Briefing AI on a campaign — fact-first

The campaign brief skeleton

Every campaign brief that produces usable AI output has the same skeleton. You can write it on the back of a notebook in seven minutes. Most marketing teams that complain the model “doesn’t get Nepal” are giving the model two sentences and a vague mood, then judging the output as if they had given it the brief in their head.

1. Brand and product. Not the brand alone. The brand and the specific product or offering this campaign exists to move. “Khalti” is not the brand for a brief; “Khalti, with a focus on the Bhai Tika gift-sending feature this Tihar” is. The model needs to know what action the audience is being nudged toward.

2. Audience, concrete. Not “Nepali consumers.” Not “young people.” A specific audience the marketer can picture: “urban Kathmandu young families on Facebook, mid-twenties to mid-thirties, smartphone-primary, comfortable with digital payments, sending money to siblings working abroad.” The more concrete the audience, the less generic the output. Vague audience descriptions are the single most common reason AI campaign output reads like stock material.

3. The specific moment. Bhai Tika. Mha Puja. Day three of Tihar. The year-end appeal week. Mid-Saun rainy slump. A moment carries cultural and emotional weight the model cannot infer from “festive season.” Naming the specific day or week tells the model which references are in play — jamara, the doko, the brother-sister relationship, the late-night Deusi-Bhailo — and which are not. A campaign briefed for “the festive season” gets you a confused Diwali-Dashain-Tihar mashup. A campaign briefed for the morning of Bhai Tika gets you tika, oil, marigold, and a sister waiting.

4. The emotional beat. This is the single most under-specified element in most briefs and the single biggest lever on output quality. One sentence: what feeling do we want the viewer to recognise, and what feeling do we want them to leave with? For a Bhai Tika campaign aimed at remittance-earning families, the beat might be: “recognition of distance — the brother who could not come home this year — and the comfort of being remembered, which the product makes possible.” That one sentence tells the model which emotional grammar to write inside.

5. Constraints, named. Must work as a 30-second video. Must end with a clear product appearance in the final three seconds. Must run on Facebook and Instagram Reels. Must be safe for a family audience watching together. No on-screen Devanagari longer than four words because text inside video gets garbled on small screens. Each constraint cuts out work the model would otherwise waste effort on.

6. Tone — three voice adjectives plus two contrast adjectives. “Warm, intimate, grounded — not sentimental, not commercial.” Five words. The contrast adjectives do as much work as the positive ones, because they tell the model where the cliff edge is.

7. What to avoid. The cliché list. “No mountain B-roll. No generic dhaka topi shorthand. No Dashain-Tihar conflation. No tear-on-cheek close-up. No sales voiceover at the end.” The avoid list is where the marketer’s experience of past campaigns becomes visible to the model.

A worked example — Khalti, Bhai Tika

Here is what a complete, fact-first brief looks like, taking less than a page, for a Khalti Bhai Tika campaign.

Brand: Khalti. Product focus: the Bhai Tika gift-sending feature — sisters send tika gift money to brothers in-app with a digital envelope. Audience: urban Kathmandu and Pokhara women aged 24-38 on Facebook and Instagram, smartphone-primary, comfortable with mobile wallets, many with brothers either out of the country or out of the city for work. The specific moment: the morning of Bhai Tika, day five of Tihar, the year the brother could not come home. Emotional beat: recognition of distance + comfort of being remembered — the sister wants to perform tika even when the brother is absent, and the product makes presence possible. Constraint: 30-second vertical video for Reels, ending with a clear in-app product moment in the final three seconds; safe for family viewing. Tone: warm, intimate, grounded — not sentimental, not commercial. Avoid: mountain B-roll, dhaka topi shorthand, Dashain-Tihar conflation, tear-on-cheek close-up, sales voiceover.

Now give that brief to the model and ask for three 30-second script options. What you get back is no longer generic. The first draft might open with a sister setting out the puja plate at a Kathmandu apartment window, marigold garland in soft morning light, phone face-up beside the plate. The brother appears on a video call from Doha. The sister places tika on her own forehead, then holds the camera close so the brother sees her smile; she taps the in-app envelope and sends the gift in a single motion. The model has produced a draft that respects every element of the brief — and the marketer’s job is now the work the marketer was trained for: selecting between three drafts, sharpening the close, replacing the line that does not land, deciding whether the apartment is the right setting versus a Tarai family courtyard.

