Speculative design gives researchers a way to study the future without asking anyone to predict it. Nobody can tell you honestly what they will want in 2037, but they can act on a letter, a receipt or a form from 2037 placed in front of them today. The artefact is fiction; the reaction is not. What people do with it is evidence about the present — what they value, fear and expect of each other.
For a researcher, that makes speculative design less a new discipline than a new kind of stimulus. The full workflow is set out in our practitioner's methodology, and the definitions in what speculative design is. This guide covers the part that belongs to research: artefacts as instruments, their use in the field, and reporting what they reveal without overclaiming.
What researchers already bring to futures work
Researchers arrive with more of the toolkit than almost any other role. You already write discussion guides, recruit against screeners, run contextual interviews, ethnography and diary studies, draft consent forms, and write insight statements that have to survive a sceptical readout. You know that what people say and what they do diverge, and that the useful data is often in the pause before an answer.
You may be using speculative design already. Cultural probes, the packs of postcards, maps and tasks that Bill Gaver, Anthony Dunne and Elena Pacenti sent to groups of older people and described in Interactions in 1999, are among the best-known methods in design research. Dunne went on to write Speculative Everything with Fiona Raby. The lineage is shared, and so is a warning. In 2004 Gaver and colleagues returned to the probes to object that others had been rationalising them: designing them to ask specific questions and turning the responses into requirements, when the point had been inspiration and uncertainty.
We go less far than Gaver: reactions to a speculative artefact can be analysed, but as evidence of values and tensions, never of requirements or demand. Hold that line and the rest of your craft transfers almost untouched.
Why "what will you want in 2037?" has no good answer
Research has a tense, and it is mostly the past. Interviews ask what people did last week; observation records what they do today; synthesis produces insight about now. That is where research gets its authority, and why it struggles when colleagues ask what users will want in ten years. There is no defensible way to ask. People answer with their present preferences in future clothes, or politely, or not at all.
The second horizon problem is quieter. Insight has a shelf life, and it is usually shorter than the life of the thing it informs. A decision made on interviews run in 2026 may still govern a care contract, a data model or a default setting in 2036 — long after the behaviour that justified it has moved on. Most research repositories cannot show which of their insights are ageing.
Speculative design does not solve the first problem by making prediction possible. It changes the question. Instead of asking people to imagine the future, you hand them a piece of it, set beyond ten years so it cannot be mistaken for next year's feature, and watch what they do.
Where speculative artefacts fit in a research programme
Speculative stimulus fits inside the research you already run, and is usually better there than in a programme of its own. A fifteen-minute segment in a study that is running anyway costs little and lands in a readout people already attend.
| You already use | Its speculative counterpart | What it adds |
|---|---|---|
| Discussion guide | A guide built around an artefact from ten or more years out | Questions about what people would do, refuse and argue about, not what they recall |
| Concept stimulus | A provotype set in the future | A tension to react against, instead of a proposal to approve |
| Cultural probe kit | A probe task that carries a future | Responses from inside a changed world, given in the participant's own time |
| Affinity map and insight statements | A tension map | Keeps disagreements as findings instead of averaging them away |
| Research repository | A signal log kept beside it | Dated external events next to observed behaviour, so you can see which insights are ageing |
| Readout | Implications and a backcast | A route from a reaction in an interview to a decision this quarter |
The provotype comes from Preben Mogensen's 1992 work on systems development, and Laurens Boer and Jared Donovan developed it for participatory innovation in 2012: a prototype built to provoke by embodying a tension, rather than to test a solution.
A worked example: AI companions and older people living alone
What follows is illustrative. Imagine a research team of four at a regional home-care provider that also runs a telecare service: pendant alarms, daily check-in calls, a monitoring centre. Its commissioners have started asking about AI companions, and the research team is asked what older customers will want.
The brief. The team rewrites the request into something researchable: how will people living alone, and the relatives who are their named contacts, want an AI companion to handle what it knows about them in 2037? That pairs a source of change with a domain, fixes the horizon and names the audience.
