Voice and terms

1 · Voice

Start from who is reading.

Our readers are students, engineers and researchers, policy and law people, and funders. Most are new to AI safety, many read English as a second language, and all are busy. The voice stays the same for all of them; only the tone changes.

Plain

Short sentences, common words, the main point first.

MeansWrite so a second-year student can follow it on a phone. Define a term the first time it appears.

Doesn’t meanDumbing down the ideas, or dropping the technical detail an engineer needs.

ExampleA free six-week online course on how to make AI systems safe.

Warm

Welcoming to newcomers, generous with credit, human in tone.

MeansTalk to the reader as a peer. Say who runs the session. Make the next step easy to find.

Doesn’t meanExclamation marks, emoji in body copy, hype, or jokes in error messages.

ExampleNo background in AI? That’s fine. Most people in our reading groups started here too.

Exact

Specific facts over adjectives. Dates, numbers, names and sources.

MeansSay how long, how much, who, when and where. Say what we don’t know.

Doesn’t meanJargon, hedging every sentence, or claiming more than the evidence shows.

Example12 weeks, mentored, 4–6 hours a week. Applications close Fri 7 Nov.

2 · Tone by situation

Same voice, different moments.

Match the reader’s state of mind. Someone filling in a form wants clarity; someone who was not selected wants respect; someone reading a policy brief wants evidence.

Situation	Tone	Example line
Homepage	Confident, inviting, concrete	Learn AI safety with people across India. Free courses, reading groups and research.
Course page	Clear and practical; facts before persuasion	Six weeks, online, about 5 hours a week. Free. Next cohort starts Mon 12 Jan.
Event invite	Friendly and specific	Join us in Bengaluru on Thu 23 Oct, 6:30 pm, to discuss scalable oversight.
Application form	Neutral, helpful, brief	Tell us why you want to join. 150 words is plenty.
Rejection email	Respectful, direct, with a next step	We couldn’t offer you a place in Cohort 3. You can join a reading group this month, and applications for Cohort 4 open in March.
Error message	Calm and specific; no blame	Enter an email address, like name@example.com.
Policy brief	Measured, sourced, neutral	The EU AI Act’s rules for general-purpose AI models have applied since August 2025. How they will be enforced is still being worked out.
Social post	Lively but factual; one idea per post	Reading group this Thursday in Delhi: what interpretability can and can’t tell us. All welcome.

3 · Naming

One name for each thing.

The live site uses up to four names for the same course. Pick the official name below and use it everywhere: website, emails, certificates, slides and social posts.

Programme or thing	Use	Don’t use
Organisation	AI Safety India. After first use on a page: AISIN, sparingly.	AI Safety Initiative of New India, AISI, AISIN in a headline or as the first mention
Technical course	Technical AI Safety (course)	Fundamentals of Safe AI, TAIS bootcamp
Governance course	AI Governance (course)	Frontier AI Governance, Fundamentals of AI Governance
Reading groups	Reading groups; with a place: Delhi reading group, Bengaluru reading group, online reading group	Book club, study circle, Bangalore
Speaker sessions	Speaker sessions	Talks series, fireside chats, masterclass
Research	Research fellowship (the 12-week mentored programme); research groups (ongoing teams)	Research intensive, research accelerator
Career advising	One-to-one career advising; button: Book a call	1:1 career coaching in prose, career counselling
Hubs	Delhi AI Safety Collective; AI Safety NIT Agartala	AI Safety - NITA, DASC without first spelling it out
University groups	University groups; a named group is “AI Safety [university name]”	Campus chapters, college clubs
Cohorts	Cohort 3 (capital C, numeral)	3rd cohort, Batch III, cohort three
Dates and times	Thu 23 Oct, 6:30 pm (all times IST; add “IST” when readers may be abroad)	10/23, Oct 23rd, 18:30 hrs, 6:30PM
Numbers	Numerals for 10 and above and for all measures: 6 weeks, 12 weeks, 5 hours. Large figures: 1,200 learners; ₹5 lakh, with “(about US$6,000)” for international readers	1200+ without context, thousands of when you know the figure
Price	Free	100% free, FREE, no cost to you!

4 · Spelling and style

Indian English, sentence case.

Indian English follows British spelling. Choose one form and keep it; mixed spellings on one page look careless and confuse search.

Spellingprogramme (but “computer program”), organisation, centre, enrol / enrolment, behaviour, recognise, colour, licence (noun), practise (verb), judgement
CapitalisationSentence case everywhere: headings, buttons, navigation, labels, email subjects. Capitalise only proper nouns and official names (AI Governance, Cohort 3). No tracked all-caps labels.
Oxford commaNo Oxford comma, unless a sentence is unclear without it: “students, engineers and researchers”.
NumeralsOne to nine in words in prose; numerals for 10 and above, for all units (6 weeks, 5 hours) and in UI. Use Indian grouping with lakh and crore for rupees.
DatesThu 23 Oct in UI and listings; Thursday 23 October 2026 in formal letters. No ordinals (23rd), no slashes.
Times6:30 pm, 12-hour, lower case, a space before “pm”. Ranges: 6:30–8 pm with an en dash. Times are IST.
Abbreviations“AI” needs no expansion. Write “AI Safety India” first, then “AISIN” if you need a short form; never more than once or twice per screen. Spell out every other acronym on first use.
One-to-one“one-to-one” in prose. “1:1” is allowed only in compact UI labels such as a tab or chip. Never “1-on-1”.
PunctuationNo exclamation marks in headings or buttons. En dash for ranges, em dash sparingly. Curly quotes.

5 · Words we use and avoid

Swap adjectives for facts.

Hype words make a free community course sound like an advertisement, and they put off the researchers and policy people we most want to reach.

