Chatbot vs Virtual Assistant: Which One Do You Actually Mean?

Three different things, one phrase

Define all three before comparing any of them. Most of the confusion in this topic is people arguing past each other because they loaded different meanings into the same two words.

Chatbot

A chatbot is software that holds a text conversation in a specific place, usually a website widget or a messaging channel, and answers questions within a defined scope.

A modern one is retrieval-based. It reads source material you give it, your website, help centre, PDFs or a knowledge base, finds the passages relevant to the question and has a language model compose an answer from them. Scope is deliberately bounded. It answers about your business, in the channels you install it in, in text.

That boundedness is the point rather than a shortcoming. A chatbot is meant to be installed in an afternoon, priced per month, and pointed at content you already have. It does not know your CRM, it does not answer the phone, and it does not decide on its own what to do next.

Examples: matram.ai, Tidio Lyro, Chatbase, Most website widgets you have used

Intelligent virtual assistant (IVA)

An intelligent virtual assistant is an enterprise conversational platform that handles voice as well as chat, integrates with back-end systems, and is built to run a contact centre's front line at scale.

The genuine technical differences from a chatbot are worth naming, because they are real. An IVA usually includes a speech layer, so the same assistant can take a phone call, transcribe it, respond in synthesised voice and pass the caller to a human agent with context attached. It ships with a dialogue builder for multi-step transactions, connectors into CRM and ticketing, agent-assist features for the humans, and analytics tuned to contact-centre metrics such as containment and average handle time.

The commercial differences are just as real. IVAs are sold, not bought. There is a discovery call, a scoping exercise, an implementation partner in many cases, and an annual contract. Deployment is measured in weeks or months rather than an afternoon, because the work is integration rather than content upload.

What actually justifies the label

• Voice, not just chat. This is the single clearest dividing line.
• Transactions against live systems: check an order, reschedule an appointment, process a return.
• Contact-centre plumbing: routing, queueing, agent handoff with full transcript and intent.
• Volume assumptions in the tens of thousands to millions of conversations a year.

Examples: NiCE Cognigy, Kore.ai, Ada, Amazon Lex, Genesys

Virtual assistant (a person)

A virtual assistant in the staffing sense is a human contractor who works remotely on administrative, scheduling, research, inbox and customer-facing tasks for one or several clients.

This meaning predates the software one and is still the more common usage outside technology circles. A VA books travel, chases invoices, cleans up a CRM, replies to email in your voice, edits a deck, calls a supplier, and makes small judgement calls all day without needing any of them specified in advance.

Comparing this to a chatbot sounds like a category error, and technically it is, but the comparison happens constantly for a practical reason. A small business owner with a growing inbox has a fixed budget and two obvious ways to spend it, and wants to know which buys more relief. That is a fair question with a real answer, covered further down.

Examples: Freelance VAs, Executive assistants working remotely, Outsourced admin agencies

Why the software labels stopped meaning much

"Intelligent virtual assistant" is partly a technical category and partly a repositioning exercise, and it helps to know which part you are reading.

Around 2016 a wave of scripted decision-tree bots gave the word chatbot a reputation for being useless. Enterprise vendors selling far more capable systems needed distance from that word, and "intelligent virtual assistant" gave it to them. The label stuck, then the same manoeuvre happened again with "AI agent". Zendesk now describes its AI agents as "autonomous systems designed to understand and autonomously resolve complex issues on any channel" that, "unlike traditional chatbots", "can reason across multi-step requests". Same move, newer noun.

It is worth having a stable definition to measure claims against. Anthropic draws the line at control: agents are "systems where LLMs dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks", while workflows are "systems where LLMs and tools are orchestrated through predefined code paths". By that standard a great many products sold as intelligent virtual assistants are workflows with a language model doing the understanding step, which is a perfectly good thing to be, and not the thing the marketing implies.

Gartner put a name to the general pattern in June 2025, coining "agent washing" for vendors rebranding assistants, RPA and chatbots without adding real agentic capability, and reported that of the thousands of vendors claiming agentic AI only around 130 were genuine. Treat the category noun as marketing until proven otherwise.

Four questions that cut through the label

• Does it answer a telephone? If yes it is an IVA or a voice platform. If no, the IVA label is doing less work than it looks.
• Where do its answers come from: your published content, or flows somebody built by hand, or both?
• Can it complete an action in another system, such as issuing a refund, without a human clicking anything?
• What does one conversation cost, and can you find that out without a sales call?

That last question is unusually diagnostic. The same distinction plays out across the whole market, which is why it is worth reading chatbot vs conversational AI and AI agent vs chatbot alongside this page. They are three cuts through the same taxonomy.

When you need none of the three

Every page like this one quietly assumes you have a problem worth buying something for. Sometimes you do not. It's worth checking before you spend a month's budget on the wrong end of a phrase.

