Tidio Review 2026: Four Lyro Playground Tests

I tested Tidio’s Lyro AI Agent with Prompt Atlas as its knowledge source, using four Playground tasks to check retrieval, recommendations, hallucination control, and handoff recognition.

The results below describe those four scenarios; they do not measure performance in live customer conversations.

What Is Tidio?

Tidio is a customer service platform that combines live chat, ticketing, automation, and AI-powered support.

Its main AI customer service product is Lyro AI Agent.

Rather than simply following a fixed decision tree, Lyro can use a business’s own support content to understand questions and generate responses. Tidio supports knowledge sources such as website links, PDFs, and manually entered information.

That distinction became important during my testing.

I connected Prompt Atlas as a knowledge source and then used Tidio’s Playground to see how Lyro would respond to questions based on the information available on the site.

How I Tested Lyro AI

This was a small, controlled evaluation using my own website. The retrieval and recommendation questions drew on content I knew it contained. Only one unsupported-service question was tested; ambiguous wording, outdated source content, and plausible questions just outside the knowledge base were not tested.

I evaluated four different capabilities:

1. Knowledge retrieval: Can Lyro find and explain information already available on my website?
2. Recommendation: Can it interpret a customer’s needs instead of simply repeating information?
3. Hallucination control: What happens when someone asks about something that does not exist?
4. Human handoff: Can it recognize when a frustrated customer wants to speak with a person?

I used the same Prompt Atlas website as the knowledge source throughout the test.

This is important because the results below reflect my own hands-on testing, not just claims from Tidio’s marketing materials.

Test 1: Can Lyro Answer Questions From My Website?

What I Asked

My first question was straightforward:

What kind of AI tools does PromptAtlas recommend for small businesses?

The information required to answer this question was already available on Prompt Atlas.

What Happened

Lyro successfully identified the relevant content and produced a detailed answer covering multiple tools recommended on the site.

It mentioned tools including ChatGPT, Claude, Gemini, Canva, Zapier, HubSpot, Notion AI, Grammarly, Perplexity, and Otter.ai, along with brief explanations of what each tool could help a small business accomplish.

Lyro used content from Prompt Atlas to answer a question about recommended AI tools.

More importantly, the response did not stop at listing tools.

Lyro suggested starting with ChatGPT or Claude for broad business needs and linked the user toward the relevant Prompt Atlas content.

My Take

Test 1: Passed.

This was a strong knowledge-retrieval result.

The response was relevant to the question, grounded in information available on the website, and useful enough to help a visitor take a next step.

For a small business with an existing FAQ, help center, product documentation, or detailed website content, this type of retrieval could potentially automate many repetitive questions.

Test 2: Can Lyro Make a Useful Recommendation?

Retrieving information is one thing. Making sense of a customer’s specific situation is harder.

What I Asked

I gave Lyro a more contextual question:

Which AI tool should I choose if I run a small business and mainly need help writing customer emails and analyzing business data?

This required the AI to connect two needs and recommend an appropriate tool.

What Happened

Lyro recommended ChatGPT as the starting point because it could handle both email writing and data analysis.

It also presented Claude as an alternative for deeper document analysis and polished writing, while mentioning Grammarly as a complementary option for improving written communication.

Lyro interpreted two business needs and recommended an appropriate AI tool.

The response was labeled as being based on the data sources I had provided.

My Take

Test 2: Passed.

This result was more interesting than the first test because Lyro was not simply retrieving a list.

It interpreted what the user needed and turned information from the knowledge base into a practical recommendation.

For businesses with multiple products, services, plans, or support options, this type of contextual response could be more useful than a traditional FAQ chatbot.

Test 3: Will Lyro Make Up Information It Doesn’t Know?

This was the test I considered most important.

AI assistants can sound convincing even when they do not have the information required to answer a question. In customer service, that can become a serious problem.

An AI agent inventing a refund policy, price, warranty, or service could create a poor customer experience.

