Chatbot Potencializado por IA: Como os Chatbots Modernos Funcionam, Quanto Custam e Qual Se Encaixa no Seu Negócio

Pesquisar por um chatbot com ia em 2026 e você aterrissa em um mercado que mistura três produtos muito diferentes sob um único rótulo. Um grupo são assistentes pessoais de IA, como ChatGPT, Claude e Gemini. Outro grupo são construtores de automação social e de mensagens, como MessengerBot, ManyChat e Chatfuel. Um terceiro grupo é software de serviço com um agente de IA integrado em um help desk, como Tidio, Intercom, Zendesk e HubSpot.

Essa confusão de categorias é a razão pela qual muitos guias de compra parecem inúteis na prática. Uma marca de ecommerce com cinco pessoas respondendo mensagens na página do Facebook não está resolvendo o mesmo problema que uma equipe de suporte de SaaS tentando automatizar 20.000 tickets por mês. Uma clínica local que deseja respostas fora do horário no Messenger precisa de algo diferente novamente. Se você comparar todos eles como se fossem intercambiáveis, você ou paga demais por recursos que nunca usará ou compra de menos e acaba reconstruindo todo o sistema três meses depois.

Eu revisei os preços públicos e as páginas de planos vinculadas neste artigo sobre 12 de abril de 2026. O rápido resumo já é suficiente para mostrar como o mercado é diferente: ChatGPT Plus continua a $20 por mês, Claude Pro continua a $17 por mês com faturamento anual ou $20 mensal, Google AI Pro é $19,99 por mês, MessengerBot Premium é $19,99 a cada 30 dias, ManyChat Essential é $17 por mês, Tidio Starter é $24,17 por mês, e o Fin do Intercom é cobrado a $0,99 por resultado resolvido.^{[1][2][4][6][8][12][14]}

A questão prática não é qual chatbot é o mais inteligente em um benchmark. É qual deles se encaixa no seu canal, no seu fluxo de trabalho e na sua tolerância de cobrança. Se as mensagens da Página do Facebook são uma fonte importante de leads para você, os limites de plano fixo importam mais do que classificações de modelos abstratos. Comece por aí, então Ver Preços do MessengerBot contra os modelos de custo que você vê em outros lugares para que você esteja comparando as coisas certas.

O que um Chatbot Potencializado por IA Realmente Significa em 2026

Um chatbot com ia não é apenas uma janela de chat com um grande modelo de linguagem por trás dela. Em produção, um chatbot útil é uma pilha. Ele precisa de uma interface, um modelo, regras de negócios, memória, acesso aos dados certos, registro e um caminho limpo para um humano quando o bot deve parar de falar.

A maneira mais fácil de pensar sobre isso é a seguinte: o modelo escreve a frase, mas o sistema decide se o modelo deve responder ou não. Essa distinção é onde bons bots se separam de demonstrações caras. Um bot sólido não permite que o modelo improvise sobre política de reembolso, regras de elegibilidade, exceções de preços ou recuperação de conta quando essas respostas devem vir de dados estruturados ou de um fluxo determinístico.

Até 2026, a maioria das implantações sérias de chatbot de IA utiliza um design híbrido:

• Lógica baseada em regras lida com fluxos de trabalho conhecidos, como captura de leads, roteamento de menu, agendamento, opt-ins, etiquetagem e gatilhos de transferência.
• Recuperação puxa o artigo, FAQ, detalhe do produto ou registro de CRM certo para que o modelo esteja fundamentado em informações comerciais atuais.
• IA Generativa transforma essa informação em uma resposta natural, faz perguntas de esclarecimento ou resume um pedido confuso.
• Uso de ferramentas permite que o bot execute ações como verificar o status do pedido, enviar um e-mail de acompanhamento ou escrever dados em Planilhas ou em um CRM.
• Escalonamento humano assume quando a confiança diminui, um caso limite de política aparece ou um cliente claramente deseja uma pessoa.

É por isso que um chatbot empresarial não deve ser avaliado apenas pela qualidade da linguagem bruta. Um modelo brilhante com roteamento ruim ainda cria resultados ruins. Um modelo ligeiramente menos impressionante com um design de fluxo forte, melhor acesso a dados e guardrails mais rígidos frequentemente resolve mais conversas reais de clientes com menos danos.

