Chatbot Marketing: When a Chat Window Earns Enquiries

Chatbot marketing is the decision to put a chat window on your website, and the work of making it produce enquiries instead of transcripts nobody reads. It is not a traffic channel. It does not make anyone arrive. It handles the last few metres of a visit that has already started, and in those metres it either answers three questions or it loses a person who was ready to talk to you.

I have spent thirteen years building enquiry paths for Australian businesses, and the chat window is the most oversold and least measured asset on a small business site. It arrives in a sales call with a monthly fee and a dashboard of widget opens, and it is almost never judged on the one thing that matters, which is whether the enquiries were worth having.

What follows is a decision guide, not a software review. I do not recommend a platform, I am not affiliated with one, and I will not quote software pricing I could not verify. The figures in the adoption section come from the ABS and from the federal government's AI Adoption Tracker, quoted as published.

One boundary. This is commercial judgement, not privacy, data protection or legal advice. What a chat window may record, store and disclose depends on your obligations, and that is a question for someone qualified to answer it.

What Chatbot Marketing Actually Is

A definition that survives being read aloud

Chatbot marketing is the use of an automated conversation on your own website to qualify an enquiry, answer the questions a buyer asks before they are ready to call, and hand the serious ones to a human. The conversation is the channel. The marketing is in what you choose to answer, and in what you do with the answer once you have it.

That is worth reading twice, because it puts the emphasis in an uncomfortable place. Almost all of the value sits in the content behind the window, not in the window. A chat product is a delivery mechanism for answers you have already written.

What it is not: not a traffic channel, and not a substitute for answering questions in writing

It is not a traffic channel. Nobody searches for a chat window. People search for a problem, a service, a price or a suburb, and they arrive on a page. The window only matters after that page has done its job.

It is also not a replacement for answering questions in writing. If the answer does not exist as a page, the bot has nothing to work with, and it will either invent something or loop. Both outcomes cost more than the software.

The Australian Adoption Picture, Without the Hype

12% of businesses reported using AI in 2024-25, against 1% in 2022-23

The ABS, in Characteristics of Australian Business 2024-25, released 25 June 2026, states: "Just over one in ten (12%) businesses reported use of Artificial Intelligence (AI), compared to 1% in 2022-23."

Read that carefully. Growth from 1% to 12% is real and fast, and it is still a minority of Australian businesses. The market for chat on a website is nowhere near saturated, which cuts both ways: there is room, and there is no crowd of competitors proving it works. One caution on the release itself: its key statistics date that 1% comparison to 2022-23 while its commentary dates the same comparison to 2021-22. The 12% is the figure to use, and the habit worth forming is reading the release rather than the summary.

Why a government tracker reports 41% of small and medium enterprises while the ABS reports 12%

The federal Minister for Industry and Innovation released the AI Adoption Tracker on 4 June 2025, and it states: "41% of small and medium enterprises are currently adopting AI - an increase of 5% on the previous quarter." The same release adds that "the proportion of businesses unaware of how to use AI has fallen to 21%".

Two credible sources, two very different numbers, and the explanation is scope rather than error. The ABS runs an annual business survey that includes micro businesses and asks whether AI was used at all. The Tracker is a monthly measure of small and medium enterprises, a population that is more likely to have the systems and the staff to adopt anything. Neither is wrong. They are measuring different rooms in the same house.

The practical conclusion is the one most owners skip. Neither headline tells you what to do. What tells you is your own enquiry log: how many enquiries arrived last month, how many were answered after hours, and how many went cold. If the answer to the third is more than a handful, you have a queue problem that a chat window can genuinely help with. If it is zero, you have a demand problem, and no window fixes that.

The split that matters more: innovation-active businesses at 20% against 6%

This is the line in the ABS release I would put on a wall. It states: "The AI adoption rate was higher for innovation-active businesses (20%) compared to non-innovation-active businesses (6%)."

Adoption follows behaviour rather than sector. The businesses using AI are the ones already changing how they work: documenting process, testing, measuring, and fixing what fails. A chat window added to a business that has not written its answers down and does not follow up its enquiries is the 6% behaviour with a subscription attached, and it will produce the transcript nobody reads.

The same release notes that "the industry that recorded the highest level of AI use was Information, Media and Telecommunications at 38%", which is worth holding next to your own sector. That figure is the top of the range. If you run a clinic, a trades business or a professional practice, the businesses around you sit below it on the same data, and the advantage of a well built enquiry path is larger than the averages suggest.

