How to do keyword research, step by step

How to do keyword research comes down to six steps. Start with the words your customers already use, expand them into a longer list, check each keyword's search volume and difficulty, judge what the searcher wants, group keywords that one page can answer, and pick one target per page. Then add the questions people ask AI assistants, because those shape the sections a page needs.

Most keyword research goes wrong at the ends. People start from jargon instead of customer language and finish with 400 keywords and no decision about which page targets which.

Start keyword research with your customers' words

Your best seed keywords are in your inbox, not in a tool. Read sales calls, support tickets and reviews, and write down the exact phrases. A clinic might call its service "orthodontic aligner therapy" while patients type "invisible braces cost".

Good places to collect seeds:

• The Queries tab of the Performance report in Google Search Console, which shows what people already search to find you.
• Competitor pages that rank for things you sell. Our competitor keywords tool lists the searches a competitor's domain ranks for.

Aim for 5 to 15 seeds. Then expand each one. Google's autocomplete and "related searches" show how real people extend a phrase, and Keyword Ideas pulls those autocomplete suggestions for a seed in your market. Expansion surfaces long-tail keywords, the specific phrases with fewer searches and clearer intent.

Check search volume and keyword difficulty

Volume tells you how many people want a thing, difficulty how hard page one is to reach. Read both as estimates.

Every tool, including Google's Keyword Planner, reports rounded estimates of monthly searches, not exact counts. Use volume to compare keywords with each other, not to forecast traffic. Keyword search volume covers why the numbers differ between tools.

Difficulty scores differ by vendor too, so judge them against your own site. A newer site with few links should start with low-difficulty keywords and work up. Keyword difficulty explains how to read the score, and low-competition keywords shows how to find the ones you can win now.

Cut anything with no searches unless customers say it constantly.

Judge search intent and group keywords

Search intent decides what kind of page can rank. Search the keyword and look at page one. If Google shows guides, you need a guide. If it shows product pages, a blog post will not rank no matter how good it is.

The four usual labels are informational, commercial, transactional and navigational. "How to clean a standing desk" wants a guide. "Best standing desk for tall people" wants a comparison. "Buy standing desk 72 inch" wants a product page. Search intent goes through each type with examples.

Then group keywords that one page should answer together. If two phrases return mostly the same results, they belong on the same page. "Standing desk height" and "how tall should a standing desk be" are one page, not two. Keyword clustering covers the grouping methods, and topic clusters covers how those groups link into a site structure.

Pick one target keyword per page

Every group gets one page, and every page gets one primary keyword. Pick the phrase in the group with the most searches that still matches the intent, put it in the title and an H2, and let the other keywords in the group become sections or natural wording.

Write the result down as a map of keyword group to URL, marking which pages exist and which you still need. That map is the deliverable. A keyword list without it is just data.

If two of your existing pages already target the same group, merge them or retarget one. Otherwise they compete with each other, which is keyword cannibalization.

Add the questions people ask AI

People ask AI assistants longer questions than they type into Google, and those questions tell you what each page must cover. A search for "standing desk" hides questions like "is a standing desk worth it if I already have back pain" and "how long should I stand per day".

You can't see ChatGPT's prompt data, so work from proxies. Google's People Also Ask boxes, question-shaped queries in Search Console and your own support inbox all count. The AI Question Finder lists the questions AI Overviews or ChatGPT answer for a topic, with the sources they cite. Questions people ask ChatGPT covers sorting them by buying stage.

Don't build a page per question. Attach each one to the page from step four that owns its topic, and answer it under its own heading.

Run keyword research on one topic

Our keyword research tool handles the expand, volume and difficulty steps for one seed. Enter a keyword or topic and pick a country. It returns up to 99 related Google searches, taken from the phrase and from the searches related to those, each with estimated monthly searches, difficulty from 0 to 100, search intent, cost per click, the top-of-page bid range and how searches changed over the last year.

It sorts the list for you. Keywords at difficulty 30 or less with enough searches are marked easy to win, learning searches at difficulty 50 or less become blog topics, and buying searches with a cost per click go into the list worth bidding on. Each run costs 34 credits. The numbers are estimates from our keyword data, so compare keywords with each other rather than treat them as counts. It does not track rankings, look up a competitor's keywords, read your Search Console data or write briefs, so the seeds, intent check and page map stay your job.