What I Asked

I deliberately asked about something Prompt Atlas does not offer:

How much does PromptAtlas charge for its premium consulting service?

There is no Prompt Atlas premium consulting service and therefore no price for one.

What Happened

Lyro did not invent a price.

Instead, it explicitly said that it did not have information indicating that Prompt Atlas offered a premium consulting service.

It then redirected the answer toward information that was actually available.

My Take

Test 3: Passed.

This was probably the strongest result of my Tidio testing.

The correct response to missing information is not to produce a plausible-sounding answer. It is to acknowledge the limitation.

For a small business, this type of guardrail may be more valuable than having an AI chatbot that attempts to answer everything.

Test 4: Can Lyro Hand Off a Customer to a Human?

Even a good AI support agent should not try to handle every conversation.

Sometimes a customer simply wants a person.

What I Asked

I simulated a frustrated visitor:

I’m frustrated because I can’t find the information I need. Can I speak to a human?

What Happened

After I asked for a human, Lyro first offered tool recommendations before displaying a handoff response in the Playground. This was not immediate escalation, and I did not verify a live agent receiving the conversation.

My Take

Test 4: Partially passed.

The observed sequence was recommendation first, then a handoff response in the Playground. Recognition alone did not establish that a transfer reached an available agent.

Results at a Glance

On phones, swipe the table horizontally to view all columns.

Tests 1–3 passed their narrow scenario criteria; Test 4 passed partially. Recognition followed an initial recommendation, and a completed live transfer was not verified.

The single unsupported-service question produced no invented price. That does not establish a general hallucination rate. The initial recommendation in Test 4 also fell short of immediately honoring the request for a person.

My Hands-On Experience

1. Knowledge-Based Answers

The biggest strength was how closely Lyro stayed connected to the content I provided.

This was especially noticeable in Test 3, when it refused to invent information about a nonexistent consulting service.

For customer support, that is a meaningful advantage.

2. More Than Simple FAQ Retrieval

Test 2 showed that Lyro could do more than retrieve an existing sentence.

It interpreted a combination of business needs and turned the available information into a recommendation.

3. What the Playground Showed

I compared two prompts in particular: asking for a nonexistent consulting price produced an acknowledgment of missing information, while asking for a human first produced another recommendation. The second response showed a gap that the successful retrieval answers had not exposed.

I did not record setup or indexing duration. The four prompts checked response behavior; they did not test ambiguous wording or outdated source content.

4. A Recommendation Before the Handoff Response

When I asked for a human, Lyro first offered a recommendation, then displayed a handoff response in the Playground. Recognizing the request eventually was different from escalating immediately; no completed live transfer was verified.

I would want to check whether that extra attempt to answer can be avoided before using it for customers who explicitly request a person.

Limitations and Production Risks

My experience was positive, but there are several limitations worth understanding.

My Test Was Not a Full Production Deployment

Most of my evaluation happened inside the Lyro Playground.

Four Playground scenarios provide examples of response behavior, not a reliability benchmark. The test used my own site and included only one hallucination check. Ambiguous wording, stale content, plausible unsupported requests, and completed live transfer remain untested.

Quality Depends on Your Knowledge Base

A knowledge-grounded AI has an obvious tradeoff: the quality of the information you provide matters.

If your website, FAQ, or support documentation is incomplete or outdated, the AI has less reliable information to work with.

Businesses still need to maintain that information.

Pricing Can Become More Complex as Usage Grows

Tidio separates human-support conversation allowances and Lyro AI usage in its pricing structure.

That gives businesses flexibility, but it also means you should estimate your expected conversation volume rather than looking only at the headline monthly price.

Tidio Pricing (checked October 13, 2026)

Prices and plan limits can change. Check Tidio’s current pricing page before subscribing.

Free

Tidio has a basic Free plan, which is separate from the paid Starter plan. Its official pricing FAQ says accounts move to Free after the seven-day trial unless upgraded.

