10 Questions to Ask Before Hiring a WhatsApp AI
Updated: June 4, 2026
A vendor sends you a demo with text replies to simple questions and promises to automate your WhatsApp. It looks good enough — until your customer sends a 40-second voice note or a photo of the product they need and the AI goes silent. Choosing the wrong vendor means paying for a rigid flow that abandons the lead at the first off-script message. These ten questions help you avoid that scenario before you sign anything.
Why these questions make all the difference
Hiring a WhatsApp AI without asking the right questions means putting a system that fails whenever the customer behaves naturally on your business's most active channel. For a company whose first contact already arrives through WhatsApp, the difference between an AI that works and one that doesn't is measured in conversations lost every night.
The channel justifies it: according to the ONTSI report La Sociedad Digital en España, over 90% of Internet users in Spain use WhatsApp daily. The CNMC put daily use of instant messaging at close to 70% of the Spanish population. If the AI you deploy doesn't cover real-world cases, the cost of unanswered messages is still yours.
The 10 questions to ask any vendor
These questions separate a WhatsApp AI that resolves real cases from a menu flow that breaks at the first free-form message. Every incomplete answer is a use case your customer won't be able to resolve without human intervention.
- Does it run on Meta's official WhatsApp Business API?
The free WhatsApp Business app doesn't scale and doesn't allow real automation. Meta's official API is the only channel that guarantees stability, approved templates and compliance with the messaging rules. Without it, your number can be blocked without warning. - Does it transcribe voice notes and reply with one?
Ask whether the AI transcribes the message, detects the language and can reply with a voice note when the answer doesn't include a URL. If it can't, it ignores the communication channel a large share of your customers prefer. - Does it analyze images sent by the lead?
A customer who sends a photo of the product, the broken part or the document they received can't be given a generic reply. Ask whether the AI analyzes the image within the conversation and uses its content in the response. - Does it read URLs the lead shares in the chat?
If the lead pastes the link to a listing, a portal or a product page, the AI should read that content and answer about that specific case. Without URL reading, the customer has to repeat all the information. - Does it follow up when the lead goes quiet?
A conversation that stalls for hours is a cold lead that can still be recovered. Ask whether the system sends a follow-up message with a specific observation about the case, not a generic nudge. - How does it handle Meta's 24-hour window?
Meta's documentation sets a 24-hour window from the user's last message for free-form messages. Outside that window, only approved templates can be sent. Ask how the vendor manages that limit and which templates it uses to reactivate conversations. - What data does it process and how does it comply with GDPR?
A WhatsApp AI processes names, phone numbers and conversations. The AEPD requires informing users and having a legal basis. Ask where the data is stored and what privacy documentation is included to adapt to the channel. - How long until the first flow is live?
A vendor who promises timelines without knowing your operations, or who delivers nothing functional after months, is a red flag. Ask for an estimate based on projects similar to yours, not on the rehearsed demo. - What human handoff capability does it have?
The AI must recognize when a conversation requires human judgment and hand the case over with the context already prepared. Ask which words, situations or amounts trigger the handoff and how the thread reaches your team. - What metrics does it deliver and how are results measured?
Without data you don't know whether the AI is working. Ask which metrics are accessible: time to first response, conversations resolved without a human, leads recovered through follow-up and correct escalations. Without metrics, you can't justify or improve the investment.
Table: unverified answer vs. verified answer
This table summarizes what happens when a vendor can't answer each criterion. Use it to decide which questions are non-negotiable for your project and which can wait for a second phase without delaying the launch.
| Criterion | Without a verified answer | With a verified answer |
|---|---|---|
| Official WhatsApp Business API | Risk of your number being blocked | Stable channel under Meta's rules |
| Transcription and voice | Unanswered voice notes, lost leads | Voice transcribed and answered on the spot |
| Image analysis | Photos ignored or generic replies | The image feeds into the response context |
| URL reading | The lead repeats all the information | The AI reads the link and answers that specific case |
| Silence follow-up | Cold leads never recovered | Follow-up with a specific observation |
| 24h window management | Out-of-window messages without an approved template | Compliance with Meta's rules, no violations |
| GDPR and AEPD | Exposure to fines and complaints | Legal basis and an adapted privacy notice |
| Metrics | No way to know if the project works | Actionable data to optimize the flow |
How an honest vendor should answer these questions
A vendor that answers these questions well doesn't dodge technical limits or promise capabilities it can't demonstrate live. This conversation illustrates what a real evaluation should look like before hiring a WhatsApp AI for a service business.
