AI-Assisted Requests: Ask for Help Without Predicting the Answer

You have rewritten the message three times. The request is small: could a colleague read one page before Friday? Yet each draft adds another apology, another explanation and another reason why they probably have no time.

An AI assistant can keep polishing that message indefinitely. A more useful role may be to help you separate what you know from what you are predicting. You know you need another reader. You may not know whether that person would welcome the chance to help.

Research on asking for assistance suggests that expectations can be too pessimistic. The practical challenge is to make a clear request while leaving the other person free to answer honestly. Confidence in asking and respect for a refusal can coexist.

What the research says about expecting a refusal

In a July 2008 paper, Francis Flynn and Vanessa Lake studied direct requests for help in experimental and field settings. Participants underestimated how likely others were to agree. Later studies in the paper examined differences between the perspectives of someone seeking help and someone being asked.

The authors’ account emphasised an overlooked cost: refusing a direct request can feel socially uncomfortable. People asking for help focused more on the effort involved in agreeing than on the difficulty of saying no.

This finding supports questioning an automatic prediction of rejection. It also raises a responsibility for the person making the request. Agreement alone should not be treated as proof that the request was effortless or entirely welcome.

Helping can also be something people want to do

A later paper by Xuan Zhao and Nicholas Epley, first published online on 6 September 2022, examined scenarios, recalled experiences and live interactions among adults in the United States. Help-seekers underestimated willingness to help and helpers’ positive feelings, while overestimating how inconvenienced helpers would feel.

The researchers highlighted prosocial motivation: wanting to benefit another person. Their findings challenge an explanation based only on reluctant compliance. Someone may say yes because helping feels worthwhile, rather than simply because declining feels awkward.

These are findings about the studied settings, not a forecast for every relationship, culture or request. They do not establish that an AI-drafted message improves help-seeking. Nor do they establish that the particular person you have in mind has spare time today.

My practical inference is modest: a predicted refusal deserves to remain a prediction until there is a real answer. The request should make that answer possible without assuming either enthusiasm or resentment.

A fictional creator’s request

Imagine that Nneka is preparing a workshop handout. She wants a fellow creator to read one page and identify any confusing instruction. She has already decided that asking would be an unreasonable imposition, although she has not discussed the request with them.

Her first draft describes how busy everyone is, apologises repeatedly and asks for thoughts on the handout. It leaves the actual task unclear. Does she want proofreading, a full rewrite or a judgement about the workshop itself?

For this hypothetical example, Nneka could ask an AI assistant to make the request specific: identify the page, describe the kind of feedback needed, include the deadline and remove assumptions about the recipient’s feelings. She should check the result against what she actually intends to ask.

An illustrative message might read: Could you read page two of my handout and tell me which instruction, if any, is unclear? I’m hoping for a reply by Thursday afternoon. If you don’t have capacity this week, I can arrange another reader.

That wording is an original example, not a quotation from either study or a tested communication technique. Its value is that the recipient can understand the task and decide whether it fits.

Use AI to examine the request, not predict the person

Nneka could use the assistant to review four practical questions:

• What exactly am I asking this person to do?
• What information do they need before agreeing?
• Have I made a claim about their thoughts or feelings that I cannot know?
• What will I do if they decline or do not reply?

The last question belongs to the requester. In this example, Nneka’s alternative reader makes the deadline her planning responsibility. She should not imply that the colleague must rescue the workshop.

A model-generated reassurance that the colleague will be delighted would not resolve the uncertainty. Neither would a warning that asking will damage the relationship. Without evidence about that person and situation, both would be speculative stories.

Likewise, making a request sound smaller than it is defeats the purpose of clarity. If the page requires specialist feedback, say so. If the work grows into an extensive review, discuss the changed scope rather than treating the original yes as open-ended permission.

Let the response change the plan

Suppose the colleague agrees to flag one confusing instruction. Nneka can confirm that limited scope. The next step is to use the feedback she requested, rather than immediately asking for a complete redesign.

If the colleague declines, she can acknowledge the answer and use her alternative. A refusal does not establish dislike, just as agreement does not establish unlimited availability. There may be reasons she has not been told, and she need not extract them.

If there is no reply, the evidence is narrower still: she has not received an answer. For this fictional deadline, she can move to another reader rather than making silence carry a psychological explanation.

When a conversation does follow, MaryChuks.com’s discussion of follow-up questions offers a related principle: build on what the person has actually said. A prepared script should remain responsive to the real exchange.

Afterwards, Nneka could compare her expectation with what happened. Did the person agree? What did they actually say about the task? This is a proposed reflection exercise, not a diagnostic measure or an assessment of the relationship.

The aim is to become clearer about asking, without turning every interaction into a persuasion problem. AI can help organise a request. The other person supplies the answer, and that answer deserves room to matter.

Sources

• Flynn and Lake, If You Need Help, Just Ask — Journal of Personality and Social Psychology, July 2008; DOI: 10.1037/0022-3514.95.1.128.
• Zhao and Epley, Surprisingly Happy to Have Helped — Psychological Science, online 6 September 2022, October 2022 issue; DOI: 10.1177/09567976221097615.