AdCP Brief Generator: Paste Prose, Get a get_products Request

Paste your brief. You get back the AdCP get_products request and the AAMP CampaignBrief it produces, with the words that produced each value quoted beside it, which is the parse a seller agent performs against a published schema.

Score a brief. The box arrives holding a worked example, so there is nothing to type.

Nothing leaves the browser. There is no network call and no account. The same brief and the same reference date always produce byte-identical JSON. Relative expressions like “next quarter” resolve against the browser’s today, with no override, so absolute dates are the reproducible option; the result prints the date it used.

Use it

1. Open the generator, paste your brief over the worked example in the box, and press Score this brief or Ctrl/Cmd + Enter from inside the box. Line breaks count as sentence breaks, so a bulleted brief reads the same as a paragraph.
2. Read the two scores: transactable (can a counterparty book this?) and expressive (how many AdCP filters hold a value?). Then read the field table. All 64 concepts get a row, including the ones your brief says nothing about; a toggle hides the empty rows.
3. Work the assumptions list. Each candidate has its own checkbox and prints the rule behind it. The worked example below produces one candidate, currency.
4. Copy either payload with the button above it.

The worked example, scored

The generator is loaded with this paragraph:

Q4 holiday campaign for a CPG snack brand. Channels: CTV plus digital display and online video. Budget: $500,000 net. Market: United States. Target demo: adults 25-54. Flight: November 1 - December 31, 2026. Goal: drive incremental household penetration and retail sales lift. Requirements: premium CTV inventory, brand safety controls, frequency cap 3 per week.

The AdCP get_products request that comes out:

{
  "buying_mode": "brief",
  "brief": "Q4 holiday campaign for a CPG snack brand. Channels: CTV plus digital display and online video. Budget: $500,000 net. Market: United States. Target demo: adults 25-54. Flight: November 1 - December 31, 2026. Goal: drive incremental household penetration and retail sales lift. Requirements: premium CTV inventory, brand safety controls, frequency cap 3 per week.",
  "filters": {
    "start_date": "2026-11-01",
    "end_date": "2026-12-31",
    "countries": ["US"],
    "channels": ["ctv", "display", "olv"]
  }
}

And the AAMP CampaignBrief, with description holding the same paragraph:

{
  "channels": [{ "channel": "CTV" }, { "channel": "DISPLAY" }],
  "target_geo": [{ "geo_type": "COUNTRY", "geo_value": "US" }],
  "flight_start": "2026-11-01",
  "flight_end": "2026-12-31",
  "total_budget": 500000,
  "frequency_cap": { "max_impressions": 3, "period_hours": 168 }
}

That request is the body of a get_products call to a sales agent. The get_products operation has the request and the response it comes back with. Nothing here sends it for you.

That brief scores 5 of 10 transactable and 4 of 30 expressive. Separately, of the 64 concepts in the field table, 4 are stated, 5 derived and 55 missing. The five transactable checks it fails are advertiser, campaign name, objective, currency and the per-channel budget split, each printed with the reason. The four filters it populates are start_date, end_date, countries and channels. A populated filter is a constraint and every one of them deletes products, so the expressive score is coverage: brief vs filters has the routing rule.

Online video reaches AdCP as channels: ["olv"] and reaches AAMP as nothing. AAMP has six channel values, CTV, DISPLAY, AUDIO, NATIVE, DOOH and LINEAR_TV, and 13 AdCP channels have no AAMP equivalent. “adults 25-54” is stated and lands in neither payload, because filters.signal_targeting takes a SignalRef resolved against a signals agent rather than a demographic, and the phrase resolves to no IAB Audience Taxonomy 1.1 identifier for the AAMP side. AAMP takes max_impressions and period_hours. get_products has no frequency field at all, so “3 per week” travels to a seller only as prose inside brief. AdCP does define a frequency cap scope, but it applies to a media buy’s packages, so it cannot influence which products come back.

What you get back

Four rows of the field table, from the worked example:

The ranking has three levels:

The worked example produces 56 ranked gaps: 5 blocks, 1 drops, 50 degrades. The first three, and the one drops:

The 64 concepts are the union, so the count includes concepts only one protocol can express: frequency cap, pacing and approval gates have no AdCP discovery-time home. The completeness rubric has the concept-by-concept map.

