How to automate WhatsApp reports for clients with AI

Three steps. First, the script calls the API and gathers facts: the state of each instance, scheduled and completed campaigns with their counters, conversations with unread messages. Second, those facts go into a prompt that asks for a short summary, without adding any number. Third, the text goes to human review and then to the client.

The critical point is the separation. Language models get math wrong and invent context. If every number in the report comes from the script and the AI only writes it up, the worst error becomes a badly worded sentence, not a nonexistent sale in the report.

The script below lists the account's instances, and for each one reads the campaigns (GET sender/listfolders, which returns status and counters like log_total, log_delivered, log_read and log_failed) and the conversations with unread messages (POST chat/find, which returns chats with jid, name and unreadCount). Keep the key and secret in environment variables.

The exact fields of GET /instances are in the reference; the example uses id and name and should be adjusted to what your account's response returns. Sales results and ad lead results come from the portal and the Performance tab, or from your own system, not from this script.

const BASE = "https://api.zapzapapi.com/api/v1";
const headers = {
  "Content-Type": "application/json",
  "x-api-key": process.env.ZAPZAP_API_KEY,
  "x-api-secret": process.env.ZAPZAP_API_SECRET,
};

async function call(path, init = {}) {
  const r = await fetch(BASE + path, { headers, ...init });
  if (!r.ok) throw new Error(path + " -> HTTP " + r.status);
  return r.json();
}

const instances = await call("/instances");           // array of instances
const facts = [];
for (const inst of instances) {
  const folders = await call("/" + inst.id + "/sender/listfolders");   // campaigns with counters
  const chats = await call("/" + inst.id + "/chat/find", {
    method: "POST",
    body: JSON.stringify({ limit: 50, sort: "-wa_lastMsgTimestamp" }),
  });
  const unread = (chats.chats || []).filter((c) => c.unreadCount > 0).length;
  facts.push({
    instance: inst.name,
    campaigns: folders.map((f) => ({ status: f.status, total: f.log_total, delivered: f.log_delivered, read: f.log_read, failed: f.log_failed })),
    chats_with_unread: unread,
  });
  await new Promise((res) => setTimeout(res, 1500));   // pause between instances
}
console.log(JSON.stringify(facts, null, 2));

Give the AI only the script's facts and a strict instruction. A prompt template that works: "You write the monthly WhatsApp summary for an agency client. Use only the JSON data below. Do not calculate percentages that are not in the data, do not cite numbers that are not in the JSON and, if a data point is missing, say it was not measured. Tone: clear, friendly, in English, for a non-technical reader. Structure: what happened, what deserves attention, suggested next step. Maximum 150 words."

Always ask for "suggested next step" instead of "conclusion": the report becomes a conversation about action. And review before sending: the report carries the agency's brand.

Two outputs. Through the client's WhatsApp, with POST /{instanceId}/send/text (see sending a message with curl), during business hours and only to those who asked to receive the report. Or through the client portal, which already shows leads, sales and performance under their brand, and the report becomes a text complement.

If the report is sent from the client's support number, treat it as an ordinary message: no advertising inside it and no agency or ZapZap branding the client did not ask for.

AI summaries of real conversations involve personal data. Do not send conversation content to a model without a legal basis and without a contract with the client; prefer aggregate counters, as in the example. The LGPD applies to the agency as processor, and the contract should say what is processed and where.

To build the pipeline with a coding agent, see build with AI: Claude Code, OpenAPI and MCP and the page on AI agents.