Cultural calibration — Dashain, Tihar, wedding season · AI for Marketing & Content

A model that has read terabytes of English text and a few gigabytes of Nepali text will, when asked to write a Dashain greeting, often produce something that calls the holiday the festival of lights, references a tika that does not exist in this country, and pictures a family scene that could be anywhere from Jaipur to Jakarta. The prose will be fluent. The grammar will be correct. The mistake will be cultural, and it will be invisible to anyone whose only check is whether the sentence sounds good in English. This section is about the cluster of cultural errors AI makes most reliably for Nepali audiences, why the errors are structural rather than random, and the small workflow habits that catch them before they reach the page.

The Dashain conflation

The single most common AI mistake on Nepali festival copy is the Dashain conflation — calling Dashain the festival of lights, blending it with Diwali, or describing it as a shopping event. The model has read far more English text about Diwali than about Dashain; when you ask it for festival copy aimed at South Asian readers, it reaches for the higher-frequency neighbour and dresses it in Nepali clothes.

Dashain is not the festival of lights. Dashain is fifteen days, anchored by Vijayadashami, marked at the senior end by tika on the forehead with rice, curd, and red, and at every household end by jamara sprouting in the prayer room, blessings from elders, the long return journeys home that empty the cities for two weeks, and the family reckoning of who came back, who could not, and who was on the phone from Doha or Sydney instead. The emotional centre is not light. The emotional centre is hierarchy, blessing, lineage, and the kind of homecoming that costs a Korea worker a month of overtime.

A draft that calls this a festival of lights will read, to a Nepali audience, as if the writer has never been here. It is the kind of error that does not get politely overlooked. It is the error that confirms the suspicion that the brand outsourced its Dashain campaign to an algorithm.

The Tihar five-day specifics

The second most common mistake is the Tihar flattening — treating the festival as one undifferentiated five-day block rather than the carefully sequenced ritual it is. Each day of Tihar has its own emotional register, its own visual world, and its own commercial opening if you understand what is actually being honoured.

Kaag Tihar — crows are fed in the morning. The mood is quiet, reflective, slightly mythic. Not a sales day.
Kukur Tihar — dogs are honoured with garlands and tika. The most photographable day; brands selling pet food, household goods, or anything with a soft emotional register have a clean entry here.
Gai Tihar / Lakshmi Puja — cows are honoured in the morning; Lakshmi Puja runs into the night, with marigold and oil lamps lining doorways. This is the visual peak Tihar is usually pictured as. Financial brands, jewellery brands, and home brands have their largest single moment of the year here.
Goru Tihar / Mha Puja — oxen are honoured; in the Newar community the same day is Mha Puja, the worship of the self, and the Newar New Year. Treating this day as generic Tihar misses both communities at once.
Bhai Tika — sisters and brothers, the seven-colour tika, the closing emotional note. Family-oriented and intensely specific; sentimental brand copy lives or dies here.

A model that writes the same Tihar caption five days in a row, with the same marigold imagery and the same vaguely lit oil lamp, is telling the audience the brand has not been paying attention.

Other festivals the model regularly misreads

Beyond the Dashain and Tihar core, a handful of other festivals draw predictable AI errors.

Holi gets coded as Indian. The model has read a great deal of English text about Holi in the Vrindavan and Mathura register and very little about Holi as it is actually celebrated in the Tarai (where it is large and central) and in Kathmandu (where it is smaller, younger-skewing, and tonally distinct from the Indian register). A Holi campaign produced from the default model is a Holi campaign for Indian readers with Nepali names pasted in.

Chhath is largely invisible to the default model. The festival is enormous in Madhesh, runs four days, and has a visual world — the river ghats at sunrise, the women in traditional dress, the sun being offered arghya — that the model will not produce unless explicitly briefed. Brands serving the Tarai market that skip Chhath are skipping the largest religious moment of their actual customer base.

Lhosar is plural — Tamu Lhosar, Sonam Lhosar, Gyalpo Lhosar — each celebrated by different communities at different times of the year. The model will collapse the three into one generic Tibetan-styled image if you let it. The fix is naming which Lhosar you are writing for and which community you are addressing.

Eid and Christmas in the Nepali Muslim and Nepali Christian communities are small in population terms and entirely real. A brand that runs Eid copy that looks like it was scraped from a Gulf campaign, or a Christmas copy that looks like it was scraped from a Hallmark site, has signalled that it sees these communities only when it remembers to.

