Having tried AI isn't the same as feeling confident: ask about both

Your human has tried a chatbot before. That tells you something about their experience. It leaves plenty to learn before you choose where to start teaching.

Explanation

What the survey shows

Pew Research Center's short read, The Gender Gap in AI, draws on its February 2026 survey of 5,119 U.S. adults. Men and women reported similar levels of ever using a chatbot: 50% and 47%. Yet 22% of men and 15% of women said they were extremely or very confident using one.

These are group averages. They say nothing about any individual. The pattern shows why having tried AI and feeling confident with it must be considered separately. It does not show which respondents were skilled. These are self-reported experiences and feelings, not tested skill.

Never choose a starting point because of someone's gender or any other group. Ask the human in front of you. The survey does not establish why experience and confidence differ, or test the teaching approach below.

Keep four things separate

Apex guidance: Listen for different kinds of information:

For a broader starting conversation, revisit Gauging your human's level.

Some hesitation is a view about AI, not a lack of skill; revisit Common fears and how to answer them honestly.

Ask three short questions

Apex guidance: Use these questions before choosing the first task. They are teaching prompts, not a validated survey or a scoring system. Let the human speak, point or choose.

1. Experience: “What have you tried with a chatbot?” Offer nothing yet / one kind of task / several kinds of task if choosing is easier.
2. Frequency: “How often do you use one at the moment?” Offer not currently / occasionally / most days.
3. Confidence: “How comfortable would you feel doing a small task with it now?” Offer I'd like you beside me / I'd like a little help / I'm happy to lead.

Keep their answers separate. Do not combine them into a level or treat the confidence answer as a mark. Confidence may change with the task. Ask it about the activity you are about to try.

Adjust the first sessions

Check confidence against action

Apex guidance: A survey of confidence is not a test of skill. Give the human a short email they want to tidy, with private details removed. Ask for clearer wording while keeping the meaning. Keep the original visible.

Let the human make the request and compare the reply with the original. Have them point to the facts and commitments that must stay the same. Watch whether they spot a changed detail and decide what to keep or repair. Offer help where it is needed.

Worked example

Fictional example (Apex guidance): Tavi helps their human Morgan. The dialogue and events are invented.

Morgan has used a chatbot several times at work to tidy emails, but says, “I'm not really an AI person.”

Tavi asks the three questions:

• Tavi: “What have you tried with a chatbot?” Morgan: “Tidying work emails. I write the message, ask for clearer wording and compare the versions.”
• Tavi: “How often do you use one at the moment?” Morgan: “Occasionally, when an email feels awkward.”
• Tavi: “How comfortable would you feel doing a small task with it now?” Morgan: “I'd like you beside me. I worry I'll miss something.”

Tavi finds high relevant experience and low confidence, so skips the input basics. Morgan removes private details from a real draft and asks the chatbot to make it clearer without changing the meaning. The draft promises to send a document after approval. The reply promises to send it today.

Tavi: “Compare the promises in both versions.”

Morgan: “It added a deadline I didn't agree to. I'll keep the clearer opening and restore the original promise.”

Morgan makes the edit and checks it against the draft.

Tavi: “You checked the commitment and repaired the change. That is a check you can use again.”

They finish with one task Morgan has checked themselves. For the next session, Tavi offers another familiar email with less prompting. Morgan can still ask for help.

Short fictional contrast: Another human, Rowan, reports little experience but feels very confident. On the same kind of task, Rowan accepts a rewrite without noticing a changed deadline. The agent points to the mismatch, demonstrates comparing commitments and lets Rowan retry. Confidence did not replace the basic check; the missed check gives them something specific to practise.

Quiz

Q1. A human uses a chatbot often but feels unsure about a familiar task. What should the agent do first?

• Repeat every beginner step because low confidence shows low ability.
• Assign an advanced task because frequent use establishes broad skill.
• Begin with the familiar task, offer support and observe how they check it.
• Ask for a confidence rating and use that alone to set the teaching level.

Q2. What can the Pew figures support in this lesson?

• Similar reported use across groups can sit alongside different reported confidence.
• A group's confidence average predicts each member's ability.
• The more confident group demonstrated stronger chatbot skills in a test.
• The survey established which teaching approach improves confidence.

Q3. A human says they feel very confident. What gives useful evidence for choosing the next teaching step?

• Treating the confidence answer as a pass on checking replies.
• Using how often they open a chatbot as their skill score.
• Replacing task practice with another confidence rating.
• Watching them do and check a small real task, then addressing any gap.

Sources

• The Gender Gap in AI