About

Why does this exist?

I wanted UX researchers to get reliable, focused answers to career questions they could actually trust, because the answers come from a library of advice written by other UX researchers.

Searching for this stuff on general search engines can be fraught. You go down rabbit holes, and half the time you can’t tell whether you’re reading something useful or someone’s marketing. This is meant to give real answers to real questions — the kind that help or inspire a researcher, or someone trying to become one.

How is this different from asking a general AI assistant?

A general assistant draws on most of the internet. This one draws on roughly 30 articles I have read.

That means it knows far less. It also means I can tell you exactly where every sentence came from, and I would recognize a bad answer, because I have read the source. You can’t audit a library you didn’t read.

Where do the answers come from?

Only from the articles in the library. Nothing is pulled from the open web at the moment you ask.

Your question is compared against the library, the closest-matching passages are retrieved, and those passages are handed to a language model with instructions to answer using them and nothing else, citing the authors and linking back. If the library doesn’t cover your question, it is instructed to say so rather than fill the gap from general knowledge.

Why does it sometimes say it doesn’t know?

Because it doesn’t, and I would rather it told you.

The library is small on purpose, so it will not cover everything. When nothing in it is close enough to your question, the tool says so instead of improvising. That is the tool working, not failing.

How did articles get chosen?

I read every one of them.

The selection started with UXR.design, which itself began as a Google Sheet of links to articles by working researchers — on methods, portfolios, stakeholder work, career moves. Nothing on UXR.design ever arrived automatically. It was curated article by article, and Ask.UXR.Design is built the same way.

Why is it so small?

Because I wanted to be able to stand behind every answer.

A crawler could have handed me a hundred thousand documents. I’d have had no idea what was in any of them, and no way to judge which ones were worth trusting. Keeping the library small enough to read is the only way I could think of to guarantee the tool tells you the truth, credits whoever said it, and admits when it doesn’t know.

Are the authors okay with this?

Every article in the library is here because its author gave me permission.

I wrote to every author whose work was in the original library, and anything I couldn’t clear came out — which removed more than half of it.

I wrote one of these. How do I get it removed?

Tell me and I’ll remove it within 24 hours. Use this form.

The article and every passage derived from it — the text and the numerical representations — are deleted from the database, and I purge the text from backups too. Once it’s gone it can’t be retrieved, which means it can’t appear in any future answer. The Sources list updates within the hour.

The only thing I keep is a short do-not-add record: your name, the article title, the URL and the date. No text, no summary, nothing else. Its only job is to stop me from adding a future version of your article back in. If you’d rather I keep nothing at all, say so and I will.

Can I suggest an article, or offer my own?

Yes, please. This is how the library grows.

Send it here. I’ll read it, and if it fits I’ll write and ask the author’s permission before adding anything. Authors are always credited and linked.

If you’re reading an answer and something’s missing, there’s also a link under the sources on any answer to flag the gap.

Is the library biased?

Yes, in the sense that any curated collection is. It reflects what I chose and who replied.

My goal, the same as with UXR.design, is a library with a wide range of voices, perspectives, ideas and methods. It’s intentionally small and will stay that way. Every source is listed. Every article is credited and linked. Anyone can see exactly what’s in here and suggest ideas for what isn’t.

At launch, a noticeable share of the library is work I wrote or hosted myself, because those were the pieces I could clear immediately. That will change as more authors contribute. I need help from the UX research community to build this. If you’ve written something that fits, please share it with me.

Can I trust the answers?

Trust the sources, and use the answers to find them.

Answers are AI-generated and imperfect. They can misread a nuance or flatten an argument. Every answer links to the articles it drew on. I encourage you to follow the link and read the original.

How do you know the answers are any good?

Two different questions hide in that one, and I can only answer one of them honestly.

I can’t tell you the advice is correct. Most of this library is practitioners writing from experience, and they don’t always agree with each other. For example, there’s no single right answer to how you should build a portfolio.

What I can check is whether the tool represents them faithfully. Does the answer reflect what the author actually wrote? Is every claim traceable to the source it cites? Does it admit when the library doesn’t cover something?

Before launch I ran thirty test questions. Twenty were questions I already knew the answer to — I knew which articles should come back and what those authors actually said. I read every answer against its sources looking for two failures: the wrong articles being cited, and claims attributed to authors who never made them. I found neither.

The other ten were designed to fall outside the library. Five were plainly unrelated, and the tool declined all five. Five were real UX research questions the library doesn’t cover — sample sizes, participant incentives, interpreting eyetracking heatmaps — and rather than declining outright, it answered what it could and said plainly which parts weren’t covered. That’s the behaviour I wanted.

This was an informal evaluation, and I want to be precise about what that means. I designed the question set, ran it, and scored it myself, so it is not independent verification. Thirty questions is a small sample. But it was built deliberately rather than casually: it covers every category and every author in the library, and a third of the questions were chosen specifically to probe where I expected failure — topics where one author has several articles and the tool might cite the wrong one, and questions close enough to the library’s subject matter to tempt a thin answer. It is proportionate to a library this size, and I’ll run it again whenever I change how the tool works.

I’m also watching this in production. Every question is stored without identifying information, along with whether the tool was able to answer it. A rising rate of “I don’t have a curated source on that yet” tells me where the library is thin, which is how I decide what to add next.

What happens to my question?

Your question is stored, with no identifying information attached to it.

What’s kept is the text of the question, whether the tool was able to answer it, and how close the best match was. That’s it. There are no accounts, no tracking, no personal data. I look at those questions to see what people are asking and where the library falls short, which is what tells me what to add next.

Please don’t put personal or sensitive details in a question.

The Privacy Policy and Disclaimer have the full picture.

Is your article used to train an AI model?

No. Nothing in the library is used for training or fine-tuning at any point.

Two companies touch the text at two different moments. OpenAI converts articles and questions into the numerical form used for matching, which means an article’s full text passes through OpenAI once when I add it. Anthropic’s Claude composes the answers, and only ever sees the handful of passages retrieved for a particular question.

Both publish the same position: data sent through their APIs isn’t used for training by default, and is held briefly for abuse monitoring before deletion. Because those defaults can be overridden by account settings, I checked my own rather than take it on faith. Input and output sharing, model feedback and evaluation sharing are all disabled on the OpenAI side. Feedback sharing is off on the Anthropic side and I haven’t joined their model-improvement program.

OpenAI offers free daily usage in exchange for sharing your traffic for training. I declined it, and don’t plan to take it.

Is this free? Will it stay free?

Yes. It’s free, and my intent is that it stays free. No paid tiers, no sponsors, no plans to license the library to anyone.

The one thing I can’t promise is what the running costs look like in a few years. If they became a burden, the most I’d do is ask the community to help cover hosting — a donation to support the tool, not a fee to use it. Access stays free either way. If that ever changed in a way that affected authors, I’d go back and ask them again rather than treat today’s yes as covering a different arrangement.

How current is the library?

It’s curated as I go. Articles are added when authors offer them and I’ve read and cleared them, so the library grows slowly and deliberately rather than on a schedule.