AI Features for Your Business Website: What's Worth Adding in 2026 (and What Isn't)

Most small-business websites in 2026 do not need a chatbot. They need a page that says what you do, a way to contact a person, and copy a customer can trust. An AI feature is worth adding when it answers a repeated question or finds a page faster — and when you can still take over when it is wrong. It is not worth adding when it invents prices, books the wrong job, or swallows personal details into a public tool.

This is about features on your own site, not about ranking in Google’s AI answers. Chatbots, on-site search, booking assistants, personalisation and accessibility widgets all look like progress. The ones that pay off have a narrow job and a human behind them. The ones that fail are sold as a receptionist you never have to check.

When is a chatbot worth putting on the website?

A chatbot is worth it when the same three or four questions arrive every week — hours, suburbs, how to book, what to bring — and a person still picks up when the answer is not in that list. Give it a script from your own pages. Put a phone number and an email next to it. Label it as a bot, not as “Sarah from reception”.

It is not worth it as the only way to reach you, or as a stand-in for a price list. If the bot can quote a fee, change a booking, or describe a legal right, you are letting software speak for the business. That is the point at which accuracy stops being a nice-to-have. A bot that says “I don’t know — here is the office number” is doing its job. A bot that fills the silence with a guess is not.

The OAIC’s guidance on commercially available AI products says public-facing AI tools such as chatbots should be clearly identified as such to customers. It also says organisations should not use an AI product simply because it is available, and should check whether it has been tested for the intended use, how human oversight will work, and who can see personal information put into it. A widget that scrapes your homepage and answers anything is not that test.

What about an AI receptionist or booking assistant?

An after-hours booking assistant can take a name, a preferred time and a suburb, then hand the slot to a person in the morning. That is useful. An assistant that confirms the job, takes a card, and tells the customer they are booked — without a person seeing the calendar — is how you get two jobs in the same hour.

Keep the assistant inside the practice software you already run, or make it write a request rather than a confirmed booking. Show the customer what it has taken down, and give them a way to reach a person if the time is wrong. If it cannot do that, a click-to-call button and a form that emails the office will lose fewer jobs.

Do not paste customer names, health notes or card details into a public chatbot to “try the feature”. The OAIC’s best-practice line is not to enter personal information, and particularly sensitive information, into publicly available generative AI tools. A receptionist product that sends those fields to a vendor you have not checked is the same risk with a friendlier name.

Is AI site search worth adding?

On a five-page brochure, no. A clear menu and a contact button beat a search box that has nothing to find. On a site with dozens of posts, product pages or course modules, on-site search — including an AI version that can rephrase the query — can stop people bouncing because they cannot guess your heading.

The risk is the same as the chatbot: if search “answers” the question in a generated paragraph, that paragraph is still the business speaking. If it only points at existing pages, the downside is a miss, not a made-up fee. Prefer the pointer over the invented summary until you have a person reviewing what it says. If the library of pages is thin, write the pages — that is what Content: $1,200/mo (4 blog posts a month, 3-month minimum) is for — rather than asking a model to hallucinate a library you never published.

Should the site personalise itself for each visitor?

Changing the homepage because someone came from a suburb or a previous visit sounds clever. It is also a collection and use of personal information. The OAIC treats information that an AI system generates or infers about an identifiable person as personal information, including when it is wrong. Feeding names, emails or browsing history into a personalisation engine has to fit the Australian Privacy Principles: it must be reasonably necessary, and a secondary use needs consent or a use the person would reasonably expect.

For most service businesses, suburb pages and a clear services list do the same job without a profile of the visitor. Personalisation that decides a price, a loan, or whether someone gets a slot is the high-risk end of the OAIC’s guidance — uses that can have a legal or similarly significant effect. Do not add that to a marketing site because a vendor demoed it.

Can an AI widget make the website accessible?

A tool that drafts alt text or a transcript can save time if a person checks the result. A widget that promises to make the whole site accessible, overnight, is a performance claim. If that claim is wrong, you have told customers and regulators something the page cannot do.

Treasury’s review of AI and the Australian Consumer Law is the official account of how those rules sit on AI products. The final report finds the ACL still applies: misleading or deceptive conduct in trade or commerce is prohibited, it can be contravened without fault, and the fact that a system can “hallucinate” does not carve AI out of that prohibition. The onus is on the business to use technology that is fit for purpose, and to make the limits of the technology clear to consumers. An overlay you have not tested is not a substitute for readable pages, real headings and a person who can use the site with a keyboard.

What do privacy rules require if you add AI?

If personal information goes into the tool, or comes out of it, the Privacy Act applies. The OAIC’s commercially available AI guidance is the document to brief against. Privacy obligations cover what you type in and what the system generates about a person. Inferred or incorrect output — a hallucinated address, a guessed medical detail — is still personal information if it is about someone who can be identified.

Update the privacy policy so it says you use AI, and how. Identify the chatbot as a bot. Do not assume a customer reasonably expected their enquiry to be sent to a model trainer overseas unless you told them at collection. The OAIC’s advice for secondary, AI-related uses is blunt: if you cannot show a reasonable expectation, get consent or a real opt-out. Due diligence is not a one-off purchase. Review the product, train staff, and keep a person in the loop.

Who is liable if the bot is wrong?

You are. Treasury’s review is explicit that a small business which buys an off-the-shelf chatbot to support customers remains liable if that chatbot misleads those customers. It may have a separate claim against the supplier if the product was not fit for purpose. It does not get to tell the customer “the AI said it”. Intention does not save you: the ACL prohibition on misleading or deceptive conduct does not require fault.

So keep the bot off topics where a wrong answer costs money or creates a legal right — prices, refunds, medical or financial advice, “you are booked”, “this is covered”. Log what it says. Have a person who can correct it. If you cannot do those three things, do not switch it on.

What should you actually commission?

Start with the site, not the widget. A website that names the service, the area, the price range you are willing to publish, and how to reach a person will beat a thin page with a bot on it. WebXpert’s published starts are on the plans page: from $999 for a typical build of about one to four pages, from $2,499 for five pages. Care after launch is included on sites WebXpert builds. Content: $1,200/mo (4 blog posts a month, 3-month minimum). None of those figures is a chatbot licence. If a vendor quotes an AI add-on, treat it as a separate product with a written scope, a human handoff, and a privacy note you can stand behind.

If you want that scope written for your site, book a free consultation.

Frequently asked questions

Does a small-business website need a chatbot in 2026?

Only if the same few questions arrive every week and a person can still take over. A chatbot is not a substitute for a clear page, a phone number, or a booking form. If it can quote prices or confirm jobs, you are responsible for what it says.

Do I have to tell customers they are talking to AI?

The OAIC says public-facing AI tools such as chatbots should be clearly identified as such to external users. Update the privacy policy to say you use AI, and do not send personal or sensitive information into a public generative tool.

Who is responsible if a website chatbot gives the wrong answer?

The business that put it on the site. Treasury’s review of AI and the Australian Consumer Law found the ACL still applies: misleading or deceptive conduct does not need intention, and a hallucination does not excuse it. Buying the bot from a vendor does not shift that duty to the customer.

Is on-site AI search worth it for a small site?

Not on a short brochure site. It can help on a site with many posts or product pages if it points people at existing URLs rather than inventing an answer. Generated summaries are still statements by the business.

Does WebXpert sell an AI chatbot?

No. WebXpert websites start from $999, five pages from $2,499, on the plans page. Content: $1,200/mo (4 blog posts a month, 3-month minimum). Those are website and publishing prices, not a chatbot licence.