Marketing Prompts: 2026 Templates

In short

A marketing prompt is a ready-made text instruction that gets an AI model to produce a content calendar, a post, a value proposition or a competitor breakdown without ten rounds of clarification. Below is a library of such templates, sitting right in code blocks: copy, swap in your own {variables}, get the result. They work in any text model — GigaChat, YandexGPT, ChatGPT, Claude. A strong prompt rests on five blocks: role, context, task, constraints, format. If you want a large bank of ready-made wordings, look at /arsenal/prompts/ and the collection of ready-made AI prompts.

What is a marketing prompt?

A prompt is a request to an AI model phrased so the answer can go straight into production. The difference between "write a post about the discount" and a working prompt is roughly the difference between a brief on a napkin and a proper spec.

What makes a marketing prompt distinct is that brand parameters are baked into it: audience, tone, product, target action. The more precisely you describe those parameters once, the less you rewrite afterwards. The good news: most marketing tasks repeat, so prompts can be turned into templates and reused for months.

The economics are simple. Briefing a copywriter, or free-form back-and-forth with a model, costs 15–20 minutes per piece of content. A finished template cuts that to a couple of minutes: you change the values in the braces and hit send. At a volume of dozens of posts, emails and offers a month, the difference adds up to whole working days.

What a strong prompt is made of

A working prompt is almost always assembled from five blocks. Let's go through each — missing blocks are the most common reason an answer comes back weak.

Role tells the model who to "act as": senior copywriter, performance marketer, analyst. The role narrows the style and vocabulary of the answer. Context describes the product, the audience and the channel — without it the model averages the text out "for everyone." Task is one concrete action in a single verb: "draft," "rewrite," "compare." Constraints set the boundaries: length, banned words, tone, what not to do. Format pins down the shape of the output — table, list or finished text — so you don't have to repackage the answer by hand.

We unpack the logic of these blocks in detail in the guide on how to write prompts — here we go straight to the templates.

Ready-made prompts: content calendar and posts

The content calendar is the starting task for any social media manager. This template produces a month's grid:

You are the content strategist for the brand {what the brand is}.
Audience: {who they are, their pains, their language}.
Channel: {Telegram / VK / Dzen}.
Draft a 4-week content calendar, 3 posts per week.
For each post: date, rubric, topic, one-line key point,
format (text/video/carousel), goal (reach/engagement/sales).
Keep the balance: 60% useful, 20% engagement, 20% selling.
Answer as a table.

Next, an individual post from a row of that calendar:

You are a copywriter for the brand {brand}, writing for {channel}.
Write a post on the topic: {topic from the content calendar}.
Audience: {who}. Tone: {friendly / expert}.
Structure: a hooking first paragraph, 2–3 paragraphs of substance,
one call to action.
Constraints: under {N} characters, no corporate-speak.
Add 3 alternative opening sentences to choose from.

Note two details. The 60/20/20 ratio in the first prompt sets the balance of rubrics so the feed doesn't turn into a wall of selling — change it to match your funnel. And asking for "3 alternative opening sentences" in the second prompt saves a whole iteration: you pick the strong hook immediately instead of sending the model round in circles.

On packaging content for specific channels, see our breakdown of AI tools for social media in 2026.

Ready-made prompts: value propositions, email and briefs

A value proposition is the short formula for why people buy from you specifically. Here's a prompt to test and reword it:

You are a marketing strategist.
Product: {what we sell}. Price: {₽}. Audience: {who}.
Competitors and their promises: {1–2 examples}.
Write 5 versions of the value proposition using the formula
"benefit + pain removed + how we differ."
Each one is a single sentence, without the words "quality,"
"reliable," "individual approach."
Mark which one is strongest and why.

An email or Telegram campaign:

You are an email marketer.
Segment: {who in the list}. Occasion: {sale / announcement / re-engagement}.
Write the email: subject line (under 40 characters), preheader,
body (3 paragraphs), one CTA button.
The goal is {click through to page X}. Use {the brand's} tone.
Give 3 subject lines with different triggers:
benefit, curiosity, deadline.