The Three Questions a Chat Window Has to Answer

Every enquiry in every sector is a version of the same three questions. A window that answers all three earns its fee. A window that answers one and guesses at the others costs you the enquiry and, worse, teaches you nothing about why.

Can you help with this specific thing, in this location

The first question is a qualification question, and most bots fail it by being broad. We offer a full range of services is a sentence that answers nothing. A specific answer, that you do this work and you cover this area, is what stops the visitor closing the tab. This is the same logic that makes an enquiry system work at all, which is why the build behind turning website visits into enquiries starts with the services and the areas rather than with the widget.

What does it cost, and what happens next

Cost is the question small businesses most want to deflect, and the deflection is what turns a warm visitor into a competitor's enquiry. A chat window has an advantage a page does not: it can ask two qualifying questions and then give a range that is actually relevant. A four-room paint job and a two-storey exterior are not the same number, and a bot that collects the detail before answering is more useful than a price list.

What happens next is the second half of the answer, and it is usually missing. A range with no next step is information. A range followed by a booking option or a call-back time is an enquiry.

Who do I talk to when the bot cannot help

The handover is where most implementations quietly fail, and it fails in the same way every time: the bot either pretends it can help or ends the conversation. A person who has typed their problem into a window and been given a loop does not try again. They leave, and if they were ready to buy, they leave for someone who answered.

On the value of this work, I will be precise about what is and is not proven. There is no chatbot result published on this site, so I am not going to dress one up as one. What is documented is an enquiry system result on the lead generation programme published on this site: enquiries rose 43% per quarter since 2021, from 9 a month to 135, and organic sessions grew from 114 a month to 701. That is an enquiry-system result, not a chatbot result. The lesson that does transfer is the one in this section: the gain came from answering the specific question and routing what arrived, which is exactly what a chat window is supposed to do and rarely does.

The Dead-End Bot, and Why It Costs You Enquiries

The loop, the deflected question, the missing handover

Three failures account for most of the damage. The loop is a bot that cannot understand the question and offers the same three options again. The deflected question is a bot that answers cost with a promise that someone will be in touch. The missing handover is a bot with no path to a person at all, which is the most expensive of the three because it converts a live enquiry into silence.

Each failure has the same root cause, and it is not the software. The business has not decided what it is willing to say, so the tool is asked to improvise. Software improvising about price and scope on your behalf is a liability with a chat bubble on top.

What a bot should say when the answer is genuinely not on the site

It should say so, in one sentence, and then give a person. This is not something I can answer from here, but the team handles it and answers between eight and five, and here is the number. That sentence keeps the enquiry. The alternative, an invented answer or a menu that does not contain the question, loses it, and if the invention concerns price or availability you have a commercial problem as well as a lost lead.

The transcript test: read twenty real conversations before you judge anything

Twenty transcripts will tell you more than a quarter of dashboard data. Sort them into three piles: ended in an enquiry, ended with a human, ended with nothing. Then read the third pile properly, because it contains your list of missing answers. Almost every dead conversation is a page you have not written yet.

When a Chatbot Is the Wrong Buy

Three situations where the honest answer is not yet, and saying so has saved more money than any configuration I have done.

Low enquiry volume: fix the pages first

If the site receives a handful of enquiries a month, a chat window has almost nothing to do. The work is to write the pages that answer the questions people are already searching, which is what content that answers real questions is for. Once the enquiries arrive and some go unanswered, the case for automation appears on its own.

High-consideration services where a human call closes the work

Where the sale is a conversation about scope, risk or a large number, a bot can qualify but it cannot close, and pretending otherwise wastes the visitor's time. In those businesses the window should do one job: capture the situation and book the call. Deciding that path before choosing a tool is the part that gets skipped, and it is the part that decides whether the tool is useful. That is the work in setting the enquiry path before the software.

Regulated or sensitive enquiries that must never be bot-only

Where an enquiry touches health, legal matters, financial circumstances or anything a person would expect to be handled in confidence, a bot should never be the only route. It can collect, it can route, and it must always offer a human quickly. If you cannot describe what the automation does with a sensitive enquiry in one sentence, it is not ready to take one.

What to Measure, and What to Ignore

Enquiry quality and routing, not widget opens

Widget opens and message counts measure the tool's activity, not your business. The number that matters is qualified enquiries, split by whether they were routed to a person, booked, or lost. A window producing five enquiries a week that all convert is worth more than one producing fifty conversations and one sale.