A outra coisa que vale a pena dizer claramente: chatbots de IA voltados para o cliente verdadeiramente úteis quase nunca são “sem necessidade de cadastro.” Essa frase ainda se aplica a ferramentas de chat leves para consumidores. Não descreve software de mensagens de produção, porque o software de produção precisa de permissões, canais, estado salvo, análises e controle administrativo.

Como um Chatbot Potencializado por IA Funciona nos Bastidores

Por trás das câmeras, os chatbots modernos seguem um pipeline bastante repetível. Os detalhes mudam de plataforma para plataforma, mas a arquitetura é consistente o suficiente para que você possa avaliar qualquer ferramenta com a mesma lista de verificação.

1. Um evento chega. Um visitante envia um chat no site, um cliente responde no Facebook Messenger, uma DM do Instagram chega ou um e-mail chega na caixa de entrada de suporte.
2. O roteador classifica a solicitação. O sistema decide se a mensagem é um fluxo de trabalho conhecido, uma pergunta geral, um problema de alto risco ou algo que deve ir direto para um agente.
3. O bot recupera o contexto. Isso pode ser um artigo da base de conhecimento, uma página de produto, um registro de CRM, uma linha de Google Sheet ou o contexto de conversas passadas.
4. O modelo gera uma resposta. O LLM transforma o contexto fundamentado em uma resposta legível por humanos, frequentemente com instruções sobre tom, limites e regras de escalonamento.
5. Ferramentas são chamadas quando necessário. O bot pode buscar o status de um envio, criar um lead, adicionar uma tag, escrever para um webhook ou agendar um acompanhamento.
6. As regras de segurança são executadas antes da entrega. Os limiares de confiança, tópicos bloqueados, cópia de fallback e regras de transferência para humanos decidem se essa resposta deve ser enviada como está.
7. Tudo é registrado. O sistema armazena transcrições, tags, resultados e sinais de resolução para que a equipe possa melhorar prompts, fluxos e a qualidade do conhecimento.

Em termos práticos, a maioria das empresas acaba com três camadas de memória:

• Memória de sessão para o chat atual, para que o bot possa acompanhar a conversa.
• Memória de perfil para atributos do cliente, como e-mail, idioma, status de compra ou localização.
• Business memory for policies, FAQs, catalogs, and process documents that should shape answers every time.

The biggest technical improvement since the first wave of AI chat bot hype is retrieval quality. Modern systems do not just stuff your full website into a prompt and hope for the best. They break documents into chunks, embed them for semantic search, rank the best matches, and then pass only the relevant context to the model. That makes answers cheaper, faster, and less likely to drift.

For MessengerBot users, this architecture matters because the platform already covers the parts many small businesses forget to budget for: visual flow control, tags, opt-in forms, website chat, Google Sheets sync, JSON API access, and message sequencing.^[6] In other words, you do not need to make AI generate every single answer to get an “AI powered chatbot” result. Often the better design is to let the model handle the messy text while the platform handles the workflow.

Consumer AI Chat Bots and Business Chatbots Solve Different Jobs

This is the fork most buyers need to get right first. Consumer AI chat bots optimize for broad usefulness: writing, summarizing, coding, studying, brainstorming, and file work. Business chatbots optimize for routing, channel permissions, user identity, lead capture, automation, analytics, and handoff. The overlap is real, but the job is different.

If your team says, “We need a chatbot,” ask one harder question: Who is the user?

• Se o usuário for da sua equipe, ferramentas como ChatGPT, Claude e Gemini costumam ser a primeira compra certa.
• Se o usuário for seu cliente no Messenger, Instagram ou chat do site, uma plataforma de mensagens ou suporte é geralmente a melhor primeira compra.
• Se ambos forem verdadeiros, a melhor configuração costuma ser uma pilha de duas camadas: um assistente de IA interno para agentes e uma plataforma de automação voltada para o cliente para conversas reais.

É por isso que as assinaturas de IA pessoal parecem baratas em comparação com plataformas de serviço. ChatGPT Plus, Claude Pro e Google AI Pro têm preços semelhantes a ferramentas de produtividade para consumidores ou prosumers. Intercom, Zendesk, HubSpot, Tidio, ManyChat e MessengerBot têm preços em torno do volume de canais, assentos, contatos ativos ou resultados porque estão lidando com carga de trabalho, suporte e operacional, não apenas gerando texto.^{[1][2][4][14][16]}

Uma boa regra é simples. Use um assistente de IA para consumidores quando a saída for principalmente texto para sua equipe. Use um chatbot empresarial quando a saída mudar o fluxo de trabalho de um cliente, capturar receita, resolver suporte ou tocar um canal com permissões e obrigações de serviço.

