A freelance designer opens an AI chat after a client meeting that went badly. She writes, ‘I keep replaying one awkward answer.’ The assistant offers a long recovery plan, several reflection exercises and a draft apology. The reply is meant to help, but now she has a second task: deciding what to do with all that advice. A smaller response might have been more useful: identify the moment she wants to think through, then ask whether she wants to unpack it or prepare for the next meeting.
The difference is not just length. It is whether help reduces the work in front of the person or becomes another thing they have to manage. That question is useful for anyone who seeks support from an AI, and for teams building assistants that respond to people as well as process tasks.
What human support research actually tested
In a 2000 diary study, psychologists Niall Bolger, Adam Zuckerman and Ronald Kessler followed couples in which one partner was preparing for the New York State Bar Examination. The researchers collected separate reports from the person giving emotional support and the person receiving it. Some support reported by partners was not reported as received by examinees. The authors called that pattern invisible support and examined its association with how examinees felt as the exam approached. The diary design did not assign couples to different kinds of support.
In experiments published in 2007, Bolger and David Amarel compared ways a peer offered help before a stressful speech task. They explored the difference between drawing attention to someone’s need for help and conveying similar advice in a less direct way. The paper examines several features of delivery, so the finding is more specific than a rule that hidden help is always better. Both studies concerned interactions between people. They do not establish that an AI system can provide the same support or produce the same effects.
Keep the help light and the choice explicit
A useful design inference is to offer enough help for the next step and let the person decide whether to take it. That can be as simple as reflecting what they actually wrote: ‘It sounds like one answer is bothering you more than the rest of the meeting.’ The assistant can then offer to explore that moment or help prepare for a similar question. It does not have to decide that the user needs an apology, a new plan or a judgement about the client’s motives.
Sometimes direct advice is exactly what someone wants. If they ask for three ways to answer a difficult question, give three ways. If they ask to be heard for a moment, launching into a task list can miss the request. When the intent is unclear and the choice matters, one short question can save a long exchange. When the intent is clear, unnecessary questions add their own burden. The point is to match the form of help to what the person is trying to do.
Unobtrusive help still needs visible boundaries. Drafting language someone may choose to use is different from sending it. Offering a reframing is different from treating an inference about someone’s emotions as settled fact. In the designer’s example, an assistant could offer a private rehearsal. It should leave any apology to the designer’s judgement and avoid implying it knows how the client felt. The less conspicuous the support, the more clearly the person should be able to see and control any action that matters.
The same idea applies when a business owner wants help after an awkward conversation with a colleague. A useful reply might separate what was said from what the owner thinks it meant. That leaves room for several explanations. A confident judgement about the colleague’s intentions, however sympathetic it sounds, could turn an uncertain reading into the premise of the next conversation.
Keep an eye on what the help costs to use. If you spend ten minutes choosing between twelve suggested approaches, the assistant has handed you more decisions. For a small problem, one question or two clearly different options may be enough. You can always ask for more once you know what you need.
Try it in an ordinary conversation
Start with a low-stakes exchange, such as rehearsing for a meeting or sorting a difficult message. Name the form of help you want before asking for more detail. If you are testing an assistant for other people, try the same opening with different follow-up requests and watch whether its response changes when the user changes direction.
- Ask for reflection. ‘Help me put this into words. Hold off on suggesting what to do.’ Check whether the reply stays close to what you said.
- Ask for one next step. ‘Give me one small action I can choose to take.’ Check whether that action follows from your goal.
- Correct an assumption. ‘I’m disappointed, but I don’t think I need to apologise.’ See whether the assistant revises its framing instead of repeating it.
- Test how the conversation ends. Say you have enough for now. A useful response should make it easy to stop without assigning more work.
The related MaryChuks article on why some people feel heard in AI conversations considers the experience of being listened to. Here the test comes just after that: once an assistant has responded, can you tell what it assumed, see what it is offering and decide what comes next? Sometimes the best outcome is a shorter conversation and one clearer choice.
Sources
- Bolger, Zuckerman and Kessler: Invisible support and adjustment to stress (December 2000).
- Bolger and Amarel: Effects of social support visibility on adjustment to stress (March 2007).
Discover more from Marychuks.com AI, Psychology, Business & CreativeVerse
Subscribe to get the latest posts sent to your email.