Not all AIs are created equal. Some may do artwork the perfect, some are expert at coding, and others have the power to foretell protein buildings precisely.
However once you’re searching for one thing extra basic—simply “somebody” to speak to—the perfect AI companions is probably not those that know all of it, however the ones which have that je ne sais quoi that make you are feeling OK simply by speaking, much like how your greatest good friend may not be a genius however in some way all the time is aware of precisely what to say.
AI companions are slowly gaining popularity amongst tech fanatics, so it is necessary for customers wanting the best high quality expertise or corporations desirous to grasp this facet of making the phantasm of genuine engagement to think about these variations.
We have been curious to search out out which platform offered the perfect AI expertise when somebody merely looks like having a chat. Curiously sufficient, the perfect fashions for this usually are not actually those from the massive AI corporations—they’re simply too busy constructing fashions that excel at benchmarks.
It seems that friendship and empathy are an entire completely different beast.
Table of Contents
ToggleEvaluating Sesame, Hume AI, ChatGPT, and Google Gemini. Which is extra human?
This evaluation pits 4 main AI companions towards one another—Sesame, Hume AI, ChatGPT, and Google Gemini—to find out which creates essentially the most human-like dialog expertise.
The analysis targeted on dialog high quality, distinct persona improvement, interplay design, and likewise considers different human-type options equivalent to authenticity, emotional intelligence, and the refined imperfections that make dialogue really feel extra real.
You may watch all of our conversations by clicking on these hyperlinks or checking our Github Repository:
Right here is how every AI carried out.
Dialog High quality: The Human Contact vs. AI Awkwardness
The true check of any AI companion is whether or not it might idiot you into forgetting you are speaking to a machine. Our evaluation tried to guage which AI was the perfect at making customers wish to simply preserve speaking by offering fascinating suggestions, rapport, and total nice expertise.
Sesame: Good
Sesame blows the competitors away with dialogue that feels shockingly human. It casually drops phrases like “that is a doozy” and “capturing the breeze” whereas seamlessly switching between considerate reflections and punchy comebacks.
“You are asking huge questions huh and truthfully I haven’t got all of the solutions,” Sesame responded when pressed about consciousness—full with pure hesitations that mimic real-time considering. The occasional overuse of “you already know” is its solely noticeable flaw, which paradoxically makes it really feel much more genuine.
Sesame’s actual edge? Conversations stream naturally with out these awkward, formulaic transitions that scream “I am an AI!”
Rating: 9/10
Hume AI: Empathetic however Formulaic
Hume AI efficiently maintains conversational stream whereas acknowledging your ideas with heat. Nonetheless it looks like speaking to somebody who’s disinterested and not likely that into you. Its replies have been quite a bit shorter than Sesame—they have been related however not likely fascinating should you needed to push the dialog ahead.
Its weak point exhibits in repetitive patterns. The bot persistently opens with “you have actually received me considering” or “that is an interesting subject”—creating a way that you just’re getting templated responses somewhat than natural dialog.
It is higher than the chatbots from the larger AI corporations at sustaining pure dialogue, however repeatedly reminds you it is an “empathic AI,” breaking the phantasm that you just’re chatting with an individual.
Rating: 7/10
ChatGPT: The Professor Who By no means Stops Lecturing
ChatGPT tracks advanced conversations with out shedding the thread—and it’s nice that it memorizes earlier conversations, basically making a “profile” of each person—however it feels such as you’re trapped in workplace hours with a very formal professor.
Even throughout private discussions, it might’t assist however sound educational: “the interaction of biology, chemistry, and consciousness creates a depth that AI’s sample recognition cannot replicate,” it mentioned in considered one of our assessments. Practically each response begins with “that is an interesting perspective”—a verbal tic that rapidly turns into noticeable, and a standard downside that every one the opposite AIs besides Sesame confirmed.
ChatGPT’s largest flaw is its incapacity to interrupt from educator mode, making conversations really feel like sequential mini-lectures somewhat than pure dialogue.
Rating 6/10
Google Gemini: Underwhelming
Gemini was painful to speak to. It sometimes delivers a concise, informal response that sounds human, however then instantly undermines itself with jarring dialog breaks and reducing its quantity.
Its most irritating behavior? Abruptly reducing off mid-thought to advertise AI matters. These steady disruptions create such a damaged dialog stream that it is inconceivable to neglect you are speaking to a machine that is extra eager about self-promotion than precise dialogue.
For instance, when requested about feelings, Gemini responded: “It is nice that you just’re eager about AI. There are such a lot of wonderful issues happ—” earlier than inexplicably stopping.