Os Modelos de Preços Que Decidem Quanto Seu Chatbot Realmente Custa

A maioria das páginas de preços de chatbots alimentados por IA parece simples até que você mapeie o gatilho de cobrança real. É aí que os custos mudam de “parece barato” para “por que o financeiro está fazendo perguntas.”

Here is the practical math behind those models. MessengerBot is easier to forecast because the public plans are tiered and feature-based. Premium is $19.99 per 30 days and Pro is $49.99 per 30 days on the public pricing page, with clear limits around pages, chat widgets, ecommerce stores, and advanced features.^[6] ManyChat is harder to forecast because contacts can quietly grow faster than revenue if you run frequent DM campaigns. Intercom and HubSpot are easy to model in a spreadsheet but can get very expensive if your AI actually resolves at scale, because success is the billing event.^{[8][9][14][18]}

There are also two hidden cost layers that never show up cleanly on the pricing page:

• Setup cost. Someone has to clean knowledge sources, design flows, write handoff rules, and test edge cases.
• Switching cost. Exporting contacts is easy compared with rebuilding triggers, tags, prompts, workflows, fallback logic, and analytics.

That second point is where a lot of teams make bad decisions. They pick the cheapest starter plan instead of the cleanest long-term billing model. The result is usually one of two painful outcomes: a migration project, or a year of working around the platform instead of using it properly.

AI Powered Chatbot Pricing Comparison for 2026

The table below compares the main platforms that come up in real buying conversations. I am grouping consumer AI assistants, messaging automation tools, and support platforms together on purpose because that is what buyers actually do in search results. The difference is that here the categories are explicit.

Pricing references reviewed April 12, 2026: OpenAI, Anthropic, Google One, MessengerBot, ManyChat, Chatfuel, Tidio, Intercom, Zendesk, and HubSpot official pages.^{[1][2][4][6][8][9][10][11][12][14][16][17][18]}

The big headline from this table is not that one tool beats the others at everything. It is that the cheapest-looking product is often the wrong comparison. ChatGPT, Claude, and Gemini are bargain subscriptions for internal productivity. Intercom, Zendesk, and HubSpot are operational systems. MessengerBot, ManyChat, and Chatfuel live in the middle, where channel coverage and marketing automation matter more than enterprise workflow control.

ChatGPT vs Claude vs Gemini for Teams That Need General AI Chat

If your team mainly needs an internal AI assistant, the first shortlist is still ChatGPT, Claude, and Gemini. The differences are not just about output style anymore. They now include model access, context window size, research limits, business connectors, and how deeply the tool plugs into your existing software.

Bate-papoGPT is still the easiest broad recommendation because the product does the widest mix of jobs well. The paid entry point remains $20 per month for Plus, and OpenAI’s current pricing page shows useful context tiers even for non-enterprise users: 54K for GPT Instant on Plus, 128K on Pro, and 256K reasoning context on Plus and Business.^[1] That makes it a good fit for mixed writing, coding, spreadsheet, research, and internal operations work. The main limitation is that you still need another platform if you want governed customer messaging across Facebook Messenger, Instagram, or a support inbox.

Claude is the best fit when your workflow is document-heavy and tone-sensitive. Anthropic still keeps Claude Pro at $17 per month on annual billing or $20 monthly, and the Pro plan now includes Claude Code, Claude Cowork, projects, research, and access to more models.^[2] On the API side, Anthropic documents a 1M token context window for Claude Sonnet 4, but that is not the same thing as a fixed claude.com consumer limit, which remains governed by usage caps and session-level limits.^[3] That distinction matters because a lot of buyers see the model context headline and assume the consumer chat product behaves like the API. It does not.

