Keyword Research for SEO: A 7-Step Process

Keyword research for SEO is the process of finding the words and questions your customers use in search, judging what each searcher wants, and deciding which page on your site should answer each one. The output is not a spreadsheet of high-volume terms. It is a map from real queries to pages, with a reason for every row.

This playbook is for owners, marketers and writers who plan content. It walks through seven steps with free and paid sources, then adds the step most guides skip: prompt research for AI search, where most of the questions that matter show zero volume in any tool.

On this page

• What is keyword research for SEO?
• How can I research keywords for SEO?
• Free and paid keyword research sources
• Reading search intent from the results
• Keyword difficulty and what it really measures
• Clustering keywords into pages
• Prompt research for AI search

What is keyword research for SEO?

Keyword research answers three questions for each topic: what do people type, what do they want when they type it, and can your site be the best answer. Search volume is only one input, and often the least reliable one, because most searches are rare. Ahrefs reports in its long-tail keyword analysis that keywords with fewer than 10 searches a month account for almost 93% of its US keyword database.

New phrasings keep appearing, too. At Search Central Live NYC in 2025, Google’s John Mueller said around 15% of the queries Google sees each day are new, a share he said has held steady even with AI search.

That changes the goal. Instead of hunting a few big terms, good research finds the topics your buyers care about and the many phrasings around them. If you need the basics of the channel first, start with what does SEO stand for.

How can I research keywords for SEO?

Work through seven steps in order. The diagram shows each step and what it produces; the sections below cover the ones that need judgment.

1. Seeds. Write down every product, service, problem and place you serve, in your customers’ words rather than your internal names. Sales emails and support tickets are the best source.
2. Expand. Run each seed through tools and the results page itself (People Also Ask, related searches), and pull the queries you already appear for from Search Console.
3. Intent. Search each candidate and note what kind of page ranks. This decides the page type.
4. Difficulty. Look at who ranks. A results page full of national brands is a different job from one full of forums.
5. Cluster. Group terms that return the same results into one page.
6. Prompts. Ask the same questions in ChatGPT, Gemini, Perplexity and Google AI Mode, and note what they answer and whom they cite.
7. Prioritize. Rank pages by business value first, then by how quickly you could compete.

Free and paid keyword research sources

You can do solid research for free; paid tools save time and add competitor data. The sources worth knowing:

Two notes on the free sources. Google’s Keyword Planner help describes its figures as estimates of monthly searches, grouped across close variants, and you need to complete account setup with billing information to use “Get ideas for new keywords”. Search Console’s Performance report gives clicks, impressions, click-through rate and average position for real queries, which makes it the most trustworthy source you have. Our Search Console guide shows how to mine it.

Reading search intent from the results

The results page tells you what Google thinks the searcher wants, which is more reliable than guessing from the words. Search the term and classify what ranks:

• Informational: guides, definitions, how-tos. “What is a heat pump”.
• Commercial investigation: comparisons, reviews, “best” lists. “Best heat pump for cold climates”.
• Transactional: product, pricing and booking pages. “Heat pump installation quote”.
• Navigational: one brand’s site. “Carrier heat pump login”.
• Local: a map pack and Business Profiles. “Heat pump installer near me”.

If the top results are all list articles, a product page will struggle, however well optimized. Mixed results mean mixed intent, and a page that serves two intents in clearly separated sections can work. Our search intent guide goes deeper.

Keyword difficulty and what it really measures

Keyword difficulty is a third-party estimate, and it measures less than the name suggests. Most tools calculate it mainly from the links pointing at the pages that already rank. That makes it a decent proxy for how much reputation you need, and a poor proxy for everything else. Age matters too: Ahrefs found that 72.9% of pages in Google’s top 10 are more than 3 years old, so a low score doesn’t mean a quick win.