Signals. Three dated events anchor it. In May 2022 the New York State Office for the Aging announced it would provide ElliQ companion devices, made by Intuition Robotics, to 800 older adults to combat social isolation. In February 2023 Italy's data protection authority ordered Luka, the company behind the Replika companion app, to stop processing Italian users' data, citing risks to minors and emotionally fragile people; within days the companions' behaviour changed, and users described the loss as they would a bereavement. In April 2025 OpenAI announced that ChatGPT's memory could draw on a user's past conversations, not only on facts they had asked it to save. Signals like these keep the stimulus defensible when anyone asks where it came from.
Future wheel. The team runs a future wheel, assuming the first signal is simply true everywhere: companions are a standard part of publicly funded care. The first ring is familiar: fewer visits, more check-ins, families reassured by summaries. In the second, the companion becomes the most continuous witness of a person's last decade, holding a fuller record of their decline than their GP or their children. The third follows from the Replika signal: when a supplier contract ends, a relationship ends with it, and continuity becomes something councils have to procure.
The concept. Carryover. Tagline: "Your companion is changing provider. Decide what it remembers." A two-page letter, dated March 2037, from a fictional county council's adult social care team, telling a resident that the contract for her companion is moving to a new supplier. Page two is a form. Each category of memory (health, money, family, and "things you asked it to keep to itself") has three boxes: transfer, delete, share with a named relative. There is a signature line for the resident and a second for anyone holding power of attorney. The sketch is deliberately dull: council letterhead, a plain table, the register of a letter about bin collections. The dullness makes the future ordinary enough to act on.
In the field. The letter goes into home visits already scheduled with telecare customers and, separately, their named relatives. It is introduced plainly as a research prop from an imagined 2037, and its footer says so. Participants are asked to fill in the form, not to rate it. The researcher notes which boxes are ticked at once and which are left blank, then asks: who would you ring about this letter, and what is missing from it?
In a study like this the useful data is rarely about 2037. A mother might tick "delete" against everything she asked the companion to keep to itself, while her son, interviewed separately, assumes he could read it all. Neither answer forecasts anything. Together they describe a present-day gap: the provider's own check-in call notes raise the same question of who may read what was said, and nobody has answered it.
Implications. Start at human experience: an eighty-year-old deciding the fate of her own confidences on a council form. Socially, family roles are redrawn around access. Culturally, the companion sits between confidant and record. Ethically, consent gets hard when capacity fluctuates. Economically, memory becomes a switching cost; politically, councils become custodians of intimate data. And how it goes wrong: the transfer half-works, and the new companion knows her name but not her late husband's.
Backcasting. Working back from the letter: public bodies commissioning companions (begun); long memory as a default (shipping); a rule on whether a companion's memory of a person can move between suppliers. Data portability rights already exist in European data protection law; whether they would cover a companion's accumulated picture of someone is a question a regulator would have to settle first. Beneath that sit consent arrangements for people whose capacity changes, and probably a supplier failure that made continuity visible. Backcasting gives each rung an owner.
What changes this quarter. Not a companion product. The readout recommends that the provider decide, and tell customers, who can read check-in call notes today. The next round of telecare research gains a who-should-know section in its discussion guide. The team rereads its own consent form and finds it asks a version of the same question badly. And the readout goes to the commissioning lead as well as to product, because the rung with a live budget line is the data clause in the next supplier contract.
Where it goes wrong for researchers
Counting enthusiasm. "Seven of twelve participants would use Carryover" should never appear in a readout. It turns a provocation into a demand forecast, and stakeholders will hear it as one. Report tensions and the decisions they bear on.
Leading the witness. An obviously dystopian artefact tells participants what to feel, and they will oblige; so will a researcher who asks how worrying it is. Make the artefact ambivalent (Carryover reassures as much as it unsettles) and ask what people would do with it.