The editing layer is real. The brief gets you to a usable draft; the marketer gets you to a shippable campaign. The model never writes the line that survives the boardroom because the line that survives the boardroom carries judgement about the brand, the moment, and the audience that the model cannot hold.

Strategy belongs to you. Production can move

This is the cleanest mental split for working with AI on campaigns. Strategy is the brief. Production is the variation against the brief. The marketer owns the strategy. The model produces variations. The marketer selects, edits, and ships.

The strategy work — deciding which festival moment to lean into, which emotional beat sits with the brand, which audience to chase this quarter, which product story is overdue — does not move. The model has no view on any of it. It does not know that your brand spent last Dashain getting criticised for a tone-deaf urban-coded ad and needs to come back warmer and more grounded this Tihar. It does not know that your competitor just bought the rural-family-reunion angle and you should not chase them into it. These are decisions the human marketer makes from inside the work.

The production work — turning the brief into three script options, four caption variants, six headline tests, a Viber-broadcast cut, a YouTube pre-roll line — is exactly where the model earns its keep. Brief in, variations out. The marketer reads the variations and chooses. The model has not done your strategy; it has compressed a week of typing into an afternoon.

Multi-asset campaigns and why brief quality compounds

Modern campaigns are not single artefacts. A Tihar campaign in 2026 ships as a 30-second video, three Reels cuts, four Facebook posts, two Instagram carousels, a Viber broadcast, an email, three WhatsApp-Business message templates, and a YouTube pre-roll. Producing all of these consistently from a single brief is exactly the work AI is genuinely good at — once the brief is good.

The compounding works like this. A two-sentence brief produces ten weak assets that do not feel like the same campaign. A serious one-page brief, with the seven elements above, produces ten assets that share an emotional beat, a tone, an audience picture, and a product moment. The campaign reads as a campaign because the model is generating every cut from the same anchor. Brief quality compounds across assets; brief weakness compounds across assets in the same way, just less visibly until you ship.

This is why the time spent on the brief is not overhead — it is leverage. Twenty minutes on the brief replaces hours of inconsistency-fixing across channel cuts. It also replaces the small editorial losses that happen when the marketer has to invent the emotional beat freshly for each asset because the original brief did not name it.

Check your understanding

Quick check

—

A marketing lead at a Nepali fintech wants to use AI to produce a Tihar campaign. Which of the following best describes the brief most likely to produce usable output?

1. AA two-sentence note saying “heartwarming Tihar campaign, family theme, end with logo,” pasted into the model.
2. BA one-page brief naming the specific product, a concrete audience picture, the specific Tihar day, the emotional beat in one sentence, format constraints, three voice adjectives plus two contrast adjectives, and an explicit avoid list.
3. CA long mood document with twenty reference images and no audience or product information.
4. DA prompt that asks the model to “think like a senior Nepali creative director and surprise us.”

Check answer→

Quick check

—

A founder tells the team: “Let’s have AI write the whole Dashain campaign in an hour — strategy, creative, channel cuts, the lot.” What is the strongest professional pushback?

1. AThe model is too slow to do all of that in an hour.
2. BAI tools are not allowed to produce campaign work in Nepal.
3. CThe model can produce the production layer fast — variations, channel cuts, captions — but it cannot do the strategy: deciding which moment to lean into, which emotional beat the brand should carry this year, which competitor angle to avoid. Strategy is the brief. Without that brief from the marketer, the model produces fluent off-strategy output.
4. DDashain campaigns specifically require a six-week lead time by industry convention.

Check answer→

What comes next

A good brief is the foundation, but a campaign also needs visuals — and visuals are where current image models meet their hardest test in the Nepali context. The next section is an honest tour of what image generation can and cannot do for Nepali audiences in 2026, where the default-Indian-aesthetic problem comes from, and the prompt patterns that get you closer to images your audience reads as theirs.