Illustrative example:Lead: Does your AI run on Meta's official API or do you use another method?
Tubot: On the official WhatsApp Business API. Your number is verified with Meta, follows the messaging rules and uses approved templates when the conversation goes beyond the 24-hour window.
Lead: My customers send lots of voice notes while driving or working. Does it understand them?
Tubot: Yes. It transcribes the audio, detects the language and can reply with a voice note if the answer doesn't include a URL. If the customer sends several voice notes in a row, it groups them and replies coherently in a single message.
Lead: And if they send photos of the product or a document?
Tubot: It analyzes them within the conversation and uses the content in the response. If they send a photo of a part, a floor plan or a notice, the AI works with that specific image, not a generic reply.
Lead: What happens if a lead stops responding after receiving information?
Tubot: After a configurable period of silence, it sends a follow-up message with a specific observation from the earlier case. Not a generic text — a real detail from the previous conversation.
Lead: And if someone asks to speak with a person directly?
Tubot: Handoff is triggered by words, situations or amounts you define. The case reaches the human with the full history prepared, so they don't have to ask for the information again.
Lead: How do I know it's working after launch?
Tubot: You get metrics on time to first response, conversations resolved without a human, leads recovered through follow-up and correct escalations. If you want to see it with your real operations, we start with a free audit.
Lead: And GDPR and my customers' privacy?
Tubot: Data is handled following the AEPD's guidelines. We help you adapt the channel's privacy notice and define the legal basis for your type of business before activating the first flow.
Can a WhatsApp AI work without Meta's official API?
Technically it may appear to work, but operating outside Meta's official API violates the terms of service and exposes the business number to a permanent ban. For a business, losing its WhatsApp number means losing the conversation history and the trust of the customers who already know it. The only safe foundation is the official API, which Tubot uses for all of its clients.
The official API is also the only way to use approved templates outside the 24-hour window. Meta has managed that billing model since July 2025, with per-template-message rates by category. A vendor without official API access can't follow up outside that window without breaking the messaging rules.
To see how this plays out in specific industries, the articles on WhatsApp AI for real estate agencies, WhatsApp AI for dental clinics and the guide to automating orders through WhatsApp show how it applies in each case. To review your situation with your real operations, you can write via Tubot's contact page or start directly at Tubot.
Frequently asked questions
What's the difference between the WhatsApp Business app and the WhatsApp Business API?
The WhatsApp Business app is for managing messages manually with basic automations such as welcome messages. The WhatsApp Business API is Meta's official route for automating conversations at scale, integrating intelligence, using approved templates and running multiple flows simultaneously. A real WhatsApp AI runs on the API, not on the app.
Is it mandatory for the vendor to run on Meta's official API?
Yes. A vendor that doesn't run on the official API may offer automations that violate Meta's terms of service and expose the company's number to a permanent ban. The official API is the only channel that guarantees stability, approved templates and compliance with the 24-hour window Meta enforces.
How long does it take to get these questions answered before hiring?
A structured conversation with the vendor should resolve them in under an hour. If the vendor can't answer one of them directly, you already have useful information about the limits of the service. They are operational, concrete questions any serious vendor should be able to answer without evasions or extra slide decks.
What if the vendor promises capabilities it can't demonstrate?
Ask for a demo with one of your real cases, not the rehearsed example. Send a voice note, an image or a URL and watch whether the AI responds usefully or gives a generic reply. If the behavior in the demo doesn't match the promises, the behavior in production won't either.
How do I know the AI I hire complies with GDPR?
Ask where the data is stored, who has access, how long it is retained and what privacy documentation is included for the WhatsApp channel. The AEPD requires informing users about how their data is processed at first contact. A vendor that can't give you that information won't let you stay compliant with their tool.
Sources
- ONTSI, 2024, https://www.ontsi.es/es/publicaciones/Informe-anual-La-Sociedad-Digital-en-Espana-2024
- CNMC, 2023, https://www.cnmc.es/prensa/panel-usos-internet-servicios-ott-cnmc-20230526
- Meta, 2026, https://developers.facebook.com/docs/whatsapp/pricing
- AEPD, 2026, https://www.aepd.es/derechos-y-deberes/conoce-tus-derechos/derecho-informacion
- INE, 2024, https://www.ine.es/dyngs/INEbase/es/operacion.htm?c=Estadistica_C&cid=1254736160707
How many WhatsApp AI projects fail not because of the technology, but because nobody asked these questions before signing the contract?
Free audit. Request it on WhatsApp now: the team reviews your case and replies in under 30 seconds.
