Audienzes Pro

List the concept universes available for profiling. A domain is a named candidate universe of concepts — e.g. sports, lifestyle. It matters beyond discovery: `domain_index` in profile_segment is scored against the mean of the domain's universe, so it is SET-DEPENDENT. Adding a 21st sport changes every score with no new measurement. Each domain therefore reports a `universe_version` that any result quoting a domain_index carries. Returns: {"ok": true, "domains": [{"domain", "universe_version", "size", ...}], "count": n}. The domains feed resolve_concepts and profile_segment.

list_domains

Covers only the user's OWN connected ad accounts and live campaigns. Audience questions (who an audience is, what it is into, whether or where an audience can be tested or bought) are covered by scope_market, profile_segment and compare_audiences. List live campaigns for one connected ad account. account_id values come from list_ad_accounts. Returns each campaign's name, status and objective, up to 100 campaigns per call; next_page or next_cursor is set when more campaigns exist. Campaign status is not a performance measure. A missing or expired connection returns ok false with a connect_url for that platform.

list_campaigns

Competitor positioning, read by a senior brand strategist: who says what to whom, where the category is crowded, where it is open, and what to do. Reads every advertiser's ads across ad libraries (copy, and the words in the picture when an ad has no copy), extracts each claim, and returns: - answer_markdown: the finished reading, written in English. It opens with the verdict, then a message matrix (brands x functional claims, emotional territory, proof points, audience addressed, tone, offers and calls to action, creative format, share of observed ads), each brand's apparent strategy, crowded and open territory, threats and opportunities, two or three recommended positioning moves with rationale, first test and KPI, the basis of the reading, and the questions to ask next. - strategist: the same reading as data; every line cites ads (A1, A2 ...) whose preview pages are in `ads`. - positioning_assertions and scope: the evidence, for compare_positioning and brief. What it is for. A request to see competitors' ads is usually a means to a decision: a positioning, a campaign brief, a pitch, or a claim to own. A list of ads is not the answer; the answer is the reading and its move. `for_brand` (the user's own brand) and `objective` (e.g. "find a territory to own for a spring launch") write the moves for that brand and goal. Args: market: ISO-2 market, e.g. "DK" ("Denmark" works too). advertisers: at least two, up to twelve, all in one call. The map compares them as one set. Past twelve, the first twelve run and the rest are named. platforms: ad libraries to search. Omit for all. for_brand: the user's brand, if they are one of the advertisers or competing with them. Optional. category: what is being compared, e.g. "running shoes". Optional; ads for other categories are read as context. objective: what the user wants to decide. Optional. country / age_min / age_max / genders: an optional audience to read the claims against. background: true returns a handle that get_result resolves. Measured 10/10: the first ads are back after about 3 minutes, the full reading after about 10. creatives: "with" (default) also brings the ads' pictures for the live display; "without" skips them.

map_positioning

What this audience actually looks up. What it is for: the category side of a positioning, i.e. what people in the market search for around the category, in their own words. Seeds are PLANNED, not built from the brand name. Seeding on a brand measures the brand's own search footprint, which says nothing about these people. `include_brand=true` is available when the brand's own footprint is the question. `query_plan` goes with the results: the expansion starts from OUR seeds, so a keyword absent from the output was very likely not asked for. Unasked is not unsearched. Search volume is EVENTS, not people, and the people searching are not necessarily your segment. High volume mixes research, comparison, support and idle curiosity, and is not intent to buy. Each keyword's `intent` is OUR keyword-rule label (English trigger words), not a measure from the source: a query our rule cannot place (another language, or no trigger word) has intent null, which is unknown, and `intent_breakdown` counts the placed rows only (see `intent_coverage`). `intent_breakdown` holds shares of the keyword rows (1 means every row carries that label), not counts. The figures come back at once; the strategist's reading is written afterwards. A reply with an `advice` block has final figures and a reading still being written; get_result with the job_id in `advice.next` returns it, and `advice` is gone and `answer_markdown` is present once it is finished.