Avoid → use instead

premier, world-classSay what makes it good: “taught by researchers from…”
revolutionary, cutting-edge“new”, or name the method
sovereign“India’s own”, or describe the policy plainly
intensive“6 weeks, about 5 hours a week”
battle-tested, high-voltageDrop it, or give evidence
leverageuse
unlockget, open, start
journey (in marketing)career, path, next step
Initiative of New IndiaAI Safety India

Words that fit the voice

learn, read, discussWhat people actually do with us
free, online, in personFacts readers filter by
mentored, part-timeDescribe the format honestly
community, peersWho you will meet
risk, safety, oversightPlain field terms; define the rest
youAddress the reader directly

6 · Interface labels

Buttons start with a verb.

Active voice, sentence case, two or three words. Say what will happen. No cute labels, no “we” in labels and no “Your” in navigation (“Applications”, not “Your applications”).

ApplyMain action on a course or fellowship page while applications are open

Apply for the next cohortWhen applications are closed and you collect interest

RSVPEvents and speaker sessions

Book a callOne-to-one career advising

Join a reading groupReading group listings and the homepage

SubscribeNewsletter sign-up only

See syllabusSecondary action next to Apply

Get startedFirst step of a multi-step flow

ContinueEvery middle step; never “Next”, “Proceed” or “Let’s go”

DoneFinal step, after the action has happened

BackPrevious step; keeps what was entered

SaveKeeps changes without submitting

CancelLeaves without changes; never the primary button

Multi-step flows

The same three words in every flow, so applicants always know where they are.

Get started→Continue→Continue→Done

Link text

Do

Describe the destination: “Read the AI Governance syllabus”
Make the link make sense out of context, for screen readers

Don’t

“Click here”, “Learn more”, “Read more” on their own
Bare URLs in body text

7 · Messages

Say what happened and what to do next.

Put errors next to the field they are about. Say how to fix the problem, never blame the reader, and never write “Oops” or “Something went wrong” on its own.

Required field

Name

Enter your name.

Pattern: imperative verb + what’s needed. Not “This field is required” or “Invalid input”.

Invalid email

Email

priya@gmail

Enter an email address, like name@example.com.

Pattern: say the fix and show an example. Not “Invalid email!”

Deadline passed

Applications for Cohort 3 closed on Fri 7 Nov.Cohort 4 opens in March. Get an email when it does.

Apply for the next cohort

Pattern: the fact, then the next option.

Empty state: no events

No events scheduled yet.New sessions are added every month. Reading groups meet weekly in the meantime.

Join a reading group

Pattern: what’s empty, why, and something useful to do now.

Empty state: no results

No results for “interpretibility”.Check the spelling or try a broader word, like “interpretability” or “governance”.

Pattern: repeat the search; suggest a fix.

Success

Application received.You’ll hear from us by Fri 21 Nov at the email you gave.

RSVP confirmed.Thu 23 Oct, 6:30 pm, Bengaluru. A calendar invite is on its way.

Pattern: confirm in two or three words, then what happens next. No “Yay!”

Loading

Sending your application…

Say what is happening with a present-tense verb and an ellipsis. Not “Please wait” or “Hang tight!”. If it takes more than a few seconds, say so.

8 · Writing about AI safety

Accurate, not alarmist or dismissive.

Our credibility with researchers, policymakers and funders depends on how carefully we describe the field. Write as if an expert who disagrees with you will read it.

Be accurateDescribe risks in proportion to the evidence. Don’t call risks inevitable, and don’t wave them away.
Cite sourcesLink to the paper, report or statement. Name the author or organisation and the year.
Say what we don’t know“Researchers disagree about…”, “There is little evidence yet that…”. Uncertainty stated honestly builds trust.
Avoid jargon, or define itTerms like alignment, interpretability, RLHF and evals get a one-line definition on first use, or a link to the glossary.
No sci-fi imageryAvoid Terminator, robot uprising, doomsday, rogue AI, Skynet, apocalypse and “existential dread”. Describe the actual mechanism.
Be fair to peopleDisagree with ideas, not people or labs. Don’t attribute motives.

DoSome studies report that models can behave differently when they seem to be under evaluation (cite the paper, author and year). It is not yet clear how common this is outside the lab.

Don’tAI is already deceiving its creators. The robots are coming, and only we can stop them.

9 · Hindi and other languages

Written by people who speak it.

Plan for localisation from the start: short sentences, no idioms or wordplay that only work in English, and room in layouts for text that runs 30% longer.

Native speakers, not machines

Hindi and other Indian-language text is written or reviewed by native speakers. Machine translation can be a first draft, never the published copy.

Keep names as they are

Don’t translate AI Safety India, course names or hub names. Write them in Latin script inside Hindi text. Translate descriptions, not names.

Set Devanagari in Anek Devanagari

AI Safety India के साथ AI सुरक्षा सीखें।

Mark text with lang=”hi” so it picks up Anek Devanagari and a taller line height.

10 · Before you publish

Ten checks, two minutes.

Run through this before anything goes on the website, in an email or on social media.

Name“AI Safety India” in full first; AISIN only after, and sparingly.
Programme namesOfficial names from the naming table; no old variants.
SpellingIndian English throughout: programme, organisation, enrol.
Sentence caseHeadings, buttons and labels; no all-caps labels.
No hypeNone of the words in the avoid list; facts instead of adjectives.
Dates and times“Thu 23 Oct, 6:30 pm”, IST, correct day of the week.
Buttons and linksStart with a verb; no “click here”; standard labels.
MessagesErrors next to the field, say how to fix, no blame, no “oops”.
AI claimsSourced, proportionate, and honest about uncertainty.
Read it aloudIf a sentence is hard to say, make it shorter.