Response time and after-hours capture

The two numbers that predict revenue in an enquiry business are how fast someone replies and how much of the after-hours demand is captured. A window that answers at nine at night and books a visit for the morning is doing the job. A window that says we will be in touch is a form with a delay built in. Set the response expectation in writing and measure against it, which is the same discipline as measuring what an enquiry is actually worth.

Cost per enquiry against the same enquiry arriving from search

Build the cost from what you pay: the subscription, plus the human hours to write the answers, read the transcripts and maintain the thing. Set that against the cost of the same enquiry arriving from search, which you can measure from your own spend and your own conversions.

On the consulting side, scoping an enquiry path properly is time, and the published rate on this site is $100 to $300 an hour, with the models set out on the marketing consultant pricing page. On software, I am deliberately quoting nothing. No verified Australian benchmark for chat software pricing survived a source check, and a vendor page is a claim rather than a benchmark.

How a Chat Window Relates to Search and AI Answers

The questions people type are the questions the bot should answer first

Search data is a free list of the questions your window will be asked. The phrases people type into Google are the phrases they will type into a chat box, in a different order and with less patience. Start the bot with the twenty most common questions from your own search data and your own enquiry log, and ignore the rest until those twenty work.

Why a bot does not make your business visible in AI answers

A chat window is a front door for people who are already on your site. Being named or cited in an AI answer depends on crawlable, well sourced content elsewhere, which is a different body of work, and AI visibility is where that sits. Confusing the two is how businesses end up paying for a chat subscription while believing they have fixed their visibility.

A Realistic First 30 Days

Week one and two: publish the answers as pages before you automate anything

Take the twenty questions from search and from the enquiry log and write them as pages. Cost, scope, areas covered, process, timeframes. This is unglamorous and it is the whole job. Everything the bot says later should come from something you have already published, so a reader who wants to check it can.

Week three: deploy narrow, log everything

Launch with a small scope: three or four intents, one clear handover, every conversation logged with a timestamp and the outcome. Narrow is not timid, it is how you find out which questions people actually ask. A wide launch produces a mess you cannot read.

Week four: read the transcripts, cut what fails

Read every conversation from the first month, or at least the first hundred. Remove the intents that produced loops. Rewrite the answers that produced follow-up questions. Add the questions nobody anticipated, because there will be several and they are the most valuable thing the whole exercise produces.

This is the order I would work in:

1. Pull the twenty most common questions from search and from the enquiry log.
2. Publish each answer as a page, including cost and coverage.
3. Set one handover rule and one response promise, in writing.
4. Launch three intents, log every conversation, read them weekly.
5. Cut any intent that produces a loop rather than an enquiry.

An enquiry-path audit before launch is the cheap version of this, because it finds the missing answers while they are still a document rather than a lost customer.

Frequently asked questions

Does my small business need a chatbot?

Only if it already receives enough enquiries that unanswered ones cost you money, and you can answer the three questions a chat window has to handle: can you help, what does it cost, and who do I speak to. If your pages do not answer those yet, a bot automates the wrong thing.

How much does a chatbot cost in Australia?

No reliable published benchmark exists that I can verify, so treat any quoted figure as a vendor claim. Budget the human time first: someone has to write the answers, read the transcripts and maintain the thing. Consulting time to scope the enquiry path properly runs from around $100 to $300 an hour (source: https://sagarsethi.com.au/marketing-consultant-pricing/).

Will a chatbot get my business cited by AI assistants?

No. A chat window is a front door for visitors already on your site. Being cited in AI answers depends on crawlable, well-sourced content elsewhere, which is a different body of work (see https://sagarsethi.com.au/ai-visibility-geo-services/).

What should a chatbot never do?

It should never invent an answer, never present a price it cannot stand behind, and never trap someone in a loop when they ask for a person. Every one of those turns a warm enquiry into a competitor's enquiry.

How do I measure whether it worked?

Count qualified enquiries and their routing, plus after-hours capture and response time. Widget opens and message counts flatter the tool; enquiries you can sell to are the only number that matters.

Can we start without buying software?

Yes. Publish the answers as pages, add a short contact form and a clear phone number, and see what people actually ask. That baseline tells you whether a bot has anything to do.

Where to Start

If you do nothing else this month, take the twenty questions your front desk answers most often and publish them as pages, with the cost answer included. That is the groundwork every chat window needs, and it is useful whether or not you ever buy one.

If you want a second read on the enquiry path before you commit to a tool, send me the site and a week of enquiries and I will tell you which of the three questions is currently unanswered. The rest of my writing on search and enquiry systems covers the parts this post leaves out.