Lead Qualification Questions an AI Should Ask (and When to Hand Off)

What does 'qualified' actually mean when an AI is asking the questions?

A qualified lead, in AI terms, is someone the system has confirmed on four things: identity (who they are and how to reach them), need (what they actually want), urgency (how soon they need it), and fit (whether your business can actually serve them). Anything beyond that is a nice-to-have, not a qualification requirement.

The mistake most businesses make is importing a 12-question sales script built for a human closer and handing it to a chatbot. A lead who just submitted a form or called at 9pm isn't looking for an interview, they're looking for confirmation that someone is paying attention. The shorter the ladder, the higher the completion rate.

We cover the broader system this sits inside, intake, scoring, and routing, in our guide to what an AI ops retainer actually includes. Qualification questions are just one layer of that system.

What should every AI ask, regardless of industry?

Before any industry-specific branching, every AI qualification flow should confirm the same four things, in roughly this order:

• Reason for contact: 'What brings you in today?' or 'What can I help you with?', open-ended, not multiple choice.
• Timeline: 'When are you hoping to get this handled?' This single question separates a browser from a buyer faster than anything else.
• Contact confirmation: 'What's the best number or email to reach you at?' Said out loud or restated in text, this catches typos before they cost you a lead.
• Fit signal: one question specific to your service that rules a lead in or out (location, service type, eligibility).

That's four questions. If your AI needs more than six before it can route the lead, the ladder is too long, and you're trading completion rate for information you probably don't need yet.

A simple way to think about it: score each answer 0-2 points based on how strong the signal is (vague timeline equals 0, 'this week' equals 2), sum the four, and use that number to decide whether the lead goes straight to a calendar booking or to a human for a closer look.

What does the question ladder look like for specific industries?

Once the universal four are answered, the next one or two questions should be industry-specific, because 'fit' means something different in every vertical.

Personal injury intake: after timeline, the next questions are about the incident itself (type of accident, whether they received medical treatment, whether they're currently represented by another attorney) and about the statute of limitations clock, often the single most time-sensitive piece of information in the whole conversation. We go deeper on this in our guide to speed to lead for personal injury firms, because in this vertical the qualification ladder and the response-time clock are the same problem.

Med spa and aesthetics: the branching question is usually treatment interest (which service, first-time or returning patient) followed by a light candidacy screen, since some treatments have contraindications a provider needs to review before booking. Our AI receptionist guide for med spas walks through how that screening question gets phrased so it doesn't sound like a medical interrogation.

Home services and contractors: the fit question is almost always property type and job scope ('is this a repair or a full replacement?'), since that determines which tech or estimator should be dispatched.

B2B services: the fit question is usually company size or current tooling, since that's what typically determines whether a lead is a good account fit at all.

When should the AI stop asking questions and hand off to a person?

Handoff should never be a judgment call the AI makes in the moment. It should be a fixed list of triggers you write down ahead of time, so the AI is checking against a rule, not guessing at tone. At minimum, build in these:

• The lead asks for a human directly. No further qualification, no 'just one more question.' Route immediately.
• The same question type gets missed twice. One miss, the AI can clarify or redirect. A second miss means the AI doesn't have the answer and shouldn't keep guessing.
• The topic requires licensed judgment. Medical advice, legal advice, specific pricing that depends on case detail, anything where a wrong answer creates liability.
• The lead signals urgency or distress. Words like 'emergency,' 'right now,' or visible frustration should skip the rest of the ladder entirely.
• The fit score clears your high-value threshold. If your scoring says this lead is worth a same-day callback, don't let the AI keep them in a chat queue finishing the script.

The point of a handoff trigger isn't to protect the AI from hard questions, it's to protect the lead from getting a worse answer than they'd get from a person.

How do you actually build this into a working system?

Here's the build order that holds up in practice:

• 1. Write the universal ladder first. Four questions, same for every lead, before any branching.
• 2. Add one branch per lead source or service line. Don't build ten branches on day one, add the highest-volume one and test it first.
• 3. Write the handoff triggers as explicit rules, not vibes, so they're auditable later.
• 4. Decide where a qualified lead actually goes, a calendar slot, a call queue, a CRM stage, before you launch, not after.
• 5. Run real past transcripts through it before trusting it with live leads, and read every one.
• 6. Review weekly for the first month. Look specifically for leads who answered fine but still got routed wrong, that's where the trigger logic needs tightening.

This is the same build order we follow when we stand up a live system for a client, since ClawOps only pitches what it can actually demo working, not a theoretical flow on a slide. For more breakdowns like this one, see our full guide library.

What mistakes make AI qualification feel robotic or lose the lead?

Asking closed questions when an open one would surface more. 'Do you need a consultation?' gets a yes or no. 'What's going on that brought you in today?' gets the actual story, which often answers two or three later questions for free.

No visible path to a human. If a lead can't tell how to escalate, they'll just leave. Every automated flow should make 'talk to a person' obvious and easy, not hidden behind more questions.

Treating a text lead like a phone lead. A phone call can absorb five questions because the back-and-forth feels like a conversation. A text or web chat loses patience after two or three. Match the ladder length to the channel.

Re-asking what the lead already told you. If the form said 'roof leak,' the AI shouldn't open with 'what can I help you with today?' That one repeated question is the fastest way to make a lead distrust the rest of the interaction.