map_search_intent

Where this audience gathers, and what is actually said there. Defaults to background=true, unlike every other tool here: measured 2026-09-22, a listening run exceeded 600 seconds. get_result returns it. `sample_posts` are OBSERVED and UNREPRESENTATIVE. A retrieved post is evidence a thing was said. It is not evidence of how widely it is held, and a vivid quote is the easiest thing in this system to over-read. `sentiment_balance` is conservative by design. Without post text at volume a sentiment score is a guess wearing a number. A community's subscriber count is not your audience - members are not necessarily in this market, let alone in this segment. An empty theme list means our sources returned nothing usable, which is a retrieval fact, not an audience with no interests. The figures come back at once; the strategist's reading is written afterwards. A reply with an `advice` block has final figures and a reading still being written; get_result with the job_id in `advice.next` returns it, and `advice` is gone and `answer_markdown` is present once it is finished.

map_listening

Which of these concepts are moving, on the sources that could see them. What it is for: the timing side of a positioning or campaign, i.e. which parts of a category are gaining attention now. `rising_per_source[src].relevance` qualifies every figure. Sources split into two kinds and they are NOT the same evidence: seeded the source was asked about your concepts, so the trends are about this audience market_wide the source returned whatever is trending in the market. Real trends, but not filtered to your concepts or your audience - context only Measured for a Danish women 18-34 snowboarding/yoga query: Google returned "Nike" and "Nike Yoga"; X returned "Belarus". Both are true. Only one is about the audience. A source listed under `sources_that_saw_nothing` did not look - wrong market, rate limit, or no usable seed. That is not evidence the concept is static. `category_temperature` is a coarse read of whether the category is moving at all, not a market growth rate. The figures come back at once; the strategist's reading is written afterwards. A reply with an `advice` block has final figures and a reading still being written; get_result with the job_id in `advice.next` returns it, and `advice` is gone and `answer_markdown` is present once it is finished.