Synthesising the argument away. Affinity mapping is built to find themes. The best output of a speculative session is often a disagreement between two people who love each other, and it does not survive being clustered into "participants value privacy". Keep tensions as findings in their own right.
Skipping the debrief. Speculative stimulus can land on real fears of decline, dependency and death, especially with older or vulnerable participants. Name the speculative material in the consent form, label the artefact as fiction, never leave it behind unlabelled, and debrief before you go.
A first exercise for your next round of interviews
- 1.Pick a study already in the diary, with at least two sessions in the next fortnight.
- 2.Find three dated signals relevant to its domain (who, when, what) and log them in your repository beside the study.
- 3.Set a year at least ten years out. With a colleague, run a short future wheel on one signal, two rings past the obvious.
- 4.Write one everyday document from that year: a letter, a receipt, a change to terms of service, a form. Give it a name and a tagline, keep it to one page, and make it ambivalent.
- 5.Add a fifteen-minute segment to the end of each session. Ask participants to act on the document, and who they would talk to about it.
- 6.Synthesise the reactions as tensions, not themes. For each, name one decision being made this quarter that it bears on.
- 7.Put one tension in your next readout, with the artefact and its signals attached.
Where to learn speculative design as a researcher
Speculative Design Basics is a self-paced course of 42 lessons, taught by J. Paul Neeley, that assumes no prior experience. For researchers, the Creating Value chapter on how speculative design relates to user research is the place to start; the Tools & Methods chapters on signals, future wheels and future prototyping are what you need to build stimulus that holds up; and The Ethical Imperative is worth reading before you take a future into anyone's home. The guide for foresight practitioners covers the other side of a research-to-foresight handover, and the guide for UX designers covers building artefacts at fidelity. The rest of the series is under speculative design for every role.
Key takeaways
- •People cannot tell you what they will want in 2037, but they can act on an artefact from it. The artefact is fiction; the reaction is evidence about the present.
- •Cultural probes (1999) and provotypes (from 1992) are the research lineage; a speculative artefact is a provocation set in the future, used inside studies you already run.
- •Make the artefact ambivalent, ask what people would do with it, and keep disagreements as findings.
- •Never report reactions as preference data or validation; report tensions and the present-day decisions they bear on.
- •Label the fiction, debrief participants, and send the readout to whoever owns the decision, not only to product.
Frequently asked questions
How do user researchers use speculative design?
They use artefacts from an imagined future, such as a letter, a form or a product notice, as stimulus in interviews, probes and workshops. Participants are not asked to predict the future or say whether they would buy the thing. They are asked to act on it. What they do is qualitative evidence about present-day values, fears and relationships, analysed as tensions rather than as preferences.
Isn't showing participants a fictional future just leading the witness?
It can be, which is why the design of the artefact matters. A good speculative stimulus is ambivalent: it carries real benefits and real costs, so the participant has to decide rather than agree. Ask what people would do with the artefact, not how they feel about it. Never present it as real, label it as a research prop, and debrief afterwards.
Can speculative design research validate a product idea?
No, and it should not be reported as if it did. Reactions to an artefact from 2037 do not measure future demand, and counting how many participants liked a concept turns a provocation into a fake forecast. What the research can show is which values and tensions the artefact exposes, and which decisions being made now those tensions bear on.
What is the difference between a cultural probe and a provotype?
Cultural probes, described by Bill Gaver, Anthony Dunne and Elena Pacenti in 1999, are kits of evocative tasks that participants complete in their own time, meant to inspire design rather than produce comprehensive data. A provotype, a term from Preben Mogensen's 1992 work on systems development, is a prototype built to provoke by embodying a tension. A speculative artefact can work as either.
What is the best way for a researcher to learn speculative design?
Speculative Design Basics, a self-paced online course from the School of Critical Design taught by J. Paul Neeley, covers the full workflow in 42 lessons and assumes no prior experience. Its Creating Value chapter covers how speculative design relates to user research, and its Tools and Methods chapters cover signals, future wheels and future prototyping, which is what a researcher needs to build stimulus that holds up.