefs and 25 RFPs, most recently active first, each with its advertiser and its last activity. Pick one and run the test. A brief received before your storefront kept a copy of it can’t be tested; picking it says so. To test an older brief that isn’t in the list, choose Paste a request instead and paste the buyer’s AdCPget_products request
(JSON); Use an example fills in a small one to start from. On the main
view the test uses your active policy; while you edit, it uses the Brief
Acceptance rules you have not saved yet, so you can change a rule, test it,
and then save. Any member of your seller account can run a test. After a test,
All brief fields shows the complete request that was tested.
Each test shows two answers side by side:
- Production is what product discovery does with that brief today: whether it returns products, the reason the check gave, and the rule a decline quoted. The discovery check reads only the first 4,000 characters of the brief’s text, and it can decline only on Brief Acceptance rules of up to 4,000 characters or 80 rules; the result says when either limit applied. When discovery does not run the check, the result says so instead of guessing. That covers a Listing storefront, a storefront that answers through a sales adapter, and a wholesale or refine request.
- Beta is the answer from Cloudflare Clef, the evaluator that may replace today’s check: Pass, Needs review, Does not pass, or Inconclusive. Clef reads the complete brief, up to 128 KiB, against Brief Acceptance rules of up to 12,000 characters or 80 rules. It gives an answer and its confidence, never a written reason. An inconclusive result is never a pass.
Policy vs. approval mode
These are two separate dials. Keeping them straight is the most common point of confusion:
Brief Acceptance has two enforcement points. During product discovery, on a
storefront that composes its own products, a confident conflict with the
seller’s written policy returns no products; an unavailable or inconclusive
qualifier fails open so discovery remains available. A Listing storefront does
not run this discovery check. At
create_media_buy, manual mode evaluates the submitted buy
again: clearly on-policy buys auto-forward, while ambiguous results, evaluator
errors, and deterministic hard findings queue for an operator. Live
auto-rejection requires a separate seller opt-in or versioned rollout. auto
mode preserves the established auto-forward contract and does not add a second
create-time evaluator gate.
Turning a policy off changes neither approval setting. Under manual, it sends
more work to a person, never less: a buy your rules would have cleared
automatically now waits for review. See Turn a policy off.
How the agent reads your policy
The manual create-time pre-screen reads your written policy against the buyer’s submission. It is not a fixed category list — a rule like “we don’t accept fast-food advertisers” is honored even though “fast food” isn’t a built-in category, because the agent derives the subjects you wrote and matches them against the submission. It also recognizes common evasion (look-alike characters, light obfuscation) so a rule isn’t sidestepped by spelling. When it cannot judge a submission with confidence, it errs toward holding it for you. When no policy document is available, the AI Business Rules Page starts with a conservative draft: no category is pre-cleared, every advertiser, category, and creative needs review, and illegal, malicious, deceptive, discriminatory, exploitative, phishing, impersonation, or destination-concealing submissions are flagged for operator rejection. This draft has no effect until an account administrator reviews, saves, and activates it. While that starter remains active, its review-all contract also makes the deterministic pre-screen hold every submission for an operator; it is not just instructional copy.The three verdicts
The manual-mode create-time pre-screen sorts a managed-storefront media buy into one of three buckets:- Clearly on policy → eligible and auto-forwarded without operator review.
- Needs a look → queued for an eligibility decision. Evidence is thin, the policy is ambiguous, or a rule says this category needs review.
- Clearly against an explicit rule you wrote → queued with a strong operator-facing rejection signal, without exposing private policy text to the buyer.
No create-time verdict auto-rejects a buy in the current rollout. Deterministic
matches can block product discovery and supply a strong queue signal; an
AI-judged conflict can only escalate or fail closed.
The second opinion
On top of the deterministic manual-mode pre-screen, an AI layer gives a second opinion. It can only downgrade a proposed clear result to “needs a look” if it spots a policy conflict the rules missed (for example, a competitive- exclusivity clause you wrote in prose). It can never approve a buy on its own, never reject one, and never overrides a hold. It also fails closed within the manual approval path: if the AI layer is unavailable, errors, or is uncertain, the buy stays with the operator rather than being cleared by the evaluator.How creatives are evaluated against your policy
Creative review applies the same content rules to a buyer’s submitted creative, plus a rendered-image check that looks at the actual pixels for brand cues, prohibited categories, and third-party marks. As with media buys, the AI layer can route a creative to human review but cannot approve one over a clear policy violation the policy check already found. That check reads the text of the creative against your Creative Policy in any language, and counts a violation only when it clearly breaks a rule you wrote. If it cannot decide or does not finish, the creative goes to a person.Automatic approval
Every storefront on automatic creative approval (creativeApproval: auto) that
has a written Creative Policy gets an automated reviewer. It approves a creative
when it is definite and every check passes. It rejects a creative only when a
separate policy check confirms that the creative breaks a rule you wrote.
Otherwise it leaves the creative in your review queue for a person, so a creative
it is unsure about waits there instead of going straight to your sources.
Apostra can switch the reviewer off for an individual account. On an account
where it is off, creatives on automatic approval go straight through and the
Creative Policy is not applied to them.
A storefront on automatic approval with no written Creative Policy is unchanged:
its creatives go straight through, because the reviewer needs a policy to judge
against. A storefront on manual approval keeps every creative in the queue.