map_trends

Profile a segment against a set of concepts, across platforms. What it is for: the audience half of a positioning or a campaign. It shows what a segment cares about, which a map of advertisers' claims cannot. The analytical call. "What kind of sports are women 18-25 in Denmark interested in?" is country="DK", age_min=18, age_max=25, genders=["female"], domain="sports". CROSS-PLATFORM IS THE DEFAULT. Omit `platforms` and every platform that can answer honestly is measured. `platforms` NARROWS that set; it does not enable a platform. No head-counts are returned. Percentages and indices are the product. Three metrics, three names, and only one may be called affinity: - `penetration_pct` — the share of the segment on that platform that shows the concept. Descriptive, and NOT comparable across platforms: the platforms see different populations and each rounds differently. It is not an average across platforms. - `domain_index` — concept penetration vs the mean of the candidate universe. SET-DEPENDENT: it moves when the universe changes, with no new measurement. Only meaningful alongside the result's `universe_version`. - `affinity_index` — [segment ∩ concept / segment] / [country ∩ concept / country] × 100, same platform. 100 = PARITY with the country's adults; above 100 over-indexes, below 100 under-indexes. The definition is identical on every platform; the measurement error is not. Args: country: ISO-2 market, e.g. "DK". Required — every benchmark is same-country, same-platform. age_min: Inclusive lower age bound. Omit for all ages. age_max: Inclusive upper age bound. Omit for all ages. genders: ["female"] | ["male"] | omit for all. Platform-reported gender. theme: Free text such as "winter sports" or "sustainable fashion". The universe is built on request. It covers themes no curated domain fits, which is the normal path for strategic planning. Rows built this way are flagged in `notes`. domain: Profile a whole universe from list_domains (e.g. "sports"). concepts: Explicit canonical concepts instead of, or narrowing, a domain. resolve_concepts maps loose wording to them. platforms: Narrow the platform set, e.g. ["meta", "tiktok"]. At most 15 known platforms; unknown names are rejected. max_concepts: Cap on concepts measured (default 20). Each concept costs measurements on every platform. At most 25. for_brand / objective: whose decision this serves, in the user's own words (the user's question works well). The expert reading is written for it. industries / company_sizes / job_functions / seniorities: a B2B segment, in plain words ("construction", "ejer", "marketing") or LinkedIn URNs; company_sizes like ["11-200"]. Only LinkedIn can express them: every other platform answers `segment_unsupported`, which is not 0. The reply is the segment's share of the country's LinkedIn members; concepts are not measured on a B2B segment yet. A name that matches nothing returns `invalid_input` with the names listed. Returns a ProfileResult: `rows` (one ConceptRow per concept, ranked, each with `consensus_percentile`, `direction`, `agreement`, `evidence_grade`, `caveats` and the per-platform `platform_evidence` it was synthesised from), plus `platform_status`, `effective_segment_by_platform`, `universe_version`, `concepts_omitted`, `notes` and `partial`. `platform_status` and `effective_segment_by_platform` describe how far each number can be read. They are the honesty of the result, not boilerplate: - A platform cell is neither 0 nor []. It is `measured`, `suppressed` (measured but the metric is an artifact — e.g. a floor-pinned audience), `concept_unavailable`, `segment_unsupported`, `different_lens` (audience scale or search intent, not affinity), `source_unavailable`, `measurement_unreliable`, `rate_limited` or `failed`. - ABSENCE IS NOT ZERO. "Reddit cannot express an age segment" is not "young Danes are not on Reddit for sport". - `effective_segment_by_platform` reports what each platform could actually express. 18-25 is an unlucky boundary: Meta is exact, most others group ages as 18-24 and carry `age_distortion_years: -1`. A platform's number belongs to its effective segment, not the requested one. - `partial: true` means the consensus is held back or thin. The figures come back at once; the strategist's reading is written afterwards. A reply with an `advice` block has final figures and a reading still being written; get_result with the job_id in `advice.next` returns it, and `advice` is gone and `answer_markdown` is present once it is finished.

profile_segment

Turn loose language into canonical concepts, before spending any measurements. It maps free wording to concepts that platforms can express: "skiing" resolves to the skiing concepts, and a compound such as "skisport" (ski + sport) resolves to every concept carrying "ski", marked `compound`. The result says which platforms can carry each concept. Args: query: Free text — a concept, a synonym, a foreign-language term, or a broad ask such as "sports". Omit to list a whole domain's universe. domain: Restrict to one domain from list_domains (e.g. "sports"). Omit to search every domain. Returns: {"ok": true, "matches": [ConceptMatch], "count": n}, where each match has `concept` (canonical name), `domain`, `match` (exact | alias | fuzzy | compound | ambiguous | unresolved), `platforms_available`, `platform_availability` (per platform: available | not offered) and `suggestions`. An `ambiguous` or `unresolved` match is a real answer, not an error: it is returned with suggestions instead of a silent fuzzy match to something else. An unresolved concept has no measurement in profile_segment. `platforms_available` is a capability list, not a result — a platform missing from it will come back `concept_unavailable`, which is not zero interest. An `unresolved` phrase is not a dead end: pass it as `theme` to profile_segment, which builds a concept universe for it on request. A Danish word can be unresolved where the English term resolves (løbesko: Running shoes).