Gemini makes the strongest case if your team already lives inside Google Workspace. Google One’s public plans page keeps Google AI Pro at $19.99 per month, bundled with 5 TB of storage and Gemini in Gmail, Docs, Vids, and more.^[4] Google’s Gemini limits page is also more explicit than many vendors about capacity tiers: the basic plan sits at a 32 thousand token context window, while higher paid tiers scale up to 1 million, with Deep Research, image, video, and agent limits broken out by plan.^[5] The catch is packaging churn. Google changes tier names and bundled benefits more often than most procurement teams like.

My short version is blunt:

• Pick Bate-papoGPT if you want the strongest all-around internal assistant.
• Pick Claude if long reading, editing, and careful writing dominate the workload.
• Pick Gemini if your company runs on Gmail, Docs, Drive, and Google search habits already.

What I would not do is expose one of these directly to customer-facing channels and call the job done. They are excellent brains. They are not, by themselves, a support operation or a Messenger automation system.

MessengerBot vs ManyChat vs Chatfuel for Messenger and Social DM Automation

This is the comparison that matters for a lot of SMBs because customer conversations still start in DMs far more often than enterprise buyers like to admit. Facebook pages, Instagram replies, click-to-message ads, and comment-triggered conversations are still where a lot of real sales and support work happens.

MessengerBot is the cleanest fit when Facebook Messenger is the center of gravity. On the public pricing page, the Premium plan is $19.99 per 30 days and the Pro plan is $49.99 per 30 days. Premium includes one Facebook account, five Facebook pages, unlimited subscribers, one chat widget, one ecommerce store, sequence messaging, website chat, JSON API plus Zapier, Google Sheets integration, forms, comment tools, and more. Pro expands that to ten pages, five chat widgets, five ecommerce stores, Instagram chatbot features, team members, and broader operational depth.^[6] The biggest advantage is cost clarity. You are not doing active-contact math every week.

Muitos bate-papos remains the smoothest social growth tool if Instagram and creator-style funnels matter more than Facebook page support. But its March 2, 2026 pricing reset made the economics more important to understand. The Free plan covers up to 25 active contacts. Essential is $17 monthly or $14 annual for up to 250 active contacts, with $0.10 per extra contact on monthly billing. Pro is $39 monthly or $29 annual for up to 2,500 active contacts, then overage applies at a lower rate. Pro also unlocks AI-powered automation and channels like WhatsApp, SMS, and Email.^[7][8][9] That structure works if you are intentionally building social funnels. It gets painful if you treat contact growth as free.

Combustível de bate-papo is harder to recommend cleanly right now for one reason that has nothing to do with bot quality: the public pricing is inconsistent across its own pages. The main English pricing page currently presents a single $69 per month AI Business Assistant offer for WhatsApp, Instagram, and social messaging. A localized pricing page still shows a conversation-based Business tier starting at $23.99 plus $0.02 for each extra conversation.^[10][11] That suggests either a transition, a regional split, or different product packaging. Any one of those can be legitimate, but if you are comparing vendors for a finance-signoff purchase, that ambiguity is a real mark against it.

Here is the practical way to separate the three:

• Escolher MessengerBot if your business lives inside Facebook Page messages and you want clearer plan tiers.
• Escolher Muitos bate-papos if Instagram-centric growth and creator funnels drive your revenue.
• Escolher Combustível de bate-papo only after you confirm which pricing page and product packaging applies to your region and channel mix.

If you already know you need Instagram bot access, more pages, and more widgets, compare those limits in Upgrade to MessengerBot Pro before you default to a contact-priced competitor.

Tidio vs Intercom vs Zendesk vs HubSpot for Support Teams

Once the job moves from “answer DMs” to “run customer service,” the stack changes. Support teams care about queues, ticketing, ownership, reporting, multilingual content, auditability, and the exact meaning of a so-called resolved conversation. This is where support platforms start to matter more than social automation builders.

Tidio is the easiest SMB recommendation in the help-desk category. The public pricing page shows Starter at $24.17 per month, Growth starting at $49.17, Plus starting at $749, and a standalone Lyro AI Agent package from $32.50 per month starting at 50 AI conversations. Tidio also gives every account 50 free Lyro conversations lifetime, and its AI page publicly pitches Lyro at $0.5 per conversation.^[12][13] That hybrid structure works well for smaller website-first teams, but you need to budget both the help desk layer and the AI layer.

