How to find SEO keywords: the five sources you already own, then the tools

The fastest way to find SEO keywords is to mine the sources you already own before you open a keyword tool: your Search Console query export, your site search log, your sales call notes, your support tickets and your reviews. Those five give you the exact wording real buyers use, with evidence that someone already asked. Then expand with autocomplete, People Also Ask and a competitor gap check, and finally qualify each candidate against the search results page to see whether you can win it. Volume is the last filter, not the first.

Most keyword lists fail for the same reason: they were generated from a tool's suggestions rather than from demand that exists. A keyword is worth having when somebody with money asked that question, your site can credibly answer it, and the results page is not already owned by something you cannot beat. Here is the order that produces that list.

Step one: harvest what you already have

Source	What it gives you	How to read it
Google Search Console, Performance report	Every query that showed your site, with impressions, clicks and average position	Export the full query list for the last twelve months. Sort by impressions where clicks are near zero. Those are queries Google already associates with you and you are not winning.
Site search log	What visitors type once they are on your site	Look for repeated searches that return nothing. Each one is a page you have not written.
Sales calls and demo notes	The question before the question, in the buyer's own words	Collect the first five questions asked on the last twenty calls. Phrase each as a search.
Support tickets and chat transcripts	Post purchase language, problems, comparisons	Group by theme. Recurring themes are high intent pages, and they reduce support load as a bonus.
Reviews, yours and competitors'	The words customers use for the benefit and the objection	Pull the nouns and the complaints. Complaints make the best comparison and alternative pages.

Step two: the striking distance list

Inside the Search Console export sits the highest return work available to most sites. Filter for queries where your average position is roughly between eight and twenty five and impressions are meaningful. Those are pages Google is already willing to show and is not yet convinced by. Improving an existing page to answer the query properly is faster than ranking a new one, and the data tells you which pages to touch.

Group the queries by the page that currently ranks. If three different pages compete for the same query, that is a consolidation job, not a new keyword.

Step three: expand from the question, not the noun

• Autocomplete. Type your core term and run the alphabet after it. Do the same with "best", "vs", "how", "cost of", "near me" and "alternative to" in front of it.
• People Also Ask. Open two or three boxes and keep opening. This is a map of the question tree around your topic, taken straight from the results page.
• Related searches at the bottom of the results page, which surface the next query in the journey.
• Communities. Reddit threads, trade forums, Facebook groups and question sites show the problem stated before anyone knows the product name. These become your top of funnel pages.
• Your own glossary. Every term of art in your industry is a definition page somebody searches for, and definition pages get quoted by assistants.
• Competitor gap. Take three competitors and list the pages they rank for that you have nothing for. Any keyword research tool with an organic research view will do this, and you can approximate it free by reading their sitemaps and navigation.

Step four: qualify by intent, then by winnability

Sort every candidate into one of four intents, because intent decides what the page has to be.

• Informational. "what is", "how to". Needs a teaching page. Converts slowly, builds the topic.
• Commercial investigation. "best", "vs", "alternatives", "reviews", "pricing". The highest value content for most businesses.
• Transactional. "buy", "near me", "hire", "book". Needs a service, product or location page.
• Navigational. A brand name. Only worth targeting when the brand is yours or you are writing a fair comparison.

Then open the results page for each shortlisted query and read it, which is the step most people skip. Ask three things: what format is ranking, who is ranking, and how much of the page is taken by features such as AI Overviews, shopping units, map packs or video. If the top five are all large national sites and the page is full of features, that query is a long project. If two of the top five are thin pages or forum posts, it is available now.

Step five: score and choose

Give each candidate a simple score and build the quarter from the top of