The AdCP side is built against release 3.1.13. The AAMP side targets the CampaignBrief in the IAB Tech Lab buyer agent’s src/ad_buyer/models/campaign_brief.py, the typed and validated one; that repository defines a second, incompatible CampaignBrief in models/flow_state.py with no adapter between them, so the output names the model it emitted.

The four labels

A stated field can still miss both payloads, because neither protocol has a field to carry it.

The two scores

A score tells you which of your sentences cannot reach a protocol field.

Transactable, out of ten. Nine of the ten checks are the values a counterparty has to be given before a campaign can be booked, whichever protocol carries it, and AAMP’s CampaignBrief is where they are written down: it will not construct without them. Eight are required fields on the model itself: advertiser, campaign name, objective, total budget, currency, flight start, flight end, and at least one channel. The ninth is budget_pct, which is required on every ChannelAllocation and validated to sum to 100. If you are buying AdCP-only, read the tenth check on its own and treat the other nine as the decisions still open.

The tenth is an AdCP request check: buying_mode present and inside its enum, brief present in brief mode and absent in wholesale and refine, if_pricing_version only alongside if_wholesale_feed_version, the budget_range object contract, the minItems: 1 arrays, the ISO 4217, ISO 3166-1, ISO 3166-2 and date patterns, and the required keys of performance-standard.

That tenth check is a subset of get-products-request.json, not the schema itself. Full JSON Schema validation does not run in your browser, so a payload that clears it here can still fail a seller’s validator on a rule outside the subset.

All ten count what reached the payloads, not what your sentences promised, because a value can be read correctly, labelled stated, and still have no field to travel in.

Expressive, out of thirty. core/product-filters.json carries 30 properties at AdCP 3.1.13: 29 filters, one of them (required_axe_integrations) deprecated in favour of trusted_match, plus ext for vendor keys. The score is how many hold a value. The rest arrive at the seller empty and filter nothing. Two of the 30 are not fields you would ordinarily fill, the deprecated one and ext, so treat 28 as the working ceiling. Every filter you populate is a hard constraint that deletes products before the seller ranks anything.

It will not invent a value

The $500,000 in the worked brief does not reach the AdCP request. filters.budget_range requires an ISO 4217 currency, and $is the sign for the US, Canadian, Australian, New Zealand, Singapore, Hong Kong and Mexican dollars among others, socurrencycomes backmissingand the wholebudget_rangeobject is omitted. USD sits in a separate list with the reason attached. Tick it and both payloads are emitted again,budget_rangecarrying{"currency": "USD", "max": 500000}and labelledassumed everywhere it appears, and transactable moves to 6 of 10.

AAMP makes the gap unavoidable: budget_pct is required on every channel allocation and the values must sum to 100, so a brief that never stated a split cannot transact until a person decides what it is. Write “60/25/15” with three channels and the engine pairs them positionally, says it has done so, and offers the result; the split stays assumed until you tick it.

Writecream took the same paragraph and returned a 60/25/15 channel split, a 14-day view-through window for display and 28 days for CTV, a click-through floor and a completion-rate target, none of it in the input. The completeness rubric has what the other generators returned.

A month named in an exclusion, as in “October is excluded”, is still read as an October flight window. It is labelled derived with the rule printed beside the dates, so you see it before you send.

You cannot verify what the seller did

Nothing in an AdCP response reports what a seller did with your prose. get-products-request.json sets additionalProperties: true, so a misspelled fliters key is accepted as an extension and ignored. A brief sent in wholesale mode passes validation while breaking the rule its own field description states, because that rule was never written as a conditional. The one field that reports narrowing, filter_diagnostics.excluded_by, is optional, counts-only, and keyed by the names of your filters.

Three sales agents in the public AdCP registry answer an anonymous get_products, and the registry sweep has what they returned, including a four-product catalogue that answered a US-only brief with Korean CTV.

Whether a better brief wins better inventory is untested.

Where to go next

• What a protocol-conformant brief actually is: the brief is one string typed {"type": "string"} and gated by buying_mode. Three payloads there pass a schema validator without being conformant, and the official SDK is looser than the schema it was generated from.
• The brief completeness rubric: the field map the scores implement, concept by concept, including the cells left blank because no mapping exists.
• The pre-flight checklist: the decisions to settle with the people who have to agree, before any of this is a JSON problem.