Source Influence

curl -X POST https://api.querying.ai/v1/source-influence \
  -H "Authorization: Bearer $QUERYING_API_KEY" -H "Content-Type: application/json" \
  -d '{
    "answer": "…the AI answer, as raw markdown…",
    "prompt": "best password manager for a small team",
    "engine": "CHATGPT",
    "brand": "1Password",
    "competitors": ["Bitwarden"],
    "citations": [
      { "url": "https://example.com/best-password-managers", "body": "…the page markdown…" },
      { "url": "https://another.example.com/pricing" }
    ]
  }'

​

Analyze a task you already ran

curl -X POST https://api.querying.ai/v1/source-influence \
  -H "Authorization: Bearer $QUERYING_API_KEY" -H "Content-Type: application/json" \
  -d '{ "taskId": "7c1f…", "brand": "1Password", "competitors": ["Bitwarden"] }'

​

Analyze every answer as you collect it

curl -X POST https://api.querying.ai/v1/async/task \
  -H "Authorization: Bearer $QUERYING_API_KEY" -H "Content-Type: application/json" \
  -d '{
    "taskType": "CHATGPT",
    "payload": { "prompt": "best password manager for a small team", "country": "US", "include": { "sourceInfluence": true } },
    "webhook": { "url": "https://example.com/hook" }
  }'

• Webhook: sent once, after the analysis finishes, with the answer and the analysis together.
• Polling: the answer task is COMPLETED as soon as the answer is ready. Keep polling until response.sourceInfluence.task.status is COMPLETED or FAILED.
• Batches complete when every answer and its analysis have finished.
• No analysis: when the answer cites no URL to read, or your credits or the key’s daily limit cannot cover the analysis hold, response.sourceInfluence is { "error": "…" } and the answer is delivered right away.
• Billing: the analysis is charged on its own, by usage like any Source Influence task. The answer’s charge does not change. A key limited to certain task types must also allow SOURCE_INFLUENCE.

​

Payload

​

Result

{
  "answer": "Teams plan starts at $19.95 per month for up to 10 users. The Business plan adds audit logs.",
  "sources": [
    {
      "id": "s1",
      "url": "https://example.com/pricing",
      "paragraphs": [
        { "id": "p1", "text": "Teams: $19.95 / month, up to 10 users." },
        { "id": "p2", "text": "Business adds audit logs and SSO." }
      ]
    },
    {
      "id": "s2",
      "url": "https://review.example.com/note",
      "paragraphs": [],
      "error": "…"
    }
  ],
  "links": [
    { "id": "l1", "paragraphId": "p1", "answerRanges": [{ "start": 0, "end": 57 }], "explanation": "States the same price and seat cap." },
    { "id": "l2", "paragraphId": "p2", "answerRanges": [{ "start": 58, "end": 92 }], "explanation": "Names the audit-log add-on." }
  ],
  "insights": [
    { "summary": "The pricing page covers both statements; the review page could not be read.", "linkIds": ["l1", "l2"] }
  ]
}

​

answer

​

sources[] — one entry per citation you sent

• id — opaque identifier, unique within the result.
• url — the citation URL you supplied.
• paragraphs[] — the paragraphs of that page included in the result. Each has an id (globally unique across the whole result) and the text of the paragraph. A paragraph can be included with no link at all.
• error — present only when the page could not be read. paragraphs is then empty, and the string is a short reason for that source.

​

links[] — one relationship per paragraph

• paragraphId — the paragraph this link is about. It always points at a paragraph that exists in sources[].
• answerRanges[] — the parts of the answer this paragraph corresponds to, as start/end pairs in the same UTF-16 offsets as above. A link may carry several ranges; the same answer text is not repeated across them. Each range is a whole answer unit (sentence, list item or table cell).
• explanation — a short sentence describing the correspondence: what the paragraph states about that part of the answer.

​

insights[]

​

Reading the result correctly

• No link is not a verdict. A paragraph without a link simply did not match an answer range, and an answer part with no range means no paragraph was linked to it. Neither says the page is irrelevant, unused or untrustworthy.
• An unread page stays uncertain. When sources[].error is set, that citation was excluded from the analysis; do not treat its lack of links as a negative finding. Retry it or supply its body if it matters.
• Correspondence, not causation. The result describes how an existing answer and existing page text line up after the fact. It does not measure generation influence, rank pages against each other, score the page, or report how the answer was produced.
• Ids are internal. Do not persist p…/l… identifiers as stable keys across runs; they are scoped to a single result.

​

Execution and limits

• Size is not a failure mode. Anything the request contract accepts — an answer up to 200,000 characters with up to 100 citations — is analyzed. Long answers, large source sets and very long pages are handled by splitting the work, not by rejecting the task.
• Depth scales with the answer, not the page. A long page does not produce more links than the answer can support: the analysis concentrates on the passages most likely to correspond to the answer, and every source that shows a correspondence is represented. On a large page, or on a page that repeats the same statement in navigation, listings and reviews, expect the strongest passages rather than every occurrence.
• Markup is not evidence. Sitemaps, link indexes and image galleries contribute no links; a page made only of those is reported with no paragraphs rather than with invented correspondence.
• A source that could not be read is reported in sources[].error without failing the task if another source is readable. The task fails when no source could be read, on an analysis failure, on a timeout, or on a service configuration error.
• Requests are metered by usage, like the rest of the API. The submit response shows the temporary credit hold; the final charge is settled when the task reaches a terminal state.