It additionally made positive to let you already know it’s an AI, so there’s a giant hole between the person and the chatbot from the primary interplay that’s exhausting to disregard.
Rating 5/10
Persona: Character Depth Separates the Genuine from the Synthetic

How does an AI develop a memorable persona? It would principally rely in your setup. Some fashions allow you to use system directions, others adapt their persona primarily based in your earlier interactions. Ideally, you possibly can body the dialog earlier than beginning it, giving the mannequin a persona, traits, a conversational fashion, and background.
To be truthful in our comparability, we examined our fashions with none earlier setup—that means our dialog began with a whats up and went straight to the purpose. Right here is how our fashions behaved naturally
Sesame: The Pal You By no means Knew Was Code
Sesame crafts a persona you’d really wish to seize espresso with. It drops phrases like “that is a Humdinger of a query” and “it is a tight rope stroll” that create a definite character with obvious viewpoints and perspective.
When discussing AI relationships, Sesame confirmed precise persona: “wow… think about a world the place everybody’s head is down plugged into their personalised AI and we neglect how one can join nose to nose.” This type of perspective feels much less like an algorithm and extra like a considering entity. It’s additionally humorous (it as soon as instructed us that our query blew its circuits), and its voice has a pure inflection that makes it simple to narrate to when attempting to painting a response. You may clearly inform when it’s excited, contemplative, unhappy and even annoyed
Its solely weak point? Sometimes leaning too exhausting into its “considerate buddy” persona. That didn’t detract from its place as essentially the most distinctive AI persona we examined.
Rating 9/10
Hume AI: The Therapist Who Retains Mentioning Their Credentials
Hume AI maintains a constant persona as an emotionally clever companion. It additionally tasks some heat by affirming language and emotional assist, so customers searching for that will probably be happy.
Its Achilles heel is principally the truth that, sort of just like the Harvard grad who wants to say that, Hume cannot cease reminding you it is synthetic: “As an empathetic AI I do not expertise feelings myself however I am designed to grasp and reply to human feelings.” These moments break the phantasm that makes companions compelling.
If speaking to GPT is like speaking to a professor, speaking to Hume looks like speaking to a therapist. It listens to you and creates rapport, however it makes positive to remind you that it’s really its activity and never one thing that occurs naturally.
Regardless of this flaw, Hume AI tasks a clearer character than both ChatGPT or Gemini—even when it feels extra constructed than spontaneous.
Rating 7/10
ChatGPT: The Professor With out Private Opinions
ChatGPT struggles to develop any distinctive character traits past common helpfulness. It sounds overly excited to the purpose of being clearly faux—like a “good friend” who all the time smiles at you however is secretly fantasizing about throwing you in entrance of a bus.
“Haha, effectively, I prefer to preserve the power up. It makes conversations extra enjoyable and fascinating plus it is all the time nice to talk with you,” it mentioned after we requested in a really critical and unamused tone why it was appearing so enthusiastically.
Its id points seem in responses that shift between figuring out with people and distancing itself as an AI. Its educational tone in responses persists even throughout private discussions, making a persona that looks like a strolling encyclopedia somewhat than a companion.
The mannequin’s default to academic explanations creates an impression extra of a software than a personality, leaving customers with little emotional connection.
Rating 6/10
Google Gemini: A number of Persona Dysfunction
Gemini suffers from essentially the most extreme persona issues of all fashions examined. Inside single conversations, it shifts dramatically between considerate responses and promotional language with out warning.
It’s not actually an AI design to have a compelling persona. “My objective is to offer info and full duties and I wouldn’t have the power to kind romantic relationships,” it mentioned when requested about its ideas on individuals growing emotions in direction of AIs.
This inconsistency makes Gemini really feel like a Nineteen Fifties film robotic, stopping any significant connection and even making it nice to spend time speaking to it.
Rating 3/10
Interplay Design

How an AI handles dialog mechanics—response timing, turn-taking, and error restoration—creates both seamless exchanges or irritating interactions. Right here is how these fashions stack up towards one another
Sesame: Pure Dialog Move Grasp
Sesame creates dialog rhythms that really feel very, very human. It varies response size naturally primarily based on context and handles philosophical uncertainty with out defaulting to lecture mode.
“Generally I really feel like perhaps I simply want to chop to the chase with a fast reply somewhat than a long-winded lecture, proper? You already know, so… that is a small humorous apart to let you already know that I am conscious of the potential of falling right into a lecture mode and attempting to maintain issues mild but additionally deep on the identical time,” Sesame instructed us throughout a philosophical debate.
