AI agents for lead generation: WhatsApp, web and voice (2026 guide)

An AI agent for customer service and lead generation is a conversational system that acts as well as answers: it understands what the person wants, consults your approved information and carries out a specific step, such as booking an appointment or logging the opportunity in the CRM. A classic chatbot only follows a script.

The difference shows when someone steps outside the menu. A button-based chatbot replies “I didn't understand that”; an agent keeps the context, asks for the missing detail and, if the question is beyond its remit, says so and hands over. In sales it is sometimes called an “AI SDR”: it handles AI lead qualification, not the close. This is the work of our AI agents and chatbots service.

• Answers from your approved information: services, terms and opening hours.
• Qualifies with the questions your sales team has defined.
• Acts: books, gathers the details for a quote and records the contact with their consent.
• Hands over to a person, with a summary and the history, when the case calls for it.

Your customers are already solving problems with AI assistants that are not yours. According to a Gartner survey of 3,566 customers, published on 8 July 2026, they are about three times more likely to resolve a service issue with third-party generative AI, such as ChatGPT, than with the company's own chatbot. Use of those tools nearly doubled in a year; use of company chatbots has been flat since 2022.

The conclusion is not that your own agent is redundant, but that it must do what a general assistant cannot: check your real availability, book, or pass the conversation to your team. And your help pages need to be readable by AI: that is the territory of generative engine optimization (GEO), covered in our guide on how to get recommended by ChatGPT.

What does not exist here yet: on 20 May 2026 Google presented new ad formats for AI Mode, including Business Agent for Leads. As of October 2026 they are in testing in the United States, with no date for Spain or the EU.

An agent should serve people where they already write or call: in Spain that means WhatsApp above all, alongside your website chat and the phone. According to IAB Spain's 2026 social media study, presented on 19 May 2026 and covering users aged 12 to 75, WhatsApp reaches a monthly penetration of 94%, the highest of the networks analysed.

Language is part of the channel: a business with international customers in Mallorca usually needs Spanish, English and German, plus Catalan for the public in the Balearic Islands. Start with the channel where most enquiries are being lost today, with a tightly scoped pilot.

• WhatsApp. Enquiries, quotes, appointments and follow-up in conversations the customer picks up whenever they like. How Meta bills it and which rules apply are explained on our WhatsApp AI chatbot page.
• Web chat. It serves someone who is already on your site and hesitating before a decision. It should work with a keyboard and a screen reader.
• Voice. AI voice agents answer calls out of hours or at peak times, book and transfer, with short sentences and AI disclosure in the first one.

A lead generation flow run by an AI agent has six steps: entry, disclosure, answer, qualification, action and handover or follow-up. Every step leaves useful data in your systems, and the person can ask for a human at any of them.

An example: a language school receives a WhatsApp enquiry on a Sunday evening. The agent introduces itself as an AI, answers on levels and timetables, asks when the person wants to start, offers a placement test in a real slot in the diary and leaves the contact in the CRM. On Monday the team finds a confirmed appointment, not an unread message.

The follow-up afterwards lives in your marketing automation and CRM: the agent feeds that system; it does not duplicate it.

• Entry: an ad with a WhatsApp button, the website chat, a QR code or a phone call.
• Disclosure: the agent says it is an AI and links to the privacy information.
• Answer: from approved content only; it does not improvise prices or availability.
• Qualification: a few questions, the ones your sales team has agreed.
• Action: it books, gathers the details for a quote or creates the opportunity in the CRM.
• Handover or follow-up: it passes to a person with a summary and the history, or confirms the appointment.

WhatsApp Business Platform charges per template message, not per conversation, and replying inside the customer service window is free. Meta has applied this model since 1 July 2025.

For WhatsApp chatbot lead generation the consequence is direct: an agent that replies to people who write to you almost always works inside free windows. That is why it pairs well with the Click-to-WhatsApp ads we run in paid social advertising.

Does Meta allow AI on WhatsApp? Yes, for your own business. Since 15 October 2025 for new users and 15 January 2026 for the rest, Meta has barred providers whose primary product is a general-purpose AI assistant; businesses using AI for their own customer service, bookings or sales remain allowed.

On 9 June 2026 the European Commission imposed interim measures ordering Meta to restore access for those assistants, pending its final decision. For an agent limited to your own business, the framework is unchanged.

