AI for writing texts: choosing a model, writing the prompt, pricing it

AI for writing texts saves time on the first draft, but it doesn't take responsibility for the result off the author. Below: how to write through the Zukko chat and the API, which model to pick by price and intelligence index, how to write the prompt, how to organize editing, what one article costs, and what you must check before publishing.

AI for writing texts: chat or API

Chat opens in the dashboard: zukko.pro/app/chat↗. You pick the model from a list, each answer is charged to your shared dollar balance, and no API key is needed. Chat is convenient when you write a single text and want to refine wording as you go, or when you're tuning a prompt.

API is for many or repetitive texts: product cards, descriptions, newsletters, translations. The same prompt is applied to a hundred inputs in a loop, and the result goes straight into your system.

Token prices are the same either way.

How to choose a model

Prices as of October 7, 2026, USD per 1M tokens. The intelligence index is the Artificial Analysis Intelligence Index shown in the Zukko catalog.

For the current list and prices, call GET https://api.zukko.pro/v1/models or open the Models page in the dashboard.

The index is a composite benchmark score, not a rating of writing quality. It helps shortlist candidates (the full table is in the AI model ranking for October 2026↗), but it doesn't tell you which model will write your particular text better. So choose by testing:

1. Take one real brief.
2. Send it unchanged to 3 models at different levels, for example claude-opus-5-5, gpt-6.1-sol and gemini-3.8-flash.
3. Remove the model names and ask a colleague to rate the texts blind against your criteria: accuracy, style, amount of editing needed.
4. Pick the cheapest model whose text needs an acceptable amount of editing.

In chat you can do this by hand, switching the model in the list.

A prompt for writing: what it should contain

A model writes better when it doesn't have to guess. A good brief answers seven questions:

• Who it's for — the audience and their level.
• Why — what the reader should understand or do.
• Facts — figures, names, terms that must be used. The model shouldn't invent them.
• Structure — sections, order, whether a table or list is needed.
• Style — tone, formal or informal address, sentence length, banned words.
• Length — approximate word or character count.
• What not to do — for example, don't add figures that aren't in the brief.

Template:

Write a blog article for an online store.
Audience: shoppers choosing their first robot vacuum.
Goal: help them pick a model by apartment size and floor type.
Facts (use only these): <list of specs and prices>.
Structure: intro, 3 sections with subheadings, comparison table, conclusion.
Style: polite, no bureaucratic language, no exclamation marks.
Length: 800–1000 words.
Don't invent specs, reviews or figures that aren't in the facts.

Draft, review and final version via API

A handy scheme is three requests: a draft, editor's notes, a revised version. A Python example with the official OpenAI SDK:

import os
from openai import OpenAI

client = OpenAI(base_url="https://api.zukko.pro/v1", api_key=os.environ["OPENAI_API_KEY"])
MODEL = "claude-sonnet-5"


def ask(system: str, user: str, max_tokens: int = 6000) -> str:
    resp = client.chat.completions.create(
        model=MODEL,
        messages=[
            {"role": "system", "content": system},
            {"role": "user", "content": user},
        ],

The fetch_... key lives in the OPENAI_API_KEY environment variable. You can run the editor on a different model: turn MODEL into a function parameter. Endpoint reference in the chat/completions docs↗.

What one article costs

Assumptions for an 800–1000-word article (you can get exact volumes for your own text via token counting↗):

• draft: 3,000-token brief in, 4,000 out;
• notes: 4,500 in (brief and text), 1,000 out;
• final: 5,500 in (text and notes), 4,000 out.

Total: 13,000 input and 9,000 output tokens. Formula: 13,000 × input price / 1,000,000 + 9,000 × output price / 1,000,000.

Checking the Sonnet 5 row: 13,000 × 0.4 / 1,000,000 = $0.0052, 9,000 × 2 / 1,000,000 = $0.018, total $0.0232. Prices as of October 7, 2026; totals are an example calculation. Most of the cost is output, so long answers and extra iterations affect the price more than the length of the brief.

Editing: what the human does

A model is good at getting past the blank page, but a person must read the final version. What to check:

• Logic. Each section answers its own question; no idea is repeated in different words.
• Filler. Generic intros, "it's worth noting", lists for the sake of lists — cut them.
• Voice. The text should sound like your brand, not an averaged template. One or two of your past texts in the brief help.
• Specifics. Where the model generalized, add an example from your own practice.

Ethics and fact-checking

Models can state falsehoods confidently: invent figures, quotes, study names and links. Before publishing:

1. Check every figure, date and name against the primary source. If there's no source, remove the claim.
2. Check the links — that they open and say what the text claims.
3. Don't pass off someone else's work as yours. If the text retells a specific source, cite it.
4. Follow your platform's rules on labeling AI-assisted content, if it has any.
5. Don't put unnecessary personal data into the prompt — customers, employees, correspondence — when you can do without it.

The most reliable way to reduce errors is to give the model the facts in the brief and explicitly forbid adding its own.

To try it, open the chat in the dashboard or create a key for the API — both use the same balance. More on bulk generation in the article SEO content generation via API↗.

FAQ

Which AI writes best in Russian?

We have no general data on writing quality in Russian per model. The intelligence index is a composite benchmark score, not a style rating. Compare 2–3 models blind on your own brief and pick the cheapest one that meets your quality bar.

How much does it cost to write an article with AI?

By our calculation, an 800–1000-word article in three steps costs from $0.0005 on DeepSeek V4.1 Flash to $0.0464 on Claude Opus 5.5, at prices as of October 7, 2026. Your numbers depend on brief length, text length and the number of iterations.

Can I publish the text without editing?

Better not. The model can get facts wrong and write formulaically. At minimum, check all figures, names and links and read the whole text.

Do I need an API key to write in the chat?

No. The chat runs in the dashboard and charges answers to your balance. A key is only needed for the API.