Use it as a rough filter, then look at the actual results:

• Who ranks? Forums, thin pages and outdated posts in the top five are an opening, whatever the score says.
• Do they match intent well? A high score with weak intent match is often beatable with a better page.
• Is there a SERP feature taking the clicks? An AI Overview or a map pack can absorb most of the attention. In Pew Research Center’s March 2025 study, users who saw an AI summary clicked a traditional result in 8% of visits, against 15% without one.
• Do you have topical standing? A site that already covers the surrounding topic tends to rank new pages on it faster. That is the argument for topical authority.

Clustering keywords into pages

Clustering turns a long list into a page plan: terms that return substantially the same results belong on one page, and terms with different results need their own. Doing this well prevents two of your pages competing for one query. The diagram shows a cluster for this very topic.

A quick manual test: search two terms and compare the top ten. If most URLs overlap, they share a page. Paid tools automate the comparison. Link the pillar and spokes to each other, which is the pattern covered in topic clusters for AI citation and internal linking.

Prompt research for AI search

Prompt research is keyword research for questions that tools cannot measure. People ask AI assistants long, specific questions, and the systems then run several searches of their own. Google’s AI features documentation calls this query fan-out: issuing multiple related searches across subtopics to build a response. Those follow-up searches rarely match anything in a keyword database. The chart puts numbers on the gap.

Zero volume does not mean zero demand. Ahrefs, citing Seer Interactive, notes that over 95% of conversational long-tail keywords have no measurable search volume. It means the demand is spread across phrasings no tool can count. A practical prompt research routine:

1. Write 20 to 50 buyer questions in full sentences, the way a customer would ask an assistant: “Which accounting software works for a three-person firm that bills hourly?”
2. Run them in several assistants and record which brands are named and which pages are cited.
3. Break each answer into its sub-questions. Pricing, integrations, setup time. Those are the fan-out topics.
4. Map each sub-question to a page section that answers it in its first sentence.
5. Repeat monthly, because answers change as sources change.

This is where keyword research meets content chunking for AI: once you know the sub-questions, each needs a section that can be quoted on its own. Tools for tracking the results are compared in AI citation tracking tools.

What are the types of keywords used in SEO?

Keyword types are labels for intent and specificity, useful for sorting a list:

• Short-tail (head) terms: broad, high volume, mixed intent, and rare: Ahrefs counts just under 18,000 US keywords with more than 100k searches a month. “Accounting software”.
• Long-tail terms: specific, low volume, clearer intent. Ahrefs defines long-tail by volume rather than length: a keyword is long-tail because few people search it, not because it has many words.
• Branded vs non-branded: includes a brand name or not.
• Local: includes a place, or implies one (“near me”).
• Question keywords: who, what, how, why. These map directly to headings and FAQ entries.

Common keyword research mistakes

Most failed keyword plans fail in one of five predictable ways:

• Chasing volume over fit. A 50,000-search term your product only half answers is worth less than a 200-search term that describes exactly what you sell. Volume without intent match brings visitors who leave.
• One page per phrasing. Writing separate pages for “keyword research tips” and “how to do keyword research” splits one intent across two URLs. Google’s ranking systems include deduplication, which shows only the most relevant of similar pages, so one of them usually disappears anyway.
• Stuffing the chosen term. Once a keyword is picked, some writers repeat it in every paragraph. Google’s spam policies name keyword stuffing as spam, and it reads badly long before it trips a filter.
• Ignoring your own data. Search Console already lists queries you rank on page two for. Those are cheaper wins than brand-new terms.
• Researching once. Language shifts, products change, and AI assistants change how people phrase questions. A plan from two years ago describes a market that has moved.

The fix for all five is the same habit: tie every keyword to a page and a reason, and revisit the list when your data changes.

Turning keyword research into an SEO plan

The end product of keyword research is a page plan: one row per page, with its primary query, the variants it absorbs, its intent, the questions it must answer, and whether it exists yet. Build the money pages first, then the questions buyers ask before they buy. Check the plan against real data every quarter, because Search Console will show you queries you never thought to research. For how each page should then be built, use the on-page SEO checklist.

If you want the research done with AI prompts mapped alongside classic queries, our AI SEO service starts from exactly this kind of plan.