Do your own keyword research: 6 clear steps

Keyword research maps the questions your audience asks, the intent behind those searches, and the page best suited to answer them. Start with your offer and your customers. Then check the real search results and use search volumes only as comparison data. That gives you a plan in which every important page has a clear purpose, rather than an endless list of words.

You can do the basics yourself. The six steps below take you from customer questions to a content plan whose results you can measure in Google Search Console after publishing.

What you get from keyword research

Good research connects language to intent. Someone searching for “what is a headless CMS” expects an explanation. For “have a headless CMS built”, a commercial service page is better suited. That difference determines which type of page you create, how much explanation it needs and which next step feels logical. It is also exactly the distinction Google itself pursues: content “created to benefit people, and not content that’s created to manipulate search engine rankings” (Google Search Central).

The research also helps you detect overlap. Minor wording or spelling variants usually don’t need separate pages when the search results largely show the same pages. Group those variants in one cluster. One clear owner prevents two of your own pages from competing for the same search query.

1. Collect customer questions. Start with problems, services, and the words customers actually use, so the research begins with real questions.
2. Expand the initial list selectively. Use Google Search Console, Google suggestions and one relevant keyword tool to find missing variants.
3. Check the live search results. Use the search engine results page (SERP) to determine whether someone expects an explanation, a service, a comparison, or another type of page.
4. Weigh business value and evidence. Assess relevance, estimated volume, competition and, above all, which credible answer you can add.
5. Cluster queries and choose pages. Group variants with the same intent and assign each cluster to one clear primary page.
6. Publish and measure. Take a baseline measurement, then compare the same URL and query group after 14 and 28 days in Google Search Console.

Start with ten usable search queries rather than hundreds of unfiltered suggestions. For each query, note what someone actually wants to achieve and which answer you can offer. A specific question with less volume can be more valuable than a broad term because the need is clearer and your page can answer more precisely.

Compare fixed periods in Google Search Console

If you already have a website, Google Search Console is your primary source of existing search query data. Open the Search results report, filter by one relevant page, and then look at the queries. You can see which terms already trigger that page, how many impressions it receives, and whether its search snippet attracts clicks. Google describes these reports as the way website owners understand and improve their Google Search performance (Google Search Console reports).

Always compare fixed and complete periods. For example, compare the last 28 days with the previous 28 days, or work with a baseline measurement before publication and a check after 14 and 28 days. Use the same URL, query group, and search type for both periods. Otherwise, you are comparing two different situations, and a trend quickly becomes misleading.

• Many impressions and few clicks can point to a weak title or meta description.
• An average position between roughly 8 and 20 may indicate an existing content opportunity, but check the real SERP first.
• New relevant search queries show that Google is starting to understand the topic more broadly.
• Multiple pages appearing for the same query can signal unclear page ownership.

Clicks, impressions, CTR, and average position each tell only part of the story. A page can appear for new relevant terms and therefore temporarily show a lower average position. Look at the individual queries and the page as a whole.

1. Set the comparison period. Compare two complete periods with the same URL, query group, and search type.
2. Read clicks, impressions, CTR, and average position together. A change in one metric is not a conclusion. Look at individual queries and at the page as a whole.
3. Assign each cluster to one primary page. Check for overlap and choose a clear owner only when the intent and SERP support it.

Want to set this up yourself? The guide to setting up Google Search Console helps you get started with the basic configuration.

Cluster search queries and choose a primary page

A cluster contains search queries that share the same need and largely the same search results. Don’t just check whether words look alike. “SEO keyword research” and “doing keyword research” probably fit an informative guide. “SEO advice” has commercial intent and belongs on a service page instead.

Make a simple mapping with four columns: search query, intent, primary page and role. For this guide that looks like this, for example:

• “do keyword research” → informative → /en/blog/how-to-search-for-keywords → practical step-by-step plan.
• “seo keyword research” → informative → the same guide → additional depth within the existing topic.
• “seo advice” → commercial → /en/seo → service page.
• “set up google search console” → informative → separate guide → technical configuration.