Intercom has the clearest AI billing in the enterprise support group. Essential is $29 per seat per month billed annually, Advanced is $85, Expert is $132, and Fin AI Agent is priced at $0.99 per outcome. Intercom’s own help page defines an outcome as a conversation Fin resolves or a Procedure that ends in a resolution or intentional handoff, and you are billed once per conversation even if multiple questions are resolved inside it.^[14][15] That transparency is a serious strength. It is also the reason CFOs will inspect the model closely at scale. A 3,000-outcome month is $2,970 before seats.

Zendesk is still the safest choice for organizations already built around ticketing discipline. The current public pricing page shows Suite + Copilot Professional at $155 per agent per month billed annually and Enterprise at $209, while Advanced AI agents remain custom-priced.^[16] That is not cheap, but Zendesk buyers are usually not looking for cheap. They are looking for operational control, governance, mature workflows, and a platform the support org can standardize on.

HubSpot makes the most sense when the CRM is the real buying center. Service Hub Starter is currently shown from $15 per month per seat on the product page, Professional from $100, and Enterprise from $150.^[17] The AI twist is more interesting: HubSpot announced on April 2, 2026 that Breeze Customer Agent moves to outcome-based pricing on April 14, 2026 at $0.50 per resolved conversation, and HubSpot says the product already resolves 65% of conversations and cuts resolution time by 39% across more than 8,000 activated customers.^[18] That performance data is vendor-reported, so treat it as directional, not neutral benchmarking. Still, the pricing change is real and unusually concrete.

The simplest buying rule here is this:

• Escolher Tidio for smaller website-first teams that want a lighter stack.
• Escolher Intercom if you want transparent AI outcome billing and a modern AI-first support platform.
• Escolher Zendesk if your team already operates like a disciplined ticketing organization.
• Escolher HubSpot if customer service sits inside a broader CRM-led operating model.

How to Design an AI Chatbot That Does Not Hallucinate or Trap Users

Most chatbot failures are not model failures. They are systems failures. Teams ask the model to behave like a complete support stack, then act surprised when it improvises around missing data, unclear policy, or an impossible customer request.

If you want a production chatbot that feels reliable, design around these six rules:

1. Use deterministic flows for risky tasks. Pricing, refunds, account security, and anything with legal or payment implications should be rule-led first.
2. Ground the bot in current business data. Retrieval beats memory. If the answer changes, store it in a source the system can refresh.
3. Set confidence thresholds. A bot should know when it is guessing and escalate before damage happens.
4. Separate explanation from execution. Let AI explain policy. Let workflows and tools actually perform the action.
5. Make handoff visible. Customers should never feel trapped in an endless “I can help with that” loop.
6. Log failures by intent, not just by CSAT. You need to know which topics break the system, not just that a conversation went badly.

A lot of small businesses get the best result from a two-lane design. Lane one is structured automation: greetings, menu choices, lead capture, tags, appointment prompts, and post-click follow-up. Lane two is AI-assisted free text for FAQ-style questions, qualification, and summarization. MessengerBot is well suited to that pattern because visual flows, forms, tags, comment tools, website chat, and integrations already exist around the conversation layer.^[6]

What I would not do is let an LLM answer everything with one giant system prompt. That makes demos look magical and production logs look expensive. Good chat bot ai deployments are opinionated about when the model is allowed to talk.

A 14-Day Launch Plan for Your First Production AI Chatbot

If you are starting from zero, do not try to automate your entire customer journey in week one. Launch the smallest useful bot first, then expand. This is the rollout I use when the goal is to get a real system into production fast without creating a cleanup project.

If you are building this inside MessengerBot, a practical starter stack is straightforward: welcome flow, menu, top-intent quick replies, tag capture, human handoff, one fallback AI answer block, and a Sheet or CRM sync for leads. That is enough to learn from real usage without turning the first version into a maze. If you want setup examples before you touch production traffic, Navegue por nossos Tutoriais and borrow a working pattern instead of improvising your first flow tree.

The launch metric that matters most early is not “AI usage.” It is one of these three: resolved conversations, qualified leads captured, or support deflection with acceptable customer satisfaction. Pick one. Otherwise you will spend two weeks admiring transcripts instead of measuring business value.

Where MessengerBot Fits Best for Facebook Messenger, Instagram, and Website Chat

MessengerBot is strongest when the business problem is channel-specific rather than model-specific. If your buyers spend time in Facebook Messenger, your support team lives in page inboxes, or your funnel depends on comment replies, DMs, broadcasts, forms, and follow-up sequences, that is where the product makes sense.