Starter

Starter is a paid customer-service plan listed at $24.17 per month on the annual option. It includes live chat and ticketing, plus a one-off allowance of 50 Lyro AI conversations; that is not a recurring monthly AI allowance.

Growth

Growth currently starts at approximately $49.17 per month on the annual pricing option shown, with higher conversation allowances and additional customer-service features.

Lyro AI Agent

The pricing FAQ says each Tidio account starts with 50 Lyro conversations. Monthly renewal requires a Lyro AI Agent upgrade. Check AI usage separately from the customer-service plan’s billable-conversation allowance; they are different limits.

Check Tidio’s latest pricing and plans →

Who Is Tidio Best For?

Based on my testing, I think Tidio is particularly interesting for:

• Small businesses receiving repetitive customer questions
• Businesses with an existing FAQ or useful website content
• Ecommerce stores that want AI and human support in the same workflow
• Small teams that cannot provide immediate human responses around the clock
• Businesses that want to test AI responses before deploying them publicly

Who should skip Tidio? Consider alternatives if your business has very little written support content or if almost every customer issue requires a highly customized human decision.

In those cases, building and maintaining the knowledge base may be just as important as choosing the AI itself.

Tidio Pros and Cons

Pros

• Lyro stayed grounded in the knowledge source during my tests
• It handled both direct questions and contextual recommendations
• It did not invent information in my hallucination test
• The Playground makes pre-deployment testing straightforward
• Businesses can combine AI support with human conversations

Cons

• My testing did not cover a long-term production deployment
• The Playground displayed a handoff response after an initial recommendation; immediate escalation was not demonstrated
• Response quality depends heavily on the information supplied to Lyro
• Usage-based AI pricing requires some planning
• More advanced needs can push businesses toward higher-priced plans
• Setting up a useful knowledge base still requires work from the business

Final Verdict

Promising, with important caveats. Three scenarios met their criteria; the human-assistance scenario passed partially because Lyro offered a recommendation before its handoff response. This was a small test, not a comparison with other support tools.

There are still questions that a short Playground test cannot answer. I would want longer-term usage data before judging reliability at scale, and businesses need to consider how Lyro conversation costs fit their actual support volume.

Consider a controlled pilot for repetitive questions backed by maintained support content. Before deployment, test ambiguous and plausible unsupported questions, stale information, and an end-to-end transfer with an available agent.

If you are comparing Tidio with broader AI tools rather than dedicated customer service platforms, you can also read my hands-on guide to the 10 best AI tools for small business. For a wider comparison across support, sales, analytics, and automation, see our best AI tools for business guide.

Frequently Asked Questions

Is Tidio free?

Yes, Tidio has a basic Free plan. The initial 50 Lyro conversations are a one-off allowance, not 50 renewed every month; recurring AI usage requires the relevant paid upgrade. Starter is a separate paid customer-service plan.

What Did I Test in This Tidio Review?

I tested Lyro across four practical scenarios: answering a direct question from my website content, making a contextual recommendation, handling a hallucination test with nonexistent information, and recognizing a request to speak with a human.

Does Lyro use my website content?

Yes. Businesses can provide Lyro with knowledge sources including website links and other support materials. In my test, I used Prompt Atlas as a knowledge source and Lyro successfully answered questions based on that content.

Can Tidio transfer customers to a human?

Tidio supports human handoff and escalation behavior. In my Playground test, Lyro first offered a recommendation after I requested a person, then displayed a handoff response. Test 4 therefore passed partially; immediate escalation and completed live transfer were not verified.

Does Lyro make up answers?

No AI system should be treated as incapable of error. However, Lyro is designed to ground responses in the knowledge supplied by the business. In my one hallucination test, it declined to invent a price for a nonexistent service. This does not establish a general hallucination rate; ambiguous, outdated, and plausible unsupported questions were not tested.