Prompts for finding blog topics that match local demand
The problem is not that an AI cannot produce topics. It is that it produces twenty identical generic ones. Here are prompts that yield specifics.
Prompt one: from customer questions
"Here are twenty questions customers ask us by phone and on Telegram: [list]. For each, propose an article headline that answers it completely. The headline should be phrased the way someone would search for it, without marketing language. Mark which questions can be merged into one article and which need their own."
This is the most effective prompt of the set because you are feeding the model real demand. Collect two weeks of questions from your messages: half an hour of work and the best topic source you have.
Note the instruction to flag mergeable questions. Without it you get five articles competing with each other for the same query.
Prompt two: from objections
"Our service is [description]. List fifteen objections and doubts a customer in Uzbekistan has before buying. For each, propose an article topic that resolves it through explanation rather than persuasion. Avoid topics of the 'why choose us' kind."
Objection-driven articles convert better than overviews: the reader arrives with intent and only needs the last doubt cleared. These are usually topics about price, timelines, guarantees, and what happens if it does not work out.
Prompt three: localising a generic topic
"Topic: [generic topic]. Rewrite it as five specific articles for an audience in Uzbekistan. Account for the local payment methods Payme and Click, Telegram's dominance as a communication channel, bilingual websites, and pricing in UZS. Do not use examples or realities from other countries."
This prompt turns 'how to improve site conversion' into 'why a seven-field form fails when your customer would rather message you on Telegram'. Local grounding is exactly what competitors' translated articles lack.
Check the output: the model happily imports Russian market realities. Delete anything referencing services or practices that do not exist here.
Prompt four: analysing the existing results
"Here are the titles of the top ten search results for [query]: [list]. Identify which question each one answers and what is missing from this set of results. Propose three topics that fill the gap rather than repeating what exists."
The last sentence carries the value. Without it the model proposes an eleventh copy of the same material. Gaps usually sit in specifics: numbers, step-by-step detail, local context, edge cases.
Prompt five: a series instead of scattered topics
"Arrange these topics into a logical series of six articles, from basic to advanced. For each, state which earlier article it builds on and which internal link belongs inside it. Topics: [list]."
A series works better than a pile of unrelated posts: it creates natural internal linking and keeps readers moving. Sequencing is a structural task, which is where these models are genuinely strong.
Validating demand before writing
Never start writing without validating. Run the phrasing through Yandex Wordstat filtered to the Uzbekistan region and Google autocomplete in both languages. Zero volume in every language means the topic interests you, not the market.
Check the Uzbek phrasing separately: it may have demand where the Russian one does not, and the reverse. This is the most underrated way to find low-competition topics.
Keep a master topic sheet: topic, language, volume, status, publication date. Without it, idea generation becomes an endless process that never reaches actual writing.