• Templates: marketing templates are always charged; utility templates only outside the customer service window.
• The 24-hour window: it opens when the customer writes to you; inside it, non-template messages and utility templates carry no charge.
• The 72-hour window: it is free and opens when the conversation starts from a Click-to-WhatsApp ad or from the call-to-action button on your Facebook Page.

Since 2 August 2026, anyone talking to a chatbot or a voice agent in the European Union must be told at the start that they are dealing with an AI, unless it is obvious. This comes from Article 50 of the EU AI Act, in force since 1 August 2024.

The duty falls on providers of AI systems that interact with people, with fines of up to €15 million or 3% of worldwide turnover. In Spain, the supervisory authority is AESIA; Spain's own AI law is still going through parliament, where the bill was sent on 26 May 2026.

The disclosure matches what the public expects: in the same IAB Spain study, 87% of users considered it essential to label AI-generated content. In practice it is one sentence:

• Chat or WhatsApp: “Hello, I'm the restaurant's AI virtual assistant. I can help with your booking or put you through to a person.”
• Voice: the first sentence of the call identifies the agent as an AI, before it asks for any details.
• Design: no names or photographs that suggest a specific person is behind it.

An AI agent should never be a dead end: the person must be able to ask for a human at any point. In Spain the law already requires this of some companies, which have until 28 December 2026 to adapt.

Ley 10/2025, Spain's customer service law, has been in force since 28 December 2025 and gives companies a year, until 28 December 2026, to adapt. It applies to providers of basic services, such as energy, telecoms and financial services, and to large companies selling to consumers: 250 or more employees, turnover above €50 million or a balance sheet above €43 million.

SMEs that do not provide basic services are not directly covered, but it makes sense to apply the same standard: according to a Gartner survey published on 4 August 2026, 87% of customers say that companies using generative AI in customer service must provide access to a human agent. Handover has to be designed: what triggers it, who receives it and with what summary.

• An automated system cannot be the only channel: the customer has the right to speak to a person on request.
• Complaints must be resolved within 15 working days.
• Service must be offered in co-official languages where applicable; in the Balearic Islands, Catalan.

A conversation with an agent contains personal data from the first message. The GDPR and Spain's data protection law (LOPDGDD) apply to the processing, and Spain's information society services law (LSSI) to commercial communications, just as they do on a form.

This summary is no substitute for legal advice. If you need to put your team's use of AI in order before deploying an agent, we cover that in AI marketing strategy and training.

• Information at the start: controller, purpose, legal basis, recipients, retention and rights (Articles 13 and 14 of the GDPR), in a short notice with a link to the full policy.
• Minimisation: the agent asks only for what it will use.
• Separate marketing consent: the LSSI prohibits unsolicited commercial communications by email or SMS without a prior request or express consent, except to existing customers, about similar products and with an easy opt-out. We apply the same standard to WhatsApp marketing templates.
• Records: consent, its date and its origin are stored in the CRM.

AI agents for customer service work best on frequent, well-bounded tasks with a clear next step. These are the usual cases, with the limit worth setting in each.

• Hotels and restaurants: availability, group bookings and opening hours, in several languages and out of hours. More in hospitality and tourism.
• Clinics: appointments, changes and reminders. The agent gives no health advice and does not assess symptoms: that goes to the centre's staff. More in health and clinics.
• Education: admissions enquiries, enrolment deadlines and booking visits or placement tests. More in education and training.
• Professional services: it records the type of matter and books the first meeting. It gives no legal or tax advice. More in professional and B2B services.
• Ecommerce: order status, returns and product questions. The purchase is completed in your online shop. More in consumer brands and ecommerce.

An agent is measured by completed tasks, not by conversations held. Before launch you need to agree what counts as a result and record the baseline.

Rigour matters because a return is not guaranteed: according to the figures Gartner published on 8 July 2026, customer service teams put a median 12% of their 2025 budget into AI, and only 24% saw a positive financial return.

The agent's events should reach your web analytics with the appropriate consent. We do not promise a number of leads: we offer a method and data you can check. More context in the future of digital marketing; to start from your own conversations, get in touch.

• Qualified leads: those that meet the written criteria and arrive complete in the CRM.
• Appointments and bookings: those the agent confirms, and those that are then kept.
• Handovers: how many conversations pass to a person, and why.
• Times and quality: first response, resolution and accuracy reviewed by people.