{"id":23715,"date":"2026-10-14T14:50:17","date_gmt":"2026-10-14T12:50:17","guid":{"rendered":"https:\/\/www.agenciareinicia.com\/blog\/how-to-get-whatsapps-ai-to-read-only-what-it-needs-rag-cache-and-instructions\/"},"modified":"2026-10-20T11:40:39","modified_gmt":"2026-10-20T09:40:39","slug":"how-to-get-whatsapps-ai-to-read-only-what-it-needs-rag-cache-and-instructions","status":"publish","type":"post","link":"https:\/\/www.agenciareinicia.com\/en\/blog\/how-to-get-whatsapps-ai-to-read-only-what-it-needs-rag-cache-and-instructions\/","title":{"rendered":"How to Get WhatsApp’s AI to Read Only What It Needs: RAG, Cache, and Instructions"},"content":{"rendered":"\n

Table of Contents: How to Optimize an AI Chatbot on WhatsApp<\/strong><\/p>\n\n

\n
\n
    \n
  1. Why Optimizing an AI Chatbot Starts with What It Reads<\/a><\/li>\n\n\n\n
  2. Structured Knowledge (RAG)<\/a><\/li>\n\n\n\n
  3. Entry-Level Positions at OpenAI, Anthropic, and Google<\/a><\/li>\n\n\n\n
  4. RAG and Cache: When They Complement Each Other and When They Compete<\/a><\/li>\n\n\n\n
  5. Brief Instructions and Actions Based on Intention<\/a><\/li>\n<\/ol>\n<\/div>\n\n\n\n
    \n
      \n
    1. Shorter history, with buttons and Flows<\/a><\/li>\n\n\n\n
    2. Short Answers<\/a><\/li>\n\n\n\n
    3. Meta’s AI agent: what you can control<\/a><\/li>\n\n\n\n
    4. How to Determine Whether Optimization Is Working<\/a><\/li>\n\n\n\n
    5. Frequently Asked Questions<\/a><\/li>\n<\/ol>\n<\/div>\n<\/div>\n\n

      If you want to optimize an AI chatbot on WhatsApp<\/strong>, the first step isn\u2019t to switch models, but to look at what it reads before each response. In the previous article in this series, which discussed the cost of each type of AI response on the WhatsApp Business API<\/a>, we saw that almost the entire token cost comes from reading<\/strong>, and that a hastily built assistant can read five times more than necessary to provide the same response. This article explains how to avoid this and what you should ask your provider or technical team at each step (you don\u2019t need to know how to code to understand it). <\/p>\n\n

      If you’re still not sure what such an assistant is used for, start with our guide to chatbots for WhatsApp Business<\/a>. Fisio Norte is a fictional, illustrative example<\/strong>: a made-up network of physical therapy centers, with sample conversations and figures, which we use as a central theme, just as in the previous article. The providers\u2019 rates are real and were retrieved on October 14, 2026. <\/p>\n\n

      Why Optimizing an AI Chatbot on WhatsApp Starts with What It Reads<\/h2>\n\n

      An AI chatbot connected to the WhatsApp Business API<\/a> doesn’t just type out a response\u2014 it<\/strong> first has to read everything it needs<\/strong> to write it.<\/p>\n\n

      In each turn, the AI re-reads the instructions, the business knowledge, the actions it can perform, the conversation history, and the customer\u2019s message. What it reads is billed as input<\/strong>, and what it writes is billed<\/strong> as output. The output costs more per token, but it\u2019s much shorter, so it has less impact on the final bill. <\/p>\n\n

      Reminder: Almost the entire token bill consists of reading data (Fisio Norte case, for illustrative purposes)<\/h3>\n\n

      Fisio Norte answers the question \u201cHow much does a session cost?\u201d<\/em> With the initial configuration, the AI reads 20,940 tokens to generate 120<\/strong>. With the optimized configuration, the customer receives the same answer, but the AI reads 3,935 and generates 90. Here\u2019s how it breaks down, block by block (the previous article covered the first three adjustments, which brought the read count to 4,140; here, the conversation and the shorter responses are also included): <\/p>\n\n

      Block<\/th>Before (tokens)<\/th>Next (tokens)<\/th>Savings<\/th><\/tr><\/thead>
      Instructions<\/strong><\/td>2.000<\/td>900<\/td>\u22121,100<\/td><\/tr>
      Knowledge<\/strong><\/td>18.000<\/td>2.500<\/td>\u221215,500<\/td><\/tr>
      Actions<\/strong><\/td>500<\/td>300<\/td>\u2212200<\/td><\/tr>
      History<\/strong><\/td>400<\/td>195<\/td>\u2212205<\/td><\/tr>
      Message from the customer<\/strong><\/td>40<\/td>40<\/td>0<\/td><\/tr>
      Total Input (Reading)<\/strong><\/td>20.940<\/strong><\/td>3.935<\/strong><\/td>\u221217,005 (\u221281%)<\/strong><\/td><\/tr>
      Response (output)<\/strong><\/td>120<\/td>90<\/td>\u221230<\/td><\/tr><\/tbody><\/table><\/figure>\n\n
      \"A
      Reading the response before and after each adjustment (case study from Fisio Norte).<\/figcaption><\/figure>\n\n

      With advanced AI (GPT-6.1 Sol, at $2 input and $10 output per million tokens; 1 \u20ac = 1.14 $) and Meta\u2019s service fee (0.0166 \u20ac per response), 1,000 responses cost 54.39 \u20ac before and 24.29 \u20ac after<\/strong>. This is the basis for the calculations in the rest of the article. <\/p>\n\n

      Structured Knowledge (RAG): How to Prepare the Information<\/h2>\n\n

      Knowledge is the heaviest component (18,000 of Fisio Norte\u2019s 20,940 tokens), which is<\/strong> why it\u2019s where the greatest savings are realized<\/strong>. The technique that achieves this is called RAG, and its results depend almost entirely on how the information is prepared. <\/p>\n\n

      What Is RAG? Explained Without Technical Jargon<\/h3>\n\n

      Think about the difference between giving a new employee the company\u2019s entire manual before every call, or providing them with a good index so they can look up only the page they need. RAG is that index<\/strong> (an acronym for Retrieval-Augmented Generation): the information is divided into fragments, stored in a search engine that understands meaning (not just exact words), and when a question comes in, only the fragments that match it are sent to the AI.<\/p>\n\n

      For the customer, the answer remains the same; what changes is the interpretation<\/strong>.<\/p>\n\n

      \"RAG
      With RAG, the AI receives only the snippets that resemble the question.<\/figcaption><\/figure>\n\n

      How to cut: by sections and approximate size of the pieces<\/h3>\n\n

      It\u2019s more important to break things down properly than to break them down extensively. Each segment should contain a single, complete idea<\/strong>: a rate with its terms, a cancellation policy, or a center\u2019s hours. If you cut off in the middle of an explanation, the AI receives half-truths; if you mix topics, it receives noise. <\/p>\n\n