Research With Adults Who May Lack Capacity: Proxy Consent, Assent, and Safeguards

Answer first: capacity is not a property of a person, a diagnosis, or an age — it is a judgement about this decision, at this moment. An adult lacks capacity to consent to your study only if an impairment or disturbance in the functioning of the mind or brain means they cannot understand the relevant information, retain it, use or weigh it, or communicate a decision. If they can do all four with support, they consent themselves and no proxy is involved. If they cannot, a relative signature does not automatically make the research lawful — in commercial user and market research the correct answer is usually to adapt the study so the person can consent, or to exclude them and record why.

The corpus already covers two neighbouring cases: research with children and teens, where age triggers parental consent and child assent, and vulnerable customer research, where FCA rules require firms to evidence good outcomes. This article covers the third case, which is different from both: an adult, presumed capable in law, whose ability to consent to this decision is in question.

This is general guidance for research teams, not legal advice. Capacity law is jurisdictional, and clinical or academic research carries approval requirements that commercial research does not.

Start from the presumption

The Mental Capacity Act 2005 (England and Wales) states five principles that travel well as a research standard regardless of where you operate:

1. A person must be assumed to have capacity unless it is established that they lack capacity.
2. A person is not to be treated as unable to make a decision unless all practicable steps to help them do so have been taken without success.
3. A person is not to be treated as unable to make a decision merely because they make an unwise decision.
4. Anything done for a person who lacks capacity must be done in their best interests.
5. Before acting, regard must be had to whether the purpose can be achieved in a way less restrictive of the person rights and freedom of action.

Principles 1 to 3 do most of the work in research. Teams get this backwards remarkably often: they treat a diagnosis, an age, or a support need as evidence of incapacity, and they treat a decision they find surprising — agreeing to a two-hour session for a small incentive — as proof of it. Neither inference is sound.

The two-stage test

Under the Act, a person lacks capacity in relation to a matter if, at the material time, they are unable to make the decision for themselves because of an impairment of, or a disturbance in the functioning of, the mind or brain. The impairment may be permanent or temporary. And critically, a lack of capacity cannot be established merely by reference to a person age or appearance, or a condition or aspect of behaviour that might lead others to make unjustified assumptions.

That is the diagnostic limb. The functional limb is the one you actually apply. A person is unable to make a decision if they are unable:

• to understand the information relevant to the decision,
• to retain that information,
• to use or weigh that information as part of the process of making the decision, or
• to communicate their decision — by talking, sign language, or any other means.

Three qualifications in the statute matter enormously in practice:

• A person is not to be regarded as unable to understand if they can understand an explanation given in a way appropriate to their circumstances, using simple language, visual aids, or any other means.
• Being able to retain the information for a short period only does not prevent someone from being able to make the decision.
• Relevant information includes the reasonably foreseeable consequences of deciding either way, or of not deciding.

So "they would not remember this tomorrow" is not a capacity failure. "They could not follow the explanation as written" is a failure of your materials, not of the person, until you have tried a better explanation.

Make consent achievable before you decide it is impossible

Principle 2 is an obligation to try. In research terms, "all practicable steps" is a concrete list:

• Rewrite the information sheet at a plain-language reading level. Short sentences, one idea per sentence, no defined terms, no legal boilerplate in the body. Put the boilerplate in an annex the person does not have to parse to decide.
• Chunk the decision. Explain, ask, confirm, then move on. Do not read four paragraphs and then ask "any questions?".
• Use teach-back. Ask the person to tell you in their own words what the study involves and what happens to their recording. Teach-back is the single most useful capacity check available, and it doubles as evidence you took the step.
• Offer voice instead of reading. For someone who struggles with written text, hearing a question and answering aloud is a substantially lower barrier.
• Remove time pressure. Let people take breaks, stop, and resume. Scheduled live sessions create pressure that suppresses questions.
• Allow a supporter to be present for explanation — while keeping the decision the participant own. A supporter explaining is help; a supporter answering is substitution.
• Choose the right moment. Capacity fluctuates with fatigue, medication, pain, and time of day. Ask when the person is at their best, not when your calendar is free.