resolve_concepts

Is there a credible, executable way to TEST this audience in this market? It answers "can we test / run / buy this audience, and where?" and needs nothing connected. NOT "is this audience good?" - addressability is not desirability. A broad cheap audience may be strategically useless; a niche expensive one may be exactly right. This is an adversarial check on whether an audience hypothesis survives contact with buying reality. It applies early, before a strategy rests on something commercially fictional, and again after profile_segment or compare_audiences, because richer analysis changes the definition and therefore its executability. `verdict` is CATEGORICAL, not a number: testable | testable_with_material_constraints | not_presently_testable | indeterminate `indeterminate` means evidence is missing, NOT that feasibility is poor. `evaluated_audience` comes before the verdict. Each source states the requested definition, the definition it can actually execute, and whether the difference is acceptable, material or disqualifying. The worst way to misread this tool is to take a low cost and a wide band from a source that cannot express the audience that was asked about - every fact correct, the conclusion wrong, because the measured object changed. Environments are grouped by `audience_role`. Search intent, household audiences and demographic-interest audiences are NOT interchangeable inventory because they share a theme; they may be compared only inside one role. Cost: a class, not a budget. CPM x impressions estimates exposure spend, not the cost of learning, so a range appears only when `test_spec` supplies a real specification, and is then labelled "estimated media required under stated test assumptions", which is not a recommended budget. `interpretation_contract` states machine-readable permitted and prohibited inferences. Notably: inventory cost may price inventory and may NOT support CPA or ROAS claims; concentration may describe where people are and may NOT support response propensity. Deliberately absent, because they are too quotable and erase the conditions underneath: feasibility_score, best_platform, cheapest_audience, recommended_budget. Args: country: ISO-2 market. age_min / age_max / genders: the audience hypothesis. concepts / theme: optional territory. Without one, it runs in provisional scoping mode and says so. platforms: narrows the sources considered. test_spec: a real test specification, if you have one. industries / company_sizes / job_functions / seniorities: a B2B segment (plain words or LinkedIn URNs; company_sizes like ["11-200"]). Only LinkedIn can run it; other platforms answer `segment_unsupported`. for_brand / objective: whose decision this serves, in the user's own words (the user's question works well). The expert reading is written for it. The figures come back at once; the strategist's reading is written afterwards. A reply with an `advice` block has final figures and a reading still being written; get_result with the job_id in `advice.next` returns it, and `advice` is gone and `answer_markdown` is present once it is finished.

scope_market

Write a behavioural brief for a segment, directly from the segment and concepts. What it is for: the last step of a positioning or targeting analysis, turning measured evidence into a creative brief a team can work from. Given a segment and concepts, brief measures the audience itself (profile_segment, then analyze_overlap), so no earlier tool results are needed. "Brief for Danish men 18-24 into snowboarding and cycling" is country="DK", age_min=18, age_max=24, genders=["male"], concepts=["snowboarding", "cycling"]. It runs as a background job; get_result returns it. `advertisers` (two or more) add advertiser positioning to the brief: map_positioning, about 10 minutes for seven. A company or job-role audience (a title such as CISO or IT-chef, a company-size range, B2B) is answered at once with status `company_audience` and a pointer to Audienzes B2B; nothing is measured. Doubtful wording goes through as a consumer brief. Instead, `evidence` already gathered, keyed by tool: {"profile": <profile_segment>, "overlap": <analyze_overlap>, "positioning": <map_positioning>, "scope": <scope_market>, "trends": <map_trends>, "listening": <map_listening>} This does NOT summarise. It asks for statements, each carrying the fact it came from, then checks every citation. A statement whose citation does not resolve is DROPPED - it was written from the model's priors rather than from your evidence - and dropped rather than flagged, because a caveat is something a reader can skip. What comes back is a BEHAVIOURAL ARCHETYPE: an addressable group characterised by measured signals. There is no name, no quote, no motivation and no day in the life, because advertising and targeting data cannot evidence any of them. Statements using that language are refused even when correctly cited - a citation does not make a feeling measurable. SET ASIDE 12/10 on David's word (the expert layer): with evidence, the reply is a creative brief (answer_markdown first) whose audience insight and tone are labelled as the strategist's reading, each reason to believe citing the concept and index or ad refs it rests on. Names, portraits and quotes stay out. for_brand / objective (the user's question) set the decision it is written for. background=true still returns the cited-facts brief. Thin evidence returns `insufficient_evidence` rather than fewer, vaguer sentences. That is not a finding that there is little to say; it means less was gathered than a brief needs. `statements_dropped` is the most informative part of the output, because it lists what the evidence does not support.

brief