Content idea generators, the four kinds of source behind them and which one can see real demand

Every content idea generator belongs to one of four families, and the family decides whether an idea has demand behind it or just plausibility. A tool reading live query data can tell you what people typed. A tool reading competitors can tell you what already ranks. A tool reading a template library can tell you a shape. A tool generating from a language model can tell you something fluent that nobody searched for. Pick by which of those four answers your current problem, and the choice gets simple.

Google Keyword Planner shows 170 searches a month in the United States for this group of phrasings, read on 31 October 2026 against the United States geo target, counted as the group maximum rather than the sum of the phrasings.

Family one, query data generators

These take a seed word and return the questions and phrases people actually typed, grouped into patterns. The output is unglamorous and it is the only output with demand attached, which makes it the right starting point when you need the next twenty articles to be worth publishing.

What they are good at is finding the question behind a category: the comparisons, the price questions, the how-long questions, the is-it-worth-it questions. What they cannot do is tell you whether the question is winnable, because volume says nothing about who already answers it well. Our note on turning a question export into an actual plan covers the step after the export, which is where most of these tools leave you.

Family two, competitor readers

These point at a domain and return what it ranks for, what it publishes most and where its traffic appears to come from. The idea arrives as a gap: a subject a rival covers and you do not.

Two cautions. The traffic numbers are modelled estimates rather than measurements, so treat the ranking list as reliable and the traffic attached to it as directional. And a gap is not automatically an opportunity, because the rival may be ranking on the strength of a domain you are not going to match this year. Our comparison of the tools that do keyword research and competitor analysis together, and where each number comes from is the right read before buying one of these on a figure you cannot source.

Family three, template and framework generators

A library of headline shapes, article structures and prompt patterns. These are cheap, often free, and they solve a real but narrow problem: you know the subject and cannot see the angle. Used that way they earn their place in ten minutes.

Used as the primary source they produce a blog that reads as assembled, because the shape arrives before the thought. The tell is a publication where every piece is a numbered list of the same length.

Family four, model generators

Ask an assistant for thirty content ideas and you get thirty fluent, coherent, plausible ideas. The strength is breadth and speed. The limitation is specific and important: the output reflects what is common in text about your category, so it converges on what everybody in your category already writes. Hand the same prompt to three competitors and all three get broadly the same list.

There is one job this family does better than any of the others, though, and it is underused. Ask it what somebody in your buyer's position would need to know before committing, in their own words, and you get a decision tree rather than a headline list. That maps to pages far better than a list of titles does.

The combination that actually works

Use them in sequence rather than choosing one. Start with query data to get the real questions, since those have demand attached. Run the competitor reader over the two or three rivals who rank for those questions, to see which ones are genuinely contested. Then use a model generator on the remaining list to find the angle nobody has taken, and use a template library only at the point where you are writing the headline.

That order puts evidence first and invention last, which is the opposite of how most teams use these tools and it is why most idea backlogs are full of ideas nobody searches for.

What to require before paying for one

• Where the query data comes from. Clickstream, a search engine's own figures, or an inferred model. All three are legitimate and they disagree, and a vendor that will not say is a vendor whose numbers you cannot defend internally.
• Whether the free tier is enough. For a site publishing four pieces a month it frequently is, and the paid tier is buying volume rather than better ideas.
• Export. If the output cannot leave as a spreadsheet, the tool owns your plan.
• Whether it checks what you already published. Almost none of them do, which is the single biggest source of wasted output: a generator will happily hand you a subject you covered eighteen months ago.

That last point is worth a manual step whatever you buy. Before an idea enters the backlog, search your own site for it. Two of your own pages answering one question split the result between them and neither wins it.

A test that takes an afternoon

Pick one subject you know well. Run it through one tool from each of the four families. Put the four outputs side by side and mark each idea with one of three labels: somebody searches for this, somebody would pay attention to this, or neither. Count the labels per family. You will find the query tools win on the first label, the model tools win on the second, and the template tools produce mostly the third. That tells you which one to put your subscription money behind for your particular publication, on your own evidence rather than on a review.

Questions about content idea generators

Is a free content idea generator good enough

For finding real questions, often yes, because the free tiers of the query tools return the patterns and limit the volume. Pay when you are planning dozens of pieces a month and the export limits start costing you more time than money.

Why do two generators give me completely different ideas

Because they are reading different sources. One is reading what people typed, another is reading what a rival ranks for, a third is generating from a model. Different inputs, different outputs, and none of them is wrong on its own terms.

Can I just ask an assistant for ideas instead

You can, and you should, for angles rather than for demand. An assistant has no access to search volume, so the ideas arrive without evidence that anybody is looking for them. Combine it with a query source rather than replacing one.

How many ideas do I actually need

Fewer than any generator gives you. A publication doing four pieces a month needs twelve good subjects a quarter, and a list of two hundred ideas is mostly a way of avoiding the decision about which twelve.

How do I stop the backlog filling with subjects we already covered

Make a site search a required step on every idea before it is approved, and record the covering page when you reject one. A rejected idea with a reason attached never costs a second review.

The question list nobody generates from

There is a source of real buyer questions that none of these four families reads: the questions people put to an assistant when they are deciding who to buy from, and the shortlist of businesses that comes back. Those questions are more commercial than anything in a keyword export, because somebody asking which supplier to use is further along than somebody asking what a term means. AI Knows Us works from exactly that set, measuring which answers name your business and which pages they quote, and the gaps it finds are content ideas with a buyer already attached to them.