home/resources
the shelf · updated july 2026
A working library for conversation design, technical writing and AI documentation, UX writing, and the research methods used to test content. Every entry earned its place by changing something I do.
how a title earns a place here
I triage on the table of contents, never on the title. Four separate batches taught me the same lesson: a book with a verifiable publisher or an author with a public history rewards the read, and a title promising the complete guide to everything usually has nothing inside it.
So this page holds two kinds of entries. Core titles changed a specific decision I make in production, and I say which one. Shelf titles are the canon I open when a project asks a question I cannot answer from practice. Anything I could not verify, or read closely enough to describe, stayed off.
Books carry no purchase links, by choice. Free standards and open frameworks link straight to the source.
The craft layer: how a system takes turns, repairs itself, and holds a persona. The first four are where the discipline actually lives.
core · book · 2018
Erika Hall · A Book Apart
The argument that conversation is the original interface and that most software is a poor conversationalist. Hall turns cooperative principles into design constraints, which is the move that makes the rest of this shelf usable.
core · book · 2021
Diana Deibel & Rebecca Evanhoe · Rosenfeld Media
The most practical book here. Persona, sample dialogs, error handling, and inclusive language, each with scripts concrete enough to lift into a spec. This is the one I recommend first to writers moving into conversation design.
core · book · 2016
Cathy Pearl · O'Reilly
The canonical voice text, and still current where it counts. Pearl's confidence thresholds for confirmation and error handling are the tiers I design against whenever a system has to decide between acting, confirming, and handing off.
core · book · 2019
Robert J. Moore & Raphael Arar · ACM Books
The Natural Conversation Framework, out of IBM Research. Its expandable sequences give dialog a structure that survives contact with real users, because every exchange has room for the repair the user will actually need.
core · book · 2024
Michael McTear & Marina Ashurkina · Apress
McTear is the academic authority of the field. Chapter 8 on evaluation, which separates lab from in-the-wild and expert from simulated users, is the taxonomy sitting under every eval plan I write. Chapter 10 asks what happens to the conversation designer's role, and answers it with evidence.
shelf · book · 2025
Victor Dibia · self-published
Dibia builds AutoGen at Microsoft Research. The chapter on UX principles for multi-agent systems names the reliability problem that a generator-plus-reviewer architecture creates for the person waiting on the other side.
shelf · book · 2026
Mezba Uddin · Apress
Tiered fallback, persona protection, and A/B testing on tone, described inside a tool that ships to enterprises. Rare for showing conversation craft at the level where it survives a release.
core · paper · 1974
Harvey Sacks, Emanuel Schegloff & Gail Jefferson · Language
Where turn-taking was first described from recordings of real talk. The finding with teeth: listeners predict the end of a turn from syntax and prosody, so silence is a poor signal for it. A system that cuts in at silence breaks a norm users feel as rudeness without being able to name it.
core · paper · 1991
Herbert H. Clark & Susan E. Brennan
Common ground and the principle of least collaborative effort. This is the theory underneath every confirmation step, and the reason a well-placed acknowledgement costs less than the misunderstanding it prevents.
shelf · paper · 1975
H. P. Grice
Quantity, quality, relation, manner. I run the four maxims as a lint pass over system prompts: most bloated agent replies are a violation of quantity wearing a friendly voice.
shelf · paper · 2021
Langevin et al. · CHI
A heuristic set built specifically for agents. Four evaluators and one day produce a defensible review before you have a single user, which makes it the cheapest quality gate available.
Research methods applied to words: comprehension, task success, and dialog quality. This is the part of my practice people are most surprised is measurable, so the shelf leans towards method and statistics.
core · paper · 1997
Marilyn Walker et al. · ACL
Task success and dialog cost combined into a single measurable model. Nearly thirty years old and still the cleanest way to argue that one dialog design is objectively better than another rather than merely newer.
core · book
Jeff Sauro & James R. Lewis · Morgan Kaufmann
Which test, which sample size, which confidence interval. The book that lets me put numbers on a content change and state the limits of those numbers in the same breath, which is what keeps a finding credible with engineers.
core · book
Erika Hall · A Book Apart
Research sized for teams that ship. I hand the evaluative research chapter to anyone who says there is no time to test, because it reframes the question as how small a study can be and still tell you something true.
shelf · book
Jeffrey Rubin & Dana Chisnell · Wiley
The procedural baseline: recruiting, task design, moderation, reporting. When a study needs to hold up to scrutiny, this is the reference that says how it should have been run.
shelf · book
Steve Portigal · Rosenfeld Media
Interviewing treated as a technique with failure modes. The chapters on silence and on following the thread improved my moderation more than any template did.
shelf · book
Tom Tullis & Bill Albert · Morgan Kaufmann
The metric catalogue: performance, self-reported, behavioural, and how to combine them without double counting. Useful when a stakeholder asks for one number and deserves a better answer.
shelf · book
Mike Kuniavsky · Morgan Kaufmann
The wide survey of methods, strongest on choosing between them. I reach for it when the research question is still soft and the method has to be argued rather than assumed.
core · standard · free
ISTQB · free download
Testing an AI system is a statistical problem: one passing example proves nothing about a probabilistic system, and the suite has to be sampled from real operational traffic. These syllabi are free, and they give you shared vocabulary with the QA team, which is worth more than being right alone.
shelf · standard · free
W3C · free
Read as a content standard rather than a code standard, it sets testable rules for link text, error messages, headings, and language. Most of what it asks of writers is legible in an afternoon.