So a save that leaves Creative Policy empty or turns it off while creative
approval is automatic, or that sets creative approval to automatic in the same
call, asks you to confirm that nobody will review new creatives. Creatives
already waiting when Creative Policy is turned off stay with a person: the
reviewer reads the policy again before it decides, finds none, and leaves them in
your queue. In
save_business_rules, send acknowledgeNoHumanReview: true with the save; on
the AI Business Rules Page, tick the confirmation the Page shows and save again.
Without it the save is refused and nothing changes. A save_business_rules
call that also sets creative approval to manual needs no confirmation, because
a person then reviews new creatives. An earlier version with no
Creative Policy cannot be made active again while creative approval is
automatic: save it again as a new version and confirm, add Creative Policy
rules, or set creative approval to manual first.
For locked assets — a product shot, logo, or book cover you’ve marked as
exact-fidelity — there’s a stronger, deterministic guarantee that the AI layer
cannot override:
- A creative whose final render goes through the platform’s locked-asset compositor preserves your supplied asset exactly, and clears that check. The compositor places your real image where the generated scene drew the product, at the same angle and perspective. When it cannot find the product, or the scene covers part of it, the creative is held back instead.
- A final-stage creative that carries your locked asset but did not go through that deterministic path — for example, one that re-generated or edited your product from a reference — is blocked from auto-finalizing and held back for a human to inspect. A buyer’s generated creative cannot quietly substitute a look-alike stand-in for your locked product at final approval.
What the agent never does on its own
- It never hard-rejects a submitted media buy in the current rollout. Deterministic Brief Acceptance conflicts can block product discovery and supply a strong operator-facing queue signal; AI judgments can only escalate.
- It never changes an approval gate for you.
- Within manual create-time evaluation, it never promotes an uncertain result to eligible. If the AI layer can’t corroborate, the buy stays with the operator. Automatic storefronts do not run this second create-time evaluator.
Request declined by platform policy
Apostra checks some requests against a platform safety policy before it acts on them. The check reads text in any language. It is separate from Brief Acceptance, the policy a seller writes for their own storefront.Apostra is switching this check on gradually, one place and one account at a
time. Until it is on for your account, nothing in this section changes what you
see. This page describes how it behaves once it is on.
What you see at each step
Discovery and campaign writes
When you send a brief to product discovery, or create or update a campaign with a brief or name that the check declines, the request fails withINVALID_REQUEST and the same message every time. Over REST the HTTP status
is 422. Through MCP the result has isError: true and the error is in
structuredContent.adcp_error.
details.reason equal to safety_refused, not for the message text.
The check applies to every way of reaching these operations, including the REST
API, the MCP tools, and Apostra’s own assistant.
- Discovery returns no products. The brief is not used to look for products.
- A campaign write saves nothing. The campaign is not created or changed.
- A refusal can also appear while Apostra writes results or a proposal for your brief. In discovery it looks the same as above. For a proposal, no pitch is produced. If your brief caused it, support can clear it for that brief. If the text Apostra generated from your brief caused it, nothing clears it: revise the brief and retry.
SERVICE_UNAVAILABLE (HTTP 503). For discovery, details.reason is
safety_check_unavailable; for a campaign write, the message is “Campaign
moderation is temporarily unavailable”. Apostra could not run the check. That is
not a finding about your content, and nothing is recorded for support to review.
Retry the same request after a short wait. If it keeps failing, contact support.
While Apostra writes results or a proposal for your brief, an outage also stops
generation, and nothing is recorded for support to review.
A separate 422 with the code CONTENT_MODERATION_BLOCKED comes from Apostra’s
structural content checks. It is a different refusal with a different cause.
Media-buy approval
On a storefront that reviews media buys manually, a buy whose text the check holds is not forwarded automatically and is not rejected. It waits in the media-buy approval queue for a person. A seller sees one check on the buy:- Label: Platform review
- Status: warn
- Detail: “This media buy requires manual review before it can be approved.”
- Recommendation: escalate to a person
PENDING_APPROVAL, the same as any buy waiting for review, and
gets no refusal. Brief Acceptance and the AI second opinion do not judge a held buy.
If Apostra cannot run the check, the buy is held in the same way.
Creative review
When a seller asks for an evaluation of a creative and its text is held by the check, the evaluation returns the same Platform review check: status warn, “This creative requires manual review before it can be approved.”, with manual review recommended. Apostra does not approve or reject the creative on the strength of it. The seller decides in the creative review queue. The check does not say what triggered it, and rendered-image evaluation never turns it into an automatic approval. If Apostra cannot run the check, the evaluation is held in the same way.What to do about a refusal
- Revise and retry. Remove or reword the part of the brief, name, media buy, or creative text you think could be read as unsafe, then send it again. Apostra does not tell you which part.
- Ask for a review if you think it is wrong. Email support@apostra.com with your account name or ID, the time of the request, and the error code and message. Include the IDs of any campaign, media buy, or creative involved. Do not paste the content itself into the email.
- Wait for an answer. An Apostra staff member reviews the refusal. If it was wrong, staff can clear it for that exact text, and your next attempt goes through. The clearance can have an end date.
Prepare inventory source inputs
Request creative policy, approval ownership, and lifecycle evidence together.
Approvals overview
The two settings and the lowest-risk posture.
Media-buy approvals
The queue and the auto-clear vs. escalate decision.
Creative reviews
How submitted creatives move through review.