Start at the bottom, where doing it yourself is genuinely the right answer. If a handful of enquiries arrive in a week and you reply to them from your phone between other things, there is nothing here to automate. A person answering a low volume of email personally beats every option on this page, because the replies are warm and already fast enough. Buying a chatbot at that stage does not save you time. It gives you a dashboard to check.

Friction arrives quietly and it does not announce itself. You notice you have typed the same answer about delivery windows several times this week, and you have started copying it out of the last reply rather than writing it again. Nothing is broken. But the copying is the signal, and the cheapest fix at this stage is usually not software at all. Write the answer once, publish it as a page, link to it. If that holds, stop there and keep your money.

It turns into a liability when the questions start arriving faster than the person answering them. Replies land the next morning instead of within the hour. Somebody asks twice because the first message went nowhere. The person who used to enjoy answering starts to resent the inbox, and answer quality drops in a way nobody notices, because nobody is measuring it. That is the point where the repetitive slice is genuinely costing you something, and the rest of this page becomes worth reading.

The case where none of it works

And then there is the shape of business where none of this holds at any volume. If almost every enquiry is specific to one account or one contract, there is no repetitive slice to remove. A chatbot pointed at your published content will answer almost nothing, because almost nothing anybody asks you is published anywhere. An IVA will not rescue that either unless the answers live in a system it can query on the caller's behalf, which is an integration project, not a purchase. What you actually need is another pair of hands, or a decision to write down more of what you know. Both are slower than buying software and both work better.

Chatbot vs intelligent virtual assistant, side by side

Ignore the rows about features for a moment and look at the last three. The buying process, the contract length and the pricing transparency tell you more about which category a product belongs to than any capability list does.

The "$33k and up" figure is not a guess. It is the lowest of the annual IVA listings we could verify, and the next section shows the workings. matram.ai sits firmly in the left column: it is a chatbot, it has no voice channel, and its plans and limits are on a public page you can read without talking to anybody.

What IVA platforms actually cost

Cognigy and Kore.ai both return an HTTP 404 for the pricing URL on their own websites. Ada's pricing page exists but shows no rates, only a link to book a consultation. That is not an accident, it is how the category sells. The figures below come from AWS Marketplace, where the same vendors are obliged to list a price, checked on 20 July 2026.

Two things follow from this table. First, the entry point for a real IVA platform is roughly $33,000 to $43,000 a year, and that is the advertised list price before implementation work, not after. Second, unit economics only look sane at volume: $43,080 across 60,000 conversations is about $0.72 each, which is reasonable if you genuinely have 5,000 conversations a month and ruinous if you have 300.

Compare that with the outcome-based pricing the helpdesk vendors have moved to. Fin, the company formerly known as Intercom, charges $0.99 per outcome with a 50-outcome monthly minimum, and bills one outcome per conversation. Both models are defensible. Both are worth putting into the chatbot ROI calculator with your own volume before you sign anything, because the ranking flips depending on how many conversations you actually have. If you are pricing the helpdesk route specifically, Intercom pricing and Zendesk chatbot pricing break those down.

Chatbot vs a human virtual assistant

Different comparison entirely. We are not going to quote hourly rates, because they vary by region, seniority and scope so widely that any single number would be a fabrication dressed as research. Compare them on structure instead.

A human virtual assistant is good at work that is varied, low in volume per task type, and full of small judgement calls. Deciding which of three suppliers to chase first. Noticing that an email is from an angry customer and handling it differently. Reformatting a spreadsheet that arrived in an unexpected shape. None of this can be specified in advance, which is exactly why a person does it well and software does not.

A chatbot is good at the opposite shape of work: high volume, low variety, no judgement required. Forty questions asked four thousand times. Opening hours, shipping policy, does it integrate with X, what is in the mid tier, do you ship to Ireland. A chatbot answers all of them simultaneously at three in the morning and gives the same answer every time, which is a feature for a policy question and a limitation for anything requiring discretion.

Where each one actually wins

• Concurrency. A person handles one conversation at a time. Software handles all of them at once, and your bill does not change because two hundred visitors arrived together instead of one after another.
• Coverage. A VA works their hours in their timezone. A chatbot covers the nights and weekends, which is when questions arrive at a site whose staff have gone home.
• Judgement. A VA can tell that a request is unusual and stop. A chatbot answers the question it was asked, which is why escalation to a human matters more than raw answer quality.
• Ramp-up. A VA needs weeks of context and gets better indefinitely. A chatbot is useful the day you point it at your documentation and gets better only when you improve that documentation.
• Breadth. A VA does dozens of unrelated tasks. A chatbot does one thing.

