Straight answer: a chatbot answers questions on your website, an AI assistant does actual work across your tools, and most businesses buy the expensive one when they only need the cheap one. The difference isn’t intelligence, it’s what the thing is allowed to touch. Get that wrong and you’ll overpay for a tool that still can’t book an appointment.
What the two things are (and why the names get muddled)
A chatbot lives on one surface, usually your website or a Facebook Messenger box, and answers questions from a script or a knowledge base. Think Intercom Fin, Drift, or Tidio. It’s reactive. Someone types, it replies, the conversation ends.
An AI assistant is different because it has access to more than one system and it can take multi-step action. ChatGPT with a connected calendar, Claude with a Projects folder full of your contracts, or Microsoft Copilot pulling from your Outlook, Teams and SharePoint all at once. You ask it to draft a proposal, pull last month’s figures, and email a client, and it does all three without you copying and pasting between four tabs.
The confusion happens because vendors now slap “AI assistant” on things that are still just chatbots with better manners. I’ve seen software described as an “AI employee” that, when you poke it, can only do one job: answer FAQs. That’s a chatbot wearing a name tag.
The test I ran with my own inbox
Last spring I had a backlog of 40-odd speaking enquiries sitting in my inbox, each one needing a reply, a date check against my calendar, and a follow-up if nothing came back in five days. I tried two setups over two weeks.
Week one: a website chatbot handled the first reply only, then handed everything to me. I still did the calendar check and every follow-up by hand. It took roughly 25 minutes per enquiry once you count the back and forth.
Week two: I set up an assistant workflow using Claude connected to my calendar and email drafts, with a simple rule: check availability, draft a reply, flag anything needing a human decision, and queue a follow-up after five days of silence. That dropped to about 8 minutes of my actual time per enquiry, mostly just approving drafts before they went out.
Same volume of enquiries, same me, one third of the time. That’s not a hypothetical, that’s what happened on my own calendar, and it’s the same pattern I saw when I let an AI answer my business phone for 30 days: the tool that could act across systems saved hours, the tool that could only reply saved minutes.
Where a plain chatbot still wins
I’m not going to pretend assistants are always the answer, because they’re not, and here’s where a chatbot beats an assistant flat out.
- Cost per interaction. Fin from Intercom charges around $0.99 per resolved conversation. A well-trained assistant workflow with tool access can run into pounds per task once you factor in the connected apps and the setup time.
- Simple, high-volume, low-stakes questions. “What are your opening hours” or “do you ship to Ireland” doesn’t need memory, tools or judgement. A chatbot answers it instantly and correctly nearly every time.
- Speed to launch. You can have a chatbot live on your site this afternoon. An assistant that touches your calendar, CRM and inbox takes days of configuration, sometimes longer if your systems don’t talk to each other cleanly.
If your business problem is “our website visitors ask the same six questions and we’re slow to reply,” don’t buy an assistant. You’re buying a hammer for a job that needed a screwdriver, and you’ll pay assistant prices for chatbot results.
Where an assistant earns its keep
The assistant model pulls ahead the moment a task needs more than one step or more than one system. Chasing an unpaid invoice needs the assistant to check the accounting software, draft a tone-appropriate email, and log the follow-up date. Preparing a client proposal needs it to pull previous work, check current pricing, and format a document. A chatbot cannot do any of that because it has nothing to reach into beyond its own script.
Memory matters here too. An assistant that remembers a client always wants invoices sent on the last Friday of the month, or that a particular supplier needs three days’ notice, behaves differently from a chatbot that starts every conversation from zero. That accumulated context is the actual value, not the fact it sounds more human.
The bit nobody wants to say out loud
Here’s the uncomfortable part. Most businesses that have bought an “AI assistant” in the last year are still using it as a chatbot, and paying assistant prices for it. They’ve connected nothing, given it no tools, and just typed questions into a chat window one at a time, same as they’d do with a free version of ChatGPT. The tool isn’t the bottleneck. The workflow design is.
