Careers Website Chatbot: What It Does, What to Measure, How to Choose

What a careers website chatbot does

A careers website chatbot answers questions and points candidates to the right page or job, at any hour, in their own words.

Done well, a chatbot removes friction. A visitor asks "do you hire in Leeds?" and gets a clear answer in seconds. Done badly, it hides the menu, gives wrong answers and sends candidates to a dead end. The difference is purpose, content quality and hand-off.

Four types: FAQ, search assistant, apply assistant and conversational navigation

Conversational navigation is a Happydance feature. Read more on the conversational navigation page.

What to measure

Count outcomes, not conversations. A busy chatbot that sends nobody anywhere is not working.

Combine with web analytics. Our employer branding metrics guide covers measuring the wider candidate journey.

Hand-off to the application

A chatbot is a doorway, not a destination. Every conversation should end somewhere useful: a job page, an application, a talent community or a person. Test the hand-off on a phone, and make sure candidate data is passed only with consent.

• A clear next step after every answer.
• Easy route to a human where possible.
• Application opens on the right job, with context.
• No candidate data passed without consent.
• A graceful message when it cannot help.

Privacy, consent and bias

Conversations can contain personal data, even when you do not ask for it. Decide what you collect, say so clearly, store it safely and delete it on a schedule. Ask how bias is tested, particularly in anything that ranks or recommends candidates. Take advice from legal and data protection teams.

A conversation flow that works

Whatever the type, a good conversation follows the same shape. It greets, understands, answers, offers a next step and learns.

1. WelcomeSay what it can help with, in one line, and offer three starting questions.
2. UnderstandLet the visitor type or tap. Confirm what you understood.
3. AnswerGive a short answer from content you control, with a link to read more.
4. Next stepOffer a job, an application, a story or a person.
5. LearnLog what could not be answered and review it every week.

When a chatbot is the wrong answer

Not every careers website needs one. If your visitors rarely ask questions, your content is thin, or your jobs data is messy, fix those first. A chatbot placed over a confusing website makes the confusion faster. Start with a clear structure, good search and honest content, then add conversation where it removes real friction.

• Visitors repeat the same questions by email or phone.
• Your jobs data is clean and up to date.
• Someone owns reviewing conversations every week.
• You can explain what data is collected and why.

A 30-day test plan

Treat a chatbot as a test, not a launch. Run a small, measured pilot and decide with evidence.

Our personalisation guide shows how to tailor the experience once the basics work.

What real conversation data shows

12 questions to ask before you buy

Use these questions with any supplier. Tick each one when you have a clear, written, demonstrated answer.

Compare how this fits your wider plans in the Happydance platform overview, and read about tailoring content in personalisation.

Chatbot requirements: a one-page brief

Before you speak to any supplier, write what the chatbot must do. Vendors sell features, and a brief keeps you buying outcomes.

A scoring grid for shortlisted tools

Score each shortlisted tool from 1 to 5 on the criteria below, multiply by the weight and total. The weights are a starting point, so change them to match your priorities.

Twenty test questions to run in every demo

Prepare your own list, drawn from real candidate emails and phone calls. Include easy questions, questions with two possible answers, questions the bot should refuse, and a few that are badly worded. Run the same list in every demo and score each answer for accuracy, tone and whether the source is shown. A tool that handles refusals well is usually safer than one that answers everything.

Conversation design rules that earn trust

• Say it is automated, in the first message.
• Offer a way to reach a person at every step.
• Keep answers short and link to the full page.
• Say when you do not know instead of guessing.
• Never ask for more personal data than the task needs.
• Write in the same tone as the rest of the careers website.

A risk register for chatbot projects

Launch checklist

• Requirements, limits and hand-off are written and signed off.
• Test questions have been run and scored.
• Privacy notice and consent have been reviewed by legal.
• A named owner reviews conversations every week for the first month.
• Success measures and a stop rule are agreed.

What a pilot should look like

A small, well-measured pilot is cheaper than a large launch you cannot evaluate, and it gives your legal, security and accessibility reviewers something concrete to inspect.

Maintaining answers over time

A chatbot is only as good as the content behind it. Assign an owner to the source pages, review the most common questions each month, add new answers when the same question appears three times, and retire answers that point to old content. Keep a change log so that you can explain why an answer changed.