Asynchronous AI-moderated interviews are genuinely useful here, and it is worth being precise about why. A Koji study can be taken by voice or by text, at whatever hour suits the participant, at their own pace, with no interviewer waiting for an answer and no scheduling. The AI asks one question at a time and follows up conversationally rather than presenting a wall of form fields. For people who find live sessions overwhelming or written surveys inaccessible, that combination lowers the barrier substantially — and lowering the barrier is exactly what principle 2 asks of you.

It also removes a safeguard, and the honest treatment is to say so: nobody is in the room to notice distress. That trade-off dictates where async AI research is appropriate — low-risk, non-sensitive topics, with a visible exit, a check-in question, and a route to a human. It is not a way to enrol someone who could not have consented in person. See Trauma-Informed User Research for the sensitive-topic case.

When capacity is genuinely absent

Here is the part most commercial research teams get wrong. The instinct is to ask a family member to sign. In most commercial user and market research, that instinct is unsound in both major frameworks.

United States. Under the Common Rule, consent may be sought from a legally authorized representative — an individual or judicial or other body authorized under applicable law to consent on behalf of a prospective subject to their participation in the procedures involved in the research. Where no applicable law addresses it, an LAR is an individual recognised by institutional policy as acceptable for providing consent in the nonresearch context on behalf of that person. Two things follow: who qualifies is set by state law and institutional policy, not by family relationship or convenience, and the framework assumes an institution with an IRB — see IRB Approval for User Research. The Common Rule also requires, where subjects are likely to be vulnerable to coercion or undue influence — explicitly including individuals with impaired decision-making capacity — that additional safeguards be included to protect their rights and welfare.

England and Wales. The MCA route is narrower still. Intrusive research on a person who lacks capacity to consent is unlawful unless the project is approved by an appropriate body and conducted in accordance with the Act. Research is intrusive if it would be unlawful to carry out on a person with capacity without their consent. Approval requires, among other things, that the research is connected with an impairing condition affecting the person, or its treatment, and that there are reasonable grounds for believing research of comparable effectiveness cannot be done with people who can consent.

Read that requirement against a typical commercial study — testing a checkout flow, exploring pricing perceptions, evaluating an onboarding experience. Almost none is connected with an impairing condition. The framework does not offer you a proxy route, because it was never designed to. The absence of a lawful proxy mechanism is the answer, not an obstacle to route around.

Where the Act does apply, it uses a consultee, not a proxy consenter — a distinction worth internalising. The researcher must take reasonable steps to identify someone engaged in caring for the person or interested in their welfare, otherwise than in a professional capacity or for remuneration, who is prepared to be consulted; if none can be found, a person with no connection to the project is nominated. The consultee is asked for advice on whether the person should take part and what the person own wishes and feelings would likely be. If the consultee advises that the person would decline or would want to withdraw, the researcher must ensure they do not take part, or are withdrawn.

That is not consent by substitute judgement in the everyday sense. It is a structured attempt to reconstruct what the person themselves would want.

The safeguards worth adopting everywhere

Sections 32 and 33 of the MCA state four safeguards that are good practice in any jurisdiction, including where nothing legally compels them:

1. Apparent objection stops the research. Nothing may be done to which the person appears to object, whether by showing signs of resistance or otherwise. Objection does not have to be verbal, and it does not have to be reasoned.
2. Their interests outweigh those of science and society. Stated in the Act as an assumption, not a balancing exercise.
3. Any indication of wanting to withdraw means withdrawal without delay. "In any way" — a look at the door counts.
4. A prior statement of wishes is binding, even if made informally, if you are aware of it.

Adopt these as house rules for every study involving anyone whose capacity might fluctuate. They cost nothing and they are the difference between a defensible programme and one that relies on nobody complaining.