Then check whether another page already gets impressions for the same query group. That is not automatically cannibalization. Sometimes two pages complement each other logically. It becomes a problem when both pages try to give the same answer and Google recognizes no clear owner.

How Straffe Sites uses AI agents

Straffe Sites uses specialized AI agent workflows on Hermes to prepare the recurring analysis work. The agents collect and compare current GSC page/query data over fixed periods. They group related queries, flag possible intent conflicts or overlap between pages, and turn the findings into a verifiable content brief, metadata options, and a measurement plan.

I make the decisions. I check the real search results, assess the primary intent, and choose which page becomes the owner of a search query. Brand voice, commercial priorities, design, code, claims, and publication are not automated. A plausible proposal is not yet a publishable change.

After an adjustment, I use the same verification chain as for other website work: readback of the built HTML, build, browser check, inspection of the live page, and a new GSC measurement for the same URL and query group. Automation helps with the research, while editorial responsibility remains human.

Below is a real run: the cluster around “website laten maken” with, per query, the impressions, the current page, and the proposal that follows from it.

How can AI help with your keyword research?

You don’t have to run every analysis in this guide by hand. In Google Search Console, open the Search results report, click Export, and download the data as CSV or XLSX, or share the table directly through Google Sheets. That file contains your queries, pages, clicks, impressions, CTR, and average position: exactly the raw material for a first analysis.

Give that export to an AI chat tool such as ChatGPT or Claude, or to any AI agent you already work with. Ask a concrete question, for example: “Group these queries by intent with the average position and CTR per group” or “Show which pages appear for each query and where those pages overlap.” Within a few minutes, you have a clustering that you then check against the real search results.

At Straffe Sites this kind of analysis runs on Hermes, an AI agent that repeats the comparison work over fixed periods. Want to get started with an agent yourself? Read the Hermes Agent documentation.

Keep one rule: the agent prepares, you decide. A clustering based on your export is a proposal, not a conclusion. Check the intent in the live search results; you still decide which page gets each topic.

Which tools do you need?

For the basics, you only need Google Search Console, the live search results, and a spreadsheet. Tools such as Semrush, Ahrefs, KWFinder, or LowFruits can collect suggestions and competition signals faster. Treat search volume and difficulty as estimates. They help you compare terms but don’t predict traffic, inquiries, or rankings.

Choose a tool that fits your question. If you already have traffic, start with Search Console. If you are starting a new site, combine customer language with Google suggestions and a limited toolset. Switching between five tools rarely yields a better choice if the intent is not yet clear.

Use keywords naturally on your website

Name the primary topic in the title and main heading when it reads naturally, then answer the query early on the page. Use descriptive H2s for the subquestions and write internal links so the destination is clear. An anchor like “technical SEO and keyword advice” says more than “read more.”

Alt text describes what is actually in an image. Don’t stuff a list of keywords into it. Keep URLs short and understandable, but don’t create a new URL just to catch a variant. The page’s content and intent matter more than forced repetition in every element.

Also, don’t work with a fixed number of keywords per page. Choose one primary question and cover the related subquestions thoroughly. Synonyms and natural variants arise automatically when you explain the topic clearly.

Measure after publication and adjust

Record the baseline measurement before the change. Note the period, property, URL, query group, clicks, impressions, CTR, and average position. Check after 14 days for early movement. After 28 days, you can assess the trend more reliably, although low search volumes and seasonality remain important limitations.

Make one clearly defined group of changes at a time. For example, first revise the title, meta description, and introduction. Then measure again before rewriting the rest of the page. This lets you see what made a difference without shifting the baseline for every measurement.

Professional help is useful when several services overlap, different pages appear for the same queries or a large export does not yet produce a usable site structure. Straffe Sites helps with technical SEO and keyword advice. Do you already have a list or an initial plan? Discuss your keyword research and we’ll identify where search intent, content and technical constraints still conflict.

Conclusion

Keyword research only works when you turn search language into clear decisions about which page owns each topic. Collect real questions, check the SERP, cluster by intent, and assign each topic to one owner. Then measure the same URL and query group again. Tools and AI agents can prepare the comparison work, for example by clustering your GSC export in an AI chat, but you still make the final publishing decision.