How to Find Blog Topics with AI Keyword Research Prompts

This page may contain affiliate links.

Every blogger hits the same wall eventually. You want to write, you have the energy, but you have absolutely no idea what to write about that anyone will actually read. Traditional keyword research tools can help, but they often cost a monthly subscription and bury you in spreadsheets of terms nobody searches for. AI changes this. Used properly, a decent large language model becomes a research assistant that never gets bored, never runs out of angles, and costs you nothing beyond the subscription you probably already pay for.

The trick is not asking "give me blog ideas." That gets you generic slop. The trick is structuring prompts that force the AI to think like a search engine, a competitor analyst, and your ideal reader all at once. Here is how to do it.

Start with audience and intent, not keywords

The biggest mistake people make is opening ChatGPT and typing "blog topics about gardening." You will get twenty obvious ideas you could have thought of yourself. Instead, give the AI context about who you are writing for and what stage of the buying or learning journey they are at.

Try a prompt like this:

"I run a UK personal finance blog aimed at people in their late twenties who have just started their first proper job. Generate 20 blog post topics that match search intent for people who are confused about workplace pensions but haven't yet opened a SIPP. For each topic, give me the likely search phrase they would type into Google and a one-line explanation of why they would search it."

Notice the difference. You have told the AI the audience, the geography, the emotional state (confused), and the specific gap in their knowledge. The results will be far more usable. You can then run the same prompt swapping "confused about pensions" for "ready to open a SIPP" and get a completely different, equally valuable set of topics.

Use AI to reverse-engineer what already ranks

AI cannot browse the live web in the way a dedicated SEO tool can, but it has absorbed an enormous amount of content and can reason about what tends to rank for given queries. A useful technique is to ask it to identify the types of content that dominate a topic.

Prompt example:

"For the search term 'best budget air fryer UK', what types of blog posts typically rank on page one? List the common formats (listicle, buying guide, comparison, recipe roundup) and for each, suggest three specific angles a smaller blog could take that the big sites tend to ignore."

This is where AI earns its keep. Big publishers chase broad head terms. You want the long-tail, the niche angle, the "air fryer recipes for one person in a small flat" type of post. Ask the AI explicitly to find the gaps.

You can push further with a follow-up:

"Now give me 15 long-tail keyword variations for that topic that a new blog could realistically rank for within six months. Include the search intent behind each one."

Mine your own niche for question-based topics

Question keywords are gold for new blogs because they are specific, often low-competition, and map directly to what people type into Google or ask voice assistants. AI is excellent at generating these at scale.

Try:

"Act as a UK-based reader who is considering buying a 3D printer for a small business. List 25 questions I might type into Google before buying, ranging from total beginner to fairly technical. Group them by theme."

Once you have the list, ask the AI to turn each question into a working blog title and suggest a rough word count. You now have a content calendar for the next two months. If you want to go deeper on any physical product angle, you can check what is actually available and well-reviewed on Amazon UK — for example, 3D printers for beginners — so your recommendations stay grounded in reality.

Validate and prioritise before you write

AI-generated ideas are a starting point, not gospel. Before you commit to a topic, do a quick sanity check. Ask the AI:

"For each of these 10 topics, estimate the likely search volume (high, medium, low), the competition level (high, medium, low), and whether it is better suited to a brand new blog or an established one. Be honest about which ones are probably too competitive."

This will not give you precise numbers, and you should treat any specific figure with suspicion. But it will help you sort the wheat from the chaff. Combine this with a free tool like Google Trends or the autocomplete suggestions in Google Search itself to confirm there is genuine interest.

A practical rule: if you cannot find at least a handful of real people asking the question in forums, Reddit threads, or Google's "People also ask" box, the AI may have invented a demand that does not exist.

If you would rather skip the trial and error of building these prompts yourself, the Content & Blogging Pack is a set of 30 proven prompts covering blog writing, SEO, newsletters and social media, starting from £9. It is essentially the done-for-you version of the workflow in this article, ready to paste into your AI tool of choice.

Turn one good topic into ten

Once you find a topic that works, do not stop there. Ask the AI to expand it:

"I have written a post called 'How to set up a UK side hustle while working full time.' Give me 10 follow-up posts that would naturally link to it, each targeting a different long-tail keyword, and explain the internal linking angle for each."

This is how you build topical authority. Search engines reward sites that cover a subject thoroughly. One strong pillar post plus ten supporting articles beats twenty random posts every time.

You can also ask the AI to suggest a content cluster diagram in plain text — a simple list of which posts link to which — so you can see the structure before you write a word.

Conclusion

AI will not magically tell you what to write. It is not a crystal ball and it does not know your audience better than you do. But it is a genuinely useful thinking partner for keyword research if you give it the right constraints: a clear audience, a specific intent, a geographic focus, and a demand for long-tail angles rather than broad head terms. Start with the audience-and-intent prompt, reverse-engineer what already ranks, mine question keywords, validate before you commit, and then expand your winners into clusters. Do that consistently and you will never stare at a blank page again.