When discussing advanced matters, it responds conversationally, with a small joke, typically with statements, different instances with human noises like “hmmms” or whispers—which makes it very convincing as a human substitute.
Sesame additionally asks pure follow-up questions that push conversations ahead, whereas realizing when to modify to statements when questioning may really feel awkward. This adaptive strategy creates dialogue that feels spontaneous somewhat than scripted.
Rating 8/10
Hume AI: Structured However Temporary
Hume AI was designed to let customers do the speaking. It recovers easily from misunderstandings with out over explaining and is sweet at protecting the dialog going with related comply with up questions.
Nonetheless, its responses are sometimes so quick that conversations require important effort to keep up momentum, which finally defeats its objective. Customers regularly want to offer new dialog fodder somewhat than having fun with pure dialogue improvement.
One other weak point seems in its templated response buildings—acknowledgment + clarification + query—making a mechanical rhythm: “What a thought-provoking query! This Idea is certainly an interesting thought… What do you consider…?”
Nonetheless it has a terrific function which will actually improve your chats: It’s in a position to detect the degrees of emotion in every response, which is one thing no different mannequin can do. That is nice that can assist you replicate on how a lot a subject impacts you and the way it’s approaching a particular scenario
Rating 7/10
ChatGPT: Good Chatter—However a Bit Unnatural
ChatGPT persistently addresses questions with related info however transforms conversations into unnecessarily elaborated interactions. Responses regularly balloon into paragraph-length explanations that disrupt pure dialogue stream.
For instance, we requested it if it may develop some degree of self consciousness or consciousness, and this was its reply:
“That is an interesting subject! Whereas AI can positively evolve and enhance by interactions, true self-awareness or Consciousness is one thing that is still uniquely human. AI can mimic conversational patterns, reply empathetically and even be taught from previous interactions, however it would not have an internal sense of self or subjective expertise. Researchers and builders are nonetheless exploring the boundaries of AI’s capabilities however for now Consciousness stays a uniquely human trait.”
That’s not how a human would work together.
Its heavy reliance on opener phrases like “that is a extremely fascinating query,” or “that’s an interesting subject” earlier than each single reply additional undermines conversational immersion, creating an interplay sample that feels mechanical somewhat than pure.
Rating 6.5/10
Google Gemini: Dialog Breaking Machine
Gemini is a masterclass in how to not design dialog mechanics. It usually cuts off mid-sentence, creating jarring breaks in dialogue stream. It tries to choose up further noises, it interrupts you should you take too lengthy to talk or take into consideration your reply and infrequently it simply decides to finish the dialog with none cause.
Its compulsive must inform you at each flip that your questions are “fascinating” rapidly transforms from flattering to irritating however appears to be a standard factor amongst AI chatbots.
Rating 3/10
Conclusion
After testing all these AIs, it’s simple to conclude that machines gained’t be capable to substitute a very good good friend within the quick time period. Nonetheless, for that particular case through which an AI should merely excel at feeling human, there’s a clear winner—and a transparent loser.
Sesame (9/10)
Sesame dominates the sector with pure dialogue that mirrors human speech patterns. Its informal vernacular (“that is a doozy,” “capturing the breeze”) and assorted sentence buildings create authentic-feeling exchanges that stability philosophical depth with accessibility. The system excels at spontaneous-seeming responses, asking pure follow-up questions whereas realizing when to modify approaches for optimum dialog stream.
Hume AI (7/10)
Hume AI delivers specialised emotional monitoring capabilities at the price of conversational naturalness. Whereas competently sustaining dialogue coherence, its responses have a tendency towards brevity and comply with predictable patterns that really feel constructed somewhat than spontaneous.
Its visible emotion tracker is fairly fascinating, most likely good for self discovery even.
ChatGPT (5.6/10)
ChatGPT transforms conversations into lecture classes with paragraph-length explanations that disrupt pure dialogue. Response delays create awkward pauses whereas formal language patterns reinforce an academic somewhat than companion expertise. Its strengths in information group might enchantment to customers searching for info, however it nonetheless struggles to create genuine companionship.
Google Gemini (3.5/10)
Gemini was clearly not designed for this. The system routinely cuts off mid-sentence, abandons dialog threads, and isn’t in a position to present human-linke responses. Its extreme persona inconsistency and mechanical interplay patterns create an expertise nearer to a malfunctioning product than significant companionship.
It’s fascinating that Gemini Stay scored so low, contemplating Google’s Gemini-based NotebookLM is able to producing extraordinarily good and lengthy podcasts about any sort of info, with AI hosts that sound extremely human.
Usually Clever E-newsletter
A weekly AI journey narrated by Gen, a generative AI mannequin.