Different AI Providers and How Their Responses Differ with Same Prompting

CloseBot’s reimagined Agent Builder allows you to select different AI providers to act as your new hires (Personas). These personas will execute your lead qualification and booking (Job Flows). There are many things that set CloseBot apart, but one reason marketing agencies love CloseBot is the ability to choose between AI providers. You can even prioritize your favorites and set others as fallbacks in case your primary choice fails. This is something that few alternatives offer 😳

Naturally, the next question we usually hear is: “How are each of your providers different?”

Persona Setup

We decided to do some split testing to show you firsthand how different providers respond with the exact same persona settings. Here you see five different personas, one for each of our supported AI providers. CloseBot’s personas, which are separated from the job flow, make it easy to split test and see how different persona-related settings affect conversation conversions.

When creating a persona, we can give it a name, tell it how often to split up large messages into smaller ones, and even how often to add typos. Personas also have settings for tone (voice) and clarifying instructions on how to respond. For this test, all personas had the same settings except for the AI provider, as shown below.

Job Flow Setup

The job flow is what instructs the agent on what to do. For this test, the job flow was intentionally simple. It was designed to gather the contact’s name (first and last) and business name. The agent was then free to continue the conversation after collecting that information. We kept the job flow simple because we were primarily testing the persona in this experiment.

Conversation Testing

The test was always run with the same responses, occasionally tweaked slightly to ensure the conversation made sense based on the latest response from the AI agent.

OpenAI

OpenAI was the trailblazer in the AI provider race when it released its GPT API in 2022. CloseBot initially integrated solely with OpenAI, and most agent builder platforms—including GoHighLevel’s AI Employee—only integrate with OpenAI.

During our time working with OpenAI, we’ve noticed a few natural tendencies:

• Likes to start messages with short words like “Hi!”, “Great”, “Thanks”, etc.
• Is naturally more concise than other AI providers
• Tends to sound more formal

Gemini

Gemini, a product of Google, was formally known as Bard. At launch in 2023 Bard utilized LaMDA, then PaLM2, and ultimately Gemini (hence the name change) to enable conversational AI. LaMDA was announced in 2021, but Google refrained from providing public access to their conversational AI until OpenAI’s tools pushed them to accelerate the Bard release in ‘23. Although Gemini is not as popular as OpenAI, it’s a very capable model.

From our lead qualification tests, we’ve noticed:

• Less likely to start responses with “Great”, “Hi”, etc.
• Naturally concise, which is refreshing compared to verbose models
• Shows a bit more personality and unpredictability in its responses

DeepSeek

DeepSeek rose to popularity in 2024 after releasing a ChatGPT competitor at a fraction of the cost. It remains popular due to its affordability and the ability to run it locally—appealing to users concerned about data privacy since it’s hosted in China.

Here’s what we’ve noticed:

• Frequently starts messages with “Hi!”, “Great”, “Thanks”
• More wordy than Gemini and OpenAI, though not the most verbose
• Overuses punctuation, making responses occasionally awkward!

Grok

Grok, a product of X (formerly Twitter), was released in late 2023. Its claim to fame was its humorous tone. Elon Musk created Grok after a fallout with Sam Altman at OpenAI, citing OpenAI’s shift away from its open-source roots. Thus, Grok is open-sourced.

What we’ve noticed:

• Frequently starts responses with statements like “Hey there!” or “Nice to meet you,” which can feel repetitive
• Most verbose of the five providers tested
• When asking questions, it often explains why it’s asking (e.g., “Could you share your last name with me? I’d like to make sure I have all your details correct.”)

Anthropic

Anthropic has been riding the coat tails of OpenAI since release in terms of popularity. They are known for their focus on AI safety and have been since they released the first of their Claude models in 2023. Anthropic was the second AI Provider added to the CloseBot Classic app and it’s what we chose to run the chat widget on our home page before the widget was removed to accommodate our own migration to CloseBot V2.

During our time playing with Anthropic we’ve noticed:

• Similar to Grok, it likes to start responses with a short statement.  This tends to make the conversation feel more choppy.
• It likes to use people’s names in responses… sometimes on almost every response if left un-edited.
• It can feel quite formal by default, although not quite as formal as OpenAI’s models

Summary

CloseBot’s newest update releasing soon brings many new features that we are excited about.  One of the main ones is the ability for you to select any of these 5 different AI Providers 💪🤖

Grab your free account today and you’ll instantly get access to both OpenAI and Anthropic (paid plans get access to the additional providers).  Test and see which ones you prefer and sign up today!