A brief is a structured assignment. A helper prompt will assemble one for you:

You are an account manager.
Put together a brief for {a landing page / an ad campaign}
for the product {name}.
Ask me 8 key questions whose answers the executor will need,
and explain what each one is for.

Banning the words "quality," "reliable" and "individual approach" isn't cosmetic: these clichés tell the buyer nothing and sound identical across every competitor. Strip them out and you force the model to hunt for a concrete benefit. By the same logic, in the brief prompt we ask not just for a list of questions but for an explanation of "what each one is for" — that way the assignment stays clear to the executor without you in the room.

A prompt for competitor analysis

Competitor analysis is a good thing to hand to an AI model — it structures what you collected by hand, fast. Important: the model does not replace research and can get facts wrong, so feed it your own data rather than asking it to "find competitors on the internet."

You are a competitive analyst.
Here is the data on 3 competitors that I collected:
{paste descriptions, prices, offers}.
Build a comparison table: product, price, value proposition,
channels, strength, weak spot.
Separately: 3 hypotheses on how we could differentiate.
Do not invent data that isn't in the input;
if something is missing, flag the gap.

The line "do not invent data" is critical here: it noticeably lowers the risk of hallucination. It also pays to explicitly ask the model to flag gaps — where the input has no price or no offer, an honest "no data" beats a plausible invention you'll later mistake for fact.

Common mistakes in prompts

Even with a finished template, the same handful of slips ruin the result. Keep them in mind while adapting a template to your own case:

How to adapt a template to your brand

Every template is built on {variables} in curly braces — placeholders you replace with your own values. The order goes like this:

1. Write your brand passport once: product, audience, tone, banned words, target action. Keep it as a separate document.
2. Paste that passport into the top of any prompt in place of the {variables} — the model answers in the brand's voice from the first try.
3. The first answer is almost never final: refine it ("shorter," "add a number," "cut the corporate-speak"). That's a normal part of the work, not a sign of a bad prompt.
4. Save the prompts that work into your own library. Note which model and which version of the wording gave the best result, so you don't lose the working variant before the next task. A ready-made base of wordings is in the qvib prompt arsenal.

Which AI model should you run these in?

The prompts above are model-agnostic: the text is the same everywhere. But availability from Russia differs. As of July 2026:

• GigaChat (from Sber) and YandexGPT operate in Russia officially, are tuned for Russian and have free tiers — handy for copy and email.
• ChatGPT, Claude and Gemini are foreign services; from Russia people usually reach them through aggregators that accept a rouble card, handle access on your behalf and return the finished answer.

The legal framing around foreign services keeps shifting, so rely on the official terms of the specific service at the moment you pay, not on rumors. And in any case use legal access routes, without circumventing blocks.

For picking a tool per task, see our pieces on free AI tools for writing and ChatGPT alternatives in Russia.

When prompts stop being big enough to hold your processes, the next step is building your own tools: landing pages, Telegram bots, automations. That's done with vibe coding in Claude Code or Cursor, and the Quest engine from qvib gives you a ready-made frame and roles for exactly this kind of work. Theory and practice are collected in the /learn/ hub.

FAQ

Can I use these prompts for free?

Yes. The templates themselves are free — copy and adapt them. Costs, if any, come from the AI side: GigaChat and YandexGPT have free tiers, foreign models have paid plans or access through aggregators (data as of July 2026).

Which model do these prompts work best in?

For Russian-language content and email, GigaChat and YandexGPT do well. For long, complex text, marketers more often reach for Claude or ChatGPT. The prompt doesn't need to change — it's universal.

How do I stop the model from inventing facts?

Ban it explicitly: add the line "do not invent data that isn't in the input." Give the model your own numbers and sources, and verify factual claims — prices and statistics above all — by hand.

Do I need to know how to code?

No. Prompts are ordinary plain text. Coding comes in later, if you want to turn chains of prompts into your own tools and automations — that's where vibe coding earns its keep.