My first discipline, and the one changing fastest. The frameworks here settle arguments about structure; the newer entries deal with documentation that machines read as well as people.
core · book · 2nd ed.
Jared Bhatti et al. · Apress
The canonical book of the profession, written by practitioners out of Google and Stripe. Two ideas earn their keep weekly: friction logs, and a content type taxonomy that stops teams writing a tutorial where a reference belongs.
core · framework · free
Daniele Procida · free
Four modes: tutorial, how-to, reference, explanation. It ends most arguments about where a page belongs, because it locates each mode by what the reader is doing at the moment they arrive.
shelf · book · 2023
Diana Lakatos · Apress
Documentation strategy end to end: editorial workflow, accessibility, community contribution, and sustainability. The chapter on measures of success is where docs metrics stop being vanity numbers.
shelf · book
Andrew Etter
Short enough to read in one sitting, and the clearest statement of the docs-as-code position: lightweight markup, version control, static sites, and docs that live beside the thing they describe.
shelf · book
Christopher Gales & the Splunk documentation team
Written by a working team about running a working team: information architecture, review cycles, and the organisational politics of owning content nobody else wants to own.
core · standard · free
open proposal · free
A convention for telling language models what a site contains and where the authoritative text lives. I implemented it on this site, which is the fastest way to learn where a proposal like this is thin and where it holds.
shelf · reference · free
Google · free
The default I inherit on most teams. Its value is being decided: a house style that already answered the small questions is worth more than a better style still under debate.
shelf · reference · free
Microsoft · free
Stronger than Google's on voice, bias-free communication, and the vocabulary of AI features, which makes it the better reference when the product speaks in the first person.
shelf · book
Robin S. Adams & Junaid A. Siddiqui (eds.) · Purdue University Press
Researchers analysing one shared corpus of design reviews through many lenses: gesture, discourse, turn-taking, professional vision. It gives review a literature, which matters once you are designing an assistant that takes part in code and document reviews.
Words inside the interface, where the constraint is the flow rather than the page.
core · book · 2020
Michael J. Metts & Andy Welfle · Rosenfeld Media
The book that argues writers belong in the design process at the point where flows are decided. Its chapter on error states is the one I have quoted most often in reviews.
core · book · 2019
Torrey Podmajersky · O'Reilly
Voice charts and a repeatable method for connecting product copy to organisational goals. The voice chart is the artefact that makes tone reviewable by people who are not writers, which is what gets it enforced.
shelf · book
Sarah Winters · Content Design London
Content design as the practice of answering a user need with the format the evidence supports, which sometimes means a table and sometimes means removing the page. Rooted in the GOV.UK work.
shelf · book · 2014
Nicole Fenton & Kate Kiefer Lee · New Riders
Plain guidance on voice, tone, and writing for people having a hard day. It ages well because it is about judgement rather than trends.
shelf · book
Janice (Ginny) Redish · Morgan Kaufmann
Writing for people who scan, backed by usability evidence rather than assertion. The before-and-after rewrites are still the most convincing teaching device I have seen for this material.
core · book
Eric Meyer & Sara Wachter-Boettcher · A Book Apart
Stress cases in place of edge cases. Working on clinical and regulated software, this reframing does the most work of anything on the shelf: the person reading your error message is often having the worst day of their year.
Linguistics and HCI: the layer I open when a project asks a question that practice cannot answer. These reward being read slowly, once.
core · book · 2015
Bing Liu · Cambridge University Press
The reference text, and the chapter on mining intentions formalises intent as action, target, intensity, holder, and time. The most rigorous treatment of intent I have found, and the source of two failure cases I now test for: pseudo-intentions, and intentions belonging to someone other than the speaker.
shelf · book · 1996
Herbert H. Clark · Cambridge University Press
Language as joint action. The long-form version of the grounding paper, for when a design argument needs the full theory rather than the summary.
shelf · book
Peter Auer (ed.) · Routledge
Conversation analysis meeting multilingualism. It matters in production because repair strategies and filler words are language dependent, so a fallback tuned in English quietly stops working when the same product ships in another language.
shelf · book · 1951
Zellig Harris · University of Chicago Press
The distributional hypothesis, written in 1951: words that occur in the same environments carry related meaning. It is the direct intellectual ancestor of embeddings, and a good reminder that this field has a longer memory than its release notes suggest.
shelf · book · 1983
Card, Moran & Newell
Where GOMS and keystroke-level modelling come from. The empirical floor under any claim about cognitive load, including the ones I make about how much a prompt should ask at once.
shelf · book · 2007
Bill Moggridge · MIT Press
Oral history of the field from the people who built it. Read for perspective on how long the current arguments have been running.
shelf · book · 6th ed.
Rogers, Sharp & Preece · Wiley
The textbook. Useful as a map when a project crosses into territory I have only practised in, and its method chapters are a fair check on whether a study design is standard or improvised.
Nothing here matches that combination yet.
A shelf is a claim about what you think the work requires.
This list is revisited every six months, and it grows the way it was built: a title enters after it changes a decision, and leaves when I stop reaching for it. If something belongs here and is missing, or you want the argument for why one of these earned its place, write to me.
The practice these feed into is documented in the case studies on the work.