The honest conclusion is that these are complements rather than substitutes, and most small businesses that run both end up with the chatbot taking the repetitive front line and the VA doing the work that survives it. The failure mode we see most often is hiring a person to answer the same twelve questions forever, which is expensive, boring for them, and the single easiest thing to automate. The reverse failure is expecting a chatbot to exercise discretion. If you are sizing this decision for a small team, the small business chatbot guide goes into what a bot realistically absorbs, and lead capture covers the part where the bot hands a qualified person to a human instead of trying to close them.

Which one you should actually buy

Three genuinely different purchases, three different triggers. The mistake is buying up a category because it sounds more capable.

If two of these describe you, buy two. A chatbot on the site and a person handling escalations is a common and sensible pairing. A chatbot and an IVA in the same organisation is usually a sign that the chatbot was bought by marketing and the IVA by the contact centre, which happens more often than either team admits.

What choosing the wrong one costs you

The invoice is the smallest part of the bill, and it is the only part anybody compares.

Buy an enterprise platform when a chatbot would have done and the contract is not the real damage. The real damage is the gap between signing and going live, measured in weeks or months, during which every question you were trying to answer is still arriving and still going unanswered. Somebody on your team becomes the owner of that implementation. Their actual job pauses. By the time the thing works, the pages it was built to explain have changed, and the first task of the new system is catching up with the business it was supposed to help.

Going the other way costs less money and more trust. Put a bot in front of work that needed a person and customers meet something that cannot do the thing they asked, decide your support is a dead end, and stop asking. You won't see that in your analytics, because the people it happened to have gone. And there is a habit cost on the other side too: hire someone to answer the same small set of questions forever and your team learns that this is work humans do here, which makes automating it later feel like taking a job away rather than giving somebody a better one.

Then there is the migration you end up doing twice. Answers typed into a vendor's configuration are answers that live inside that vendor, so moving suppliers means retyping your own knowledge, and you will find that some of it only ever existed in there. The teams that come out of this well are usually the ones whose answers were on their own website all along, because for them switching tools is a re-crawl rather than a rebuild. So ask the question the demo never asks. If this purchase turns out to be wrong in a year, what have you lost, and what do you still own?

Where matram.ai is not the right tool

We would rather say this on the page than after you have paid for a month.

matram.ai is an AI chatbot. It is not an intelligent virtual assistant and we do not describe it as one. There is no voice channel, no telephony, no IVR replacement, no speech recognition and no contact-centre routing. If the problem you are solving is a phone queue, nothing on our pricing page will help you, and the vendors to look at are NiCE Cognigy, Kore.ai and Amazon Lex, in roughly that order of hand-holding versus build-it-yourself.

Nor is it a helpdesk. If you need ticketing, SLAs, macros and a full agent workspace with an AI layer on top, that is Zendesk or Intercom territory, and we cover the trade-offs honestly in the Zendesk alternative and Intercom alternative comparisons rather than pretending the question does not arise.

What it is genuinely good at

• Answering text questions from your own content, with a citation to the source page on every answer.
• Strict mode, so the bot answers only from material you approved and says it does not know rather than improvising.
• Going live in an afternoon: crawl a URL, import a sitemap, upload PDFs up to 10MB or paste text, then one line of JavaScript on the site.
• Sitting in the channels your team already uses, including Slack, Messenger, Zendesk, Freshchat, Crisp, Zoho SalesIQ and Google Chat.
• Capturing leads, exporting them to CSV, and escalating to a person through a shared team inbox when the answer needs a human.
• Answering in 95 or more languages without a separate build per language.

That is a smaller claim than "intelligent virtual assistant", and deliberately so. The category inflation described earlier is exactly what makes these pages hard to research, and we are not going to add to it to win a comparison.

Frequently asked questions

Sources

• AWS Marketplace vendor listings (accessed 20 July 2026) - List prices for Ada Strategic, NiCE Cognigy Basic 5K pm, the Cognigy private-SaaS enterprise licence and Kore.ai per-session rates.
• Ada pricing page - Ada publishes no rates on its own site, only a consultation link.
• Cognigy pricing page (HTTP 404, 20 July 2026) - Cognigy does not publish pricing on its own website.
• Kore.ai pricing page (HTTP 404, 20 July 2026) - Kore.ai does not publish pricing on its own website.
• Fin pricing page (accessed 20 July 2026) - $0.99 per outcome, 50-outcome monthly minimum, one outcome billed per conversation.
• Zendesk, AI agents - Zendesk's own definition of its AI agents as autonomous systems that reason across multi-step requests.
• Anthropic, "Building effective agents", 19 December 2024 - The distinction between agents and workflows used to test category claims.
• Gartner press release, 25 June 2025 - The term "agent washing" and the finding that only around 130 of thousands of claimed agentic AI vendors were genuine.