A marketing chatbot is a question-answering job, not a channel

The short version

• Chatbot marketing is a question-answering job, not a broadcast channel. The bot earns attention by resolving something, not by starting a conversation nobody asked for.
• A marketing chatbot that cannot reach live data or maintained content has no advantage over the general AI assistant the customer already has open.
• The useful jobs are narrow: answer pre-purchase questions, qualify, book the next step, and log what it could not answer.
• Measure booked calls, qualified handovers and cost per resolved conversation. Conversations started goes up when the bot interrupts more.
• If two sentences on your pricing page would end most enquiries, write the sentences instead of building a chatbot.

What does a marketing chatbot do that a form does not?

It answers before it asks. A form is a collection device: it takes a name and an email and promises a reply later. A chatbot can resolve the thing that stopped the buyer, then collect details once there is a reason to give them. That order is the whole advantage, because the unanswered question is the reason the person is still on the page.

So the test for any chatbot for marketing is simple. Write down the ten questions people ask immediately before they buy from you. Whether you deliver to their country, whether it integrates with the system they already run, whether you take their insurer, what the real price is at their volume. If the bot answers those ten accurately, it is doing marketing work. If it cannot, it is a form with a typing animation.

This also tells you where to put it. Chatbots in marketing earn their place on pages where a decision is being made, which means pricing, product detail, delivery and eligibility. They earn nothing on a blog post, where the reader has no pending decision and no patience.

Which marketing jobs are worth giving a chatbot?

Four, roughly. Answering pre-purchase questions from your own content. Qualifying an enquiry enough to route it to the right person. Booking the next step, usually a call, a quote or a visit. And capturing what people asked, including the questions nobody on your team knew were being asked.

Everything else on the usual list is weaker than it looks. Outbound messages into a chat channel nobody opted into are spam with better delivery. A quiz used as a lead gate mostly annoys the people who were going to convert anyway. A bot given a personality but no data produces conversations that read like a holding pattern. Using chatbots for marketing goes wrong most often when the bot is handed a target of 'conversations started' instead of questions answered.

It is also fine to decide the answer is no. If your buyers ask two questions and both are settled by a sentence on the pricing page, write the sentence. A chatbot is worth building when the question space is wide, the answers live in content you already maintain, and the cost of a wrong answer is recoverable.

What a marketing chatbot should be able to do on day one

• •Answer the ten questions that come immediately before a purchase, from your own published content.
• •Say that it does not know, and offer a route to someone who does.
• •Hand over with the conversation attached, so the buyer does not repeat themselves.
• •Log every question it could not answer, and send that list to whoever writes the pages.
• •Tell the person they are talking to software, without being asked.

Why do customers skip the company chatbot?

Because they have something better open in another tab. A Gartner survey of 3,566 B2B and B2C customers, conducted in February and March 2026, found customers are three times more likely to use a third-party generative AI tool for customer service than the chatbot the company provides. That is the competitive position now. Your bot is not being compared to a phone queue. It is being compared to a general assistant that writes well and never loops.

A general assistant beats a weak company bot at nearly everything except the one thing that matters, which is your data. It does not know your stock level, this customer's order, your delivery cut-off for their postcode, or what your policy says this month. That is the only durable reason for a marketing chatbot to exist, and it is also the reason most of them disappoint: they are wrapped around an FAQ page the visitor could have read themselves.

The practical consequence is a build decision, not a copywriting decision. If the bot cannot reach a live system or a maintained body of content, it has no advantage, and a visitor who tests it once with a real question will not test it twice.

Does a chatbot actually sell anything?

There is evidence that it can, and it is narrower than the way it gets quoted. In a randomised field experiment targeting 6,000 customers of a financial services company, reported in Marketing Science, customers were assigned to human agents or to chatbots, and the finding was that artificial intelligence can improve sales by four times compared to some human employees. The experiment also varied the timing of bot disclosure, which tells you disclosure was expected to change the outcome.

Read it for what it is. One company, one industry, one set of scripts, published in 2019, with the release not stating the year the experiment ran. It is a reason to test a bot against your current process, not a number to drop into a business case. 'Chatbots outsell humans' is not what that study says, and a comparison against 'some' employees is doing a lot of work in that sentence.

The mechanism it supports is still worth keeping. A bot answers instantly, at 2am, without getting bored on the fiftieth identical question. Where speed and consistency are what the buyer needs, that wins. Where the buyer needs somebody to take responsibility for a decision, it does not, and no amount of prompt work changes that.

What should you measure, and what should you ignore?

Measure outcomes you would have paid for anyway. Qualified handovers. Booked calls. Questions resolved without a human. Cost per resolved conversation. Then measure the thing most teams never look at, which is the rate of confidently wrong answers, found by reading transcripts rather than by trusting a dashboard.

Ignore conversations started, messages sent and engagement rate. Those numbers go up when the bot interrupts more, which is the opposite of the behaviour you want. A marketing chatbot that opens fewer conversations and books more calls is the better bot, and the headline metric will look worse. Most marketing chatbots are judged on the wrong axis for exactly this reason.

Cost per resolved conversation is also how you compare building to buying. Vendors publish list prices: Intercom publishes a price for its Fin AI agent on a public pricing page, with a free 14-day trial. So you can get a real number for the bought option before anyone writes code. If a build cannot beat a published price on cost or on fit, buy it.

Can you trust the chatbot adoption statistics in vendor decks?

Treat them as directional, and check what sits behind each one. A U.S. Chamber of Commerce page, with research by Teneo Research for its Technology Engagement Center, reports that a majority of small businesses embrace artificial intelligence, names the research partner, and publishes no sample size and no field dates. The Census Bureau's Business Trends and Outlook Survey asks 1.2 million businesses fortnightly and receives roughly 20,000 responses per wave, with a firm-weighted four-period moving average to December 2025 published by the Federal Reserve. Zendesk's CX Trends 2026 surveyed 6,182 consumers and 5,115 CX professionals across 22 countries and does not publish the fieldwork month.

None of that is dishonest. It is simply not comparable, and a deck that puts those numbers on one slide is building a trend out of different populations, weights and questions. If a figure is deciding your budget, open the source and find the sample. Often it is not published, which is itself the answer.

For a buying decision, mechanism beats a borrowed percentage. Ask whether your buyers ask questions before purchase, whether the answers exist in writing, and whether anybody owns keeping them current. Those three answers predict whether chatbots and marketing fit together at your company. No survey of other people's businesses can.

What does a marketing chatbot need before it goes live?

A written answer set, an escalation path, a transcript log somebody reads, and a clear statement that it is software. The first is the real work. A bot inherits the quality of the content behind it, so if your pricing is explained inconsistently across four pages, the bot will explain it inconsistently too, and faster.

Escalation has to be a route to a person with the conversation attached, not a form that starts the conversation again. Logging matters because the questions it failed are the cheapest market research you will see this quarter. Disclosure matters because a buyer who works out mid-conversation that they were talking to a bot discounts everything it told them.

If the bot touches accounts, orders or anything a customer had to log in to see, permissions must be enforced before retrieval rather than filtered out of the answer afterwards. That is a different build from a public marketing bot, and it should be scoped, tested and budgeted as one.