I’ve sat with clients who paid £200 a month for an “AI assistant” subscription and used it exactly like Google, typing a question, reading an answer, closing the tab. That’s not an assistant failing to deliver, that’s a business that never set up the actions that make it one. The software vendors won’t tell you this because the subscription renews either way.
This is also why a badly configured chatbot can quietly cost you five-star reviews while everyone assumes the tech itself is the problem, when it’s usually three minutes of missing setup.
A decision framework that holds up
Ask these four questions before you buy anything:
- Does the task need more than one system? If yes, lean assistant. If it’s answered entirely from a knowledge base, lean chatbot.
- Does the task repeat with the same person over time? Memory-dependent tasks (recurring clients, ongoing projects) favour an assistant. One-off questions favour a chatbot.
- What’s the volume? High-volume, low-complexity interactions (hundreds of FAQ replies a day) are cheaper through a chatbot. Low-volume, high-complexity tasks (a handful of proposals a week) are cheaper through an assistant, because your time is the expensive part, not the software.
- Who checks the output? If nobody reviews what goes out, keep tasks simple and scripted. Assistants making judgement calls without a human glance at the draft are how you end up with an email that quotes the wrong price, which ties directly into why AI pricing advice quietly trains small businesses to undercharge when nobody’s checking the numbers it suggests.
What this costs in real terms
A chatbot subscription for a small business typically runs £20 to £80 a month depending on volume. An assistant setup, once you add the connected tools (calendar, email, CRM, accounting software) and the time to configure the workflows, tends to land between £150 and £500 a month in software alone, before you count the hours spent building it or paying someone to build it for you. That range lines up with what I found when I broke down what an AI “employee” really costs a small business in 2026, and the gap between the two numbers is exactly why so many businesses default to the chatbot even when the assistant would save them more time.
If you’re not sure which side of that line your business sits on, that’s a fair thing to get a second opinion on rather than guessing and paying twice. It’s the kind of decision I walk clients through as an AI consultant for small businesses, mapping the actual tasks before choosing the tool, not the other way round.
The mistake I see most often
Businesses buy the assistant first and figure out the workflow later, which is backwards. Map the five tasks eating the most hours in your week before you look at a single tool. In my case that was speaking enquiries, invoice chasing, and proposal drafting. For a client running a small marketing agency I worked with earlier this year, it was client onboarding emails and monthly reporting, two tasks that took her team roughly 9 hours combined every month before we set up an assistant workflow that cut it to under 2. Same principle every time: name the task, then pick the tool built for that task, not the other way round.
If you want a running sense of how fast this space moves and what’s launching versus what’s just marketing noise, I cover the real releases each week in AI news this week for small business and the week before that, because half the “new assistant” announcements are chatbots with a rebrand.
Frequently asked questions
Is ChatGPT a chatbot or an AI assistant?
On its own, with no connected tools, ChatGPT behaves like a very capable chatbot: you ask, it answers, nothing else happens. Once you connect it to your calendar, files or email through custom actions or a paid plan with integrations, it becomes an assistant because it can now take steps rather than just reply.
Can a small business afford an AI assistant, not just a chatbot?
Yes, and it’s often cheaper than expected if you scope it to one or two repeating tasks rather than trying to automate everything at once. A single-workflow assistant setup, such as invoice chasing or appointment scheduling, can cost as little as £30 to £80 a month on top of tools you may already own.
Do I need to replace my chatbot with an assistant?
No, and for most businesses the two should sit side by side. Keep the chatbot handling website FAQs and simple lead capture, and use an assistant for the internal, multi-step tasks like proposals, follow-ups and reporting where a scripted reply isn’t enough.
What’s the biggest sign a business bought the wrong one?
If staff are still copying information between the AI tool and three other systems by hand, that’s the tell. It means you paid for an assistant but set it up like a chatbot, and the connections that would have made it worth the money were never switched on.
Related reading: Where to Find Free Contract Templates You Can Edit in Word and Where to Find Genuine Remote Job Leads Online (Not the Recycled Listings).
For the bigger picture, see my full guide to AI marketing.