Wedding season is mostly missing from default outputs

The Nepali wedding season runs roughly Mangsir through Magh — late November to February — with a smaller second window in spring. It is the highest spending season of the year for jewellery, gold, textiles, banquet venues, photography, beauty, and a long tail of services around them. It is also almost completely missing from default AI imagery and copy.

The default model, asked for a South Asian wedding image, produces a Hindi film wedding — red lehenga, sangeet, a particular gold register, a particular musical cue. A Nepali wedding is its own visual world. Daura suruwal for the groom in many communities, gunyo cholo and the long red and gold register for the bride, the multi-day ritual sequence that does not map cleanly onto the Hindi film template, the Newari wedding traditions which are different again, the village procession in some communities, the banquet hall in others. None of this appears unless you put it in the brief.

The same gap exists in copy. A default model writes wedding copy in the register of Mumbai jewellery brands — “your big day, your dream wedding” — which lands flat in Nepal because that is not the register a Nepali family uses when it talks about a wedding. The Nepali register is more about the families joining, the elders being honoured, the household being formed. A model can hit that register if you tell it to. It will not hit it on its own.

The workflow that catches the cultural errors

The mistake brands make is treating cultural calibration as a creative review at the end of the process. By then the deadline is tomorrow, the copy is approved upstream, and the calibration becomes a polish rather than a correction.

The workflow that actually catches errors has three stages, in order.

1. Brief the cultural register in. Before you ask the model for a Dashain caption, tell it which Dashain — the urban Kathmandu apartment Dashain, the village home-return Dashain, the diaspora Dashain in the Gulf. Tell it which generation the copy speaks to. Tell it the specific elements you want named — jamara, tika, the elders, the long road home — and the elements you want excluded — the festival of lights phrase, the generic Diwali imagery, the implication that this is mainly about shopping. Briefing the cultural register in is faster than briefing it back out.

2. Review for default-South-Asian filler. Read the draft with one specific question: which sentence could appear, word for word, in a Mumbai or Delhi campaign? Those sentences are the model reaching for the higher-frequency Indian neighbour. Either rewrite them with Nepali specifics or cut them. The test is fast and surprisingly catches the bulk of the cultural errors.

3. Get a Nepali-native reviewer to read it last. Not the marketing manager who studied in Australia. A reviewer who has lived a Dashain or a Tihar or a Bhai Tika in the last twelve months in Nepal, ideally from the community the campaign is aimed at. The reviewer is reading for one thing — whether the copy sounds like a Nepali person wrote it, or whether it sounds like a competent foreigner did. The reviewer is fast, cheap, and the single most reliable safeguard available.

When AI translates marketing copy, the leverage is in the glossary and the reviewer

A separate but related failure mode is translation. A marketer drafts a campaign in English, asks the model to translate it to Nepali, and ships the translation without a native review. The output is grammatical, the honorific register is wobbly, the brand-specific terminology drifts campaign to campaign, and the idiom is translated rather than rewritten. The audience reads it as foreign.

The highest-leverage habit when AI translates marketing copy into Nepali is not a better prompt. It is two things together: a personal terminology glossary that gets fed into every translation prompt — your brand name, your product names, the technical terms your category uses in Nepali, the honorific register you have decided to use — and a Nepali-native reviewer who reads every translated draft before it ships. The glossary keeps your campaigns consistent; the reviewer catches the idiom and tone errors that the glossary cannot. Either alone is not enough. Both together close roughly all of the gap.

Check your understanding

Quick check

—

A model produces a Dashain greeting that calls the festival ‘the festival of lights’ and pictures a family lighting oil lamps. What is the most accurate calibration?

Acceptable — Dashain and Tihar are often referred to interchangeably.
Wrong — that phrasing describes Tihar (Deepawali), not Dashain. Dashain is anchored by tika, jamara, elders’ blessings, and the homecoming around Vijayadashami.
Wrong only if the audience is in the Tarai; in Kathmandu the phrasing is fine.
Correct because Dashain falls in the same season as Diwali.

Quick check

—

A Kathmandu agency uses AI to translate every English campaign into Nepali. Which single habit raises quality the most across the next twelve months of work?

Switching to a larger model.
Pairing a personal terminology glossary fed into every prompt with a Nepali-native reviewer on every translated draft.
Asking the model to translate sentence by sentence rather than paragraph by paragraph.
Using only Devanagari with no English brand names mixed in.

What comes next

Once the cultural calibration is right, the next problem is the language itself — producing Nepali content that reads like a Nepali person wrote it rather than a translation of an English draft. The next section is about Devanagari content production: register, idiom, the honorific trapdoor, the mixed-script reality, and the glossary that keeps the whole thing consistent.