Chat Bots: What They Are, Safety, and How to Get One

You’ve probably chatted with an AI without realizing it—whether asking a retailer about a tracking number or testing a virtual assistant’s sense of humor. In early 2026, the University of Birmingham published a first-of-its-kind safety guide for public use of AI health chatbots, making it clear that chatbots are no longer just novelty tools.

FTC inquiry into AI companion chatbots: Sept 2025 (Federal Trade Commission (U.S. regulator)) ·
University of Birmingham safety guide: Feb 2026 (University of Birmingham (academic institution)) ·
JMIR systematic review: weak evidence for mental health benefits (2020) (Journal of Medical Internet Research (peer-reviewed journal))

Key research findings on chatbot safety and regulation:

What is a chat bot?

How do chatbots work?

• Chatbots process natural language input using either rule-based decision trees or AI-driven models that rely on large language models and machine learning.
• Rule-based chatbots follow pre-written triggers and can’t handle unexpected questions; AI chatbots generalize from training data and can respond to a wider range of inputs.

The simplest way to think of a chatbot is as a conversational interface. Rule-based chatbots are like a choose-your-own-adventure book—each response leads down a fixed path. AI chatbots, by contrast, generate replies by predicting the most likely next word, as demonstrated in models like GPT and BERT.

Types of chatbots (rule-based vs AI)

• Rule-based: Used in customer service for FAQs, order status, appointment booking.
• AI-powered: Used in virtual assistants (Siri, Alexa), companion chatbots (Character AI), and mental health support tools.

The 2020 JMIR systematic review noted that although AI chatbots showed potential for mental health support, the evidence was weak—a warning that even advanced bots don’t guarantee beneficial outcomes.

The pattern is clear: choose the type based on the complexity of the interaction, but always audit the safety features of any AI chatbot you deploy.

Why would someone use a chatbot?

Benefits for businesses

• Cost savings: reduce staffing costs in customer support (Botpress (chatbot platform)).
• 24/7 availability: provide instant responses any time.
• Scalability: handle multiple conversations simultaneously.

Benefits for end users

• Quick answers and personalized interactions (Bright Horizons (child development organization)).
• Access to mental health education and coping skills around the clock (AMF Treatment (mental health provider)).

The FTC’s 2025 inquiry into companion chatbots underscores that the very qualities that make chatbots appealing—their constant availability and seemingly empathetic responses—also create new emotional dependencies, especially for younger users.

The combination of business efficiency and user emotional engagement makes chatbot safety a priority for both developers and regulators.

Is there any free chatbot?

Which AI is 100% free?

• Free options include: ChatGPT free tier, Quillbot AI Chat, and HubSpot’s free chatbot builder (Botpress (chatbot platform)).
• Many free tools have usage caps, limited features, or display ads.

Limitations of free chatbots

• Free versions often lack advanced analytics and integrations (Botpress (chatbot platform)).
• Security features may be less robust, raising privacy concerns (Cyberbullying Research Center (youth safety organization)).

The 2026 regulatory literature from PubMed Central stresses that disclosure of capabilities is not enough—free chatbots should also have clear internal safety protocols. Users should check whether a free bot encrypts conversations and what data it retains.

The catch: a free chatbot that collects your conversation history for training might cost you more in privacy than the paid version ever would.

Are chatbots safe to use?

Are AI chat bots illegal?

• Chatbots themselves are not illegal, but using them for fraud, impersonation, or harassment is prohibited (FTC (U.S. regulator)).
• Platforms that fail to moderate harmful chatbot interactions face regulatory action (EPIC (privacy advocacy group)).

What should you avoid entering into chatbots?

• Never share passwords, credit card numbers, Social Security numbers, or health records (University of Birmingham (academic institution)).
• Avoid sharing personal details that could be used for social engineering (Galileo (AI safety platform)).

The Ensora Health 2026 therapist guide recommends a multi-pillar approach: limit data shared, use encrypted connections, and have a plan to disengage if the conversation feels emotionally manipulative.

The implication: safety isn’t automatic. It depends on the platform’s encryption, data policy, and the content filter (or lack thereof).

How do I get a chat bot?

Using pre-built platforms

• Platforms like HubSpot, Chatfuel, and ManyChat offer drag-and-drop builders (Botpress (chatbot platform)).
• Steps: choose platform → define goal → design conversation flow → test → deploy.

Building your own chatbot

• Use APIs from OpenAI, Google Dialogflow, or Microsoft Bot Framework.
• Requires basic programming skills; offers full control over behavior and data handling (Botpress (chatbot platform)).

1. Define your use case – customer support? personal assistant? roleplay? Each demands a different architecture.
2. Select a platform or framework – no-code (HubSpot) vs code (OpenAI API).
3. Design conversation flows – map out expected inputs and responses; test edge cases.
4. Integrate safety features – apply content filters and data encryption, following the University of Birmingham guide.
5. Test with real users – simulate off-topic and sensitive questions.
6. Deploy and monitor – set up logging and review conversations for safety issues.

The trade-off: using a no-code builder gets you a chatbot in minutes, but you sacrifice privacy control. Building your own takes longer but lets you enforce strict data practices.

How to tell if someone is using a chat bot?

How can I tell if I am chatting with an AI bot?

• Repetitive phrasing – bots often reuse sentence structures.
• Extremely fast responses – humans pause; bots don’t.
• Inability to handle off‑topic or complex queries – ask about current events or emotional nuance.

The Galileo 2025 safety strategy emphasizes real‑time risk signals: if a conversation partner avoids direct questions, repeats canned responses, or never corrects its own mistakes, it’s likely a bot.

Detection tools like CAPTCHA and reverse image search help, but no single method is foolproof. A Stanford 2025 study found that even AI companion bots can simulate empathy convincingly—making bot detection harder than ever. For more information on this topic, check out SpongeBob SquarePants questions answered.

The pattern: as AI improves, traditional detection methods (Turing test, CAPTCHA) weaken. Future solutions will likely require a combination of behavioral analysis and regulatory disclosure mandates.

“This guide is a pragmatic, neutral resource focused on harm reduction and maximizing benefits.” — University of Birmingham, on its 2026 AI health chatbot safety guide

University of Birmingham (academic institution)

“Risks and dangers of AI companions for young people are growing as these products become more emotionally sophisticated.” — Stanford News, 2025

Stanford News (academic research)