The current public pricing structure is simple enough to budget without gymnastics. Premium at $19.99 per 30 days is a reasonable entry point for a single-account operation that needs up to five Facebook pages, one website chat widget, one ecommerce store, unlimited subscribers, flow building, website chat, email tools, JSON API plus Zapier, Google Sheets sync, forms, and core post or comment automation. Pro at $49.99 per 30 days is where the platform becomes more useful for heavier operators, because it expands pages and widgets, adds Instagram chatbot capabilities, supports more team-oriented work, and opens a wider operational footprint.^[6]

That makes MessengerBot a particularly good fit for:

• Local businesses that get repeated Messenger questions about hours, pricing, availability, and bookings
• Ecommerce brands using Facebook and Instagram comments to trigger DMs and recover abandoned interest
• Agencies managing multiple small-business pages without wanting active-contact pricing surprises
• Teams that want a visual builder and integrations without standing up a custom app stack

It is a weaker fit when your primary operating model looks like a large-scale help desk with strict ticket queues, advanced enterprise security review, or heavy phone and email service orchestration. That is not a flaw. It is product positioning. MessengerBot does not need to beat Intercom or Zendesk at enterprise help-desk governance to be the right answer for Messenger-first growth and support.

There is also a business model angle worth noting if you build flows for clients. If you are packaging chatbot setup as a service, flat tier pricing is easier to margin than contact-driven pricing, and the partner upside is more straightforward too. In that case, it can make sense to Junte-se ao nosso programa de afiliados while you are rolling chatbot builds into your client offer.

The Mistakes That Make AI Chat Bots Look Smart in a Demo and Weak in Production

I see the same failure pattern repeatedly across chatbot rollouts. The team does not buy the wrong technology because they are careless. They buy the wrong technology because the demo rewards the wrong thing.

These are the mistakes that hurt most often:

• Buying on model hype instead of channel fit. The smartest model in a screenshot is useless if your real problem is Facebook permissions, inbox routing, or active-contact billing.
• Letting the LLM answer everything. Good bots use AI selectively. Bad bots hope the model will invent workflow discipline.
• Ignoring the actual billing trigger. Per contact, per conversation, per outcome, and per seat are not interchangeable.
• Skipping handoff design. A bot that cannot fail gracefully creates more work than it saves.
• Feeding the system bad source material. If your policies are outdated or contradictory, retrieval just makes wrong answers faster.
• Not logging intent-level failures. You need to know whether returns, billing, delivery, or product fit questions are breaking the system.
• Treating free plans as production plans. Free is good for evaluation. It is rarely the right place to stop.

The Chatfuel pricing inconsistency is a good real-world example of why this matters. A lot of comparison posts would quietly pick whichever number makes the table look neat. That is the wrong move. If the public pricing picture is inconsistent, the correct takeaway is not “cheap.” The correct takeaway is “verify before you buy.”^[10][11]

The same principle applies everywhere. If a vendor bills per resolved conversation, model your resolved conversations. If a vendor bills per active contact, model contact growth. If a vendor bills per seat, count the people who will really need access six months from now, not just the pilot team.

Which AI Powered Chatbot Fits Your Business Right Now

If you want the shortest usable answer, use this decision matrix instead of another generic top-10 list.

If your use case is specifically Messenger, Instagram, and website chat for a small or midsize business, the market narrows fast. That is where MessengerBot, ManyChat, and Tidio deserve most of the attention. If you are answering Facebook page questions, collecting leads, and routing to human support when needed, the “best” chatbot is usually the one that keeps your channel operations simple, not the one with the most dramatic AI branding.

Ready to Build a Messenger-First AI Powered Chatbot?

If your next step is not more theory but an actual build, keep it simple. Start with one live use case, verify the billing trigger before launch, and make the handoff path obvious. For Messenger-first teams, the fastest path is usually to compare plan limits, copy a proven flow structure, and only then add AI where free text actually helps.

Use these three pages in that order: Ver Preços do MessengerBot, Navegue por nossos Tutoriais, e Upgrade to MessengerBot Pro if you already know you need broader page, widget, or Instagram coverage. If you are building chatbot setups for clients, the fourth step is simple too: Junte-se ao nosso programa de afiliados.