The short version
- Keyword research finds phrases with measurable volume. Topic research finds the problem the phrase is a symptom of.
- Keyword-led content targets demand that already exists. Topic-led content can address demand no tool measures.
- Assistant-led discovery favours topic research, because people ask full, contextual questions rather than typing fragments.
- The best input to both is free: your own Search Console query data and actual customer conversations.
Keyword research is the older and more codified discipline: find phrases, check volume and difficulty, pick targets. It works, and it has a blind spot — it can only show you demand that has already been expressed in a way a tool can count.
Topic research starts from the problem rather than the phrase. What is the person actually trying to do? What do they need to understand? What will they ask next? That produces content shaped around a question rather than around a string.
The short answer
Fit for your workflow
Use keywords for language and demand. Use topics to decide what the page should say.
Keyword data tells you which words your audience uses and roughly how many of them are searching — genuinely useful, and a targeting input rather than a brief. Topic research tells you what the page actually needs to contain to be the best answer. Leading with keywords alone produces pages that match a phrase and satisfy nobody.
- Established category with clear demand
- Keyword-led, topic-informed.
- New category nobody searches for yet
- Topic-led. Keyword tools will show nothing.
- Technical or expert audience
- Topic-led — this audience detects keyword-shaped content immediately.
- Local services
- Keyword-led. Intent is concrete and the phrases are stable.
What each one gives you
The data source row is the one that changes behaviour. Keyword research happens in a tool; topic research happens in conversations. Teams that do only the first end up writing about what is easy to measure rather than what their customers actually struggle with.
Why this shifted with AI assistants
People type fragments into a search box and full sentences into an assistant. "best crm small business" becomes "I run a 12-person plumbing company, we track jobs in a spreadsheet, what should we move to". The second is a topic with context, and the content that gets recommended is content that addresses that context specifically.
That shifts the balance toward topic research without eliminating keyword research, which still tells you the language people use and where demand concentrates. See ChatGPT vs Google for discovery.
Turning research into a brief
Both research modes produce inputs rather than instructions, and the step that most content processes skip is converting them into something a writer can act on. A usable brief states the question the page answers, the language the audience uses for it, what competing pages cover, and — the part that gets omitted — what this page will contribute that they do not.
That last line is what separates a brief from a keyword assignment. Without it, the writer produces a competent summary of what already ranks, which satisfies nobody’s definition of helpful content and is the first thing a generative summary replaces. Keyword research cannot supply it and neither can a tool; it comes from knowing something.
Where that contribution is genuinely absent, the honest response is to not write the page. A brief that cannot answer it is describing a summary of existing material, which will rank briefly, earn nothing, and be replaced by a generated answer. Spiking a piece at the brief stage costs an hour; discovering the same thing after publication costs a week and adds a page you will later have to prune.
The best free research source you already have
The queries report in Search Console shows what people actually searched before reaching you, including phrasings you never considered targeting. It is first-party, free, and specific to your audience rather than modelled from a general index. Most teams open it monthly and should open it before every brief.