A practical protocol

You are not conducting capacity assessments — you are running a process that avoids enrolling someone who cannot consent, without excluding people simply because supporting them takes effort.

One point on GDPR: consent is only a valid lawful basis if it is freely given, specific, informed, and unambiguous. If the person could not understand the study, the consent is not informed, and it is not rescued by a signature. Nor should you quietly switch to legitimate interests to record a session with someone who could not consent to it — that inverts the protection.

Where Koji fits

The accessibility features that make Koji a better research platform are the same ones that make consent achievable for more people:

• Voice or text, participant choice. One study, two modalities. Reading difficulty stops being an exclusion criterion.
• Asynchronous by default. No scheduling, no live pressure, breaks whenever needed — directly serving the "all practicable steps" obligation.
• AI follow-up probing that asks one thing at a time and rephrases when an answer is unclear, instead of presenting a dense form.
• Structured questions — open_ended, scale, single_choice, multiple_choice, ranking, yes_no — let you build the consent and check-in steps into the interview itself: a yes_no teach-back confirmation, a single_choice support-needs question, a mid-session check-in. Because these are typed fields, you can filter out any session where the consent confirmation was not clean, before the data reaches analysis.
• Multi-language support for participants who understand best in their first language — an accessibility question as much as a translation one.

Traditional survey tools push in the opposite direction: a wall of fields, a fixed reading level, an all-or-nothing submit button. That design excludes people quietly and calls the result a representative sample.

Start with 10 free credits and build the accessible route into the study rather than bolting it on.

Frequently asked questions

Can a family member consent on behalf of an adult participant?

Usually not, for commercial research. In the US, consent must come from a legally authorized representative as defined by applicable law or institutional policy — not by family relationship. In England and Wales the MCA uses a consultee who advises on what the person would want, and the research provisions only apply to approved projects connected with an impairing condition. For ordinary product research, no proxy route exists, and that is the answer.

Does a diagnosis of dementia mean someone cannot take part?

No. Capacity is decision-specific and time-specific, and lack of capacity cannot be established by reference to a condition or to age. Many people with a dementia diagnosis can consent to a low-risk interview, particularly with plain-language materials and a well-chosen time of day. Assess the decision, not the diagnosis.

What if capacity fluctuates during a session?

Treat consent as ongoing rather than a one-time signature. Re-confirm at the start of each session and at any natural break, watch for signs of distress or objection, and stop at any indication the person wants to withdraw. Short retention is not a bar to consenting, but a person who has lost track of what they agreed to should be re-oriented before continuing.

Are asynchronous AI interviews better or worse for participants with support needs?

Better on access, worse on supervision. Self-paced voice or text interviews remove scheduling pressure, reading barriers, and the demand of performing for a stranger. They also remove the person who would notice distress. Use them for low-risk topics with a visible exit and a route to a human, and never as a way to enrol someone who could not consent in a live session.

How should we record a capacity decision?

Record the process, not a diagnosis: which materials were used, that teach-back was attempted, and the outcome in neutral terms such as "consent not confirmed; participant not enrolled". Do not write clinical speculation into a research or CRM record. That is both a data-minimisation obligation and basic respect.

Do we need ethics approval for this kind of research?

Commercial user research does not usually require IRB or REC approval, but the presence of participants with impaired decision-making capacity is one of the strongest triggers for seeking independent review anyway. If your study is connected with a health condition, or is conducted with or on behalf of a healthcare institution, assume approval is required and check early.

Related resources

• Structured Questions Guide — building consent confirmation and check-ins into the interview itself
• User Research With Children and Teens — COPPA, parental consent, and assent
• Vulnerable Customer Research — evidencing good outcomes under FCA rules
• Trauma-Informed User Research — interviewing safely on sensitive topics
• Research Ethics and Informed Consent — the general framework this article specialises
• IRB Approval for User Research — when independent review is required
• Research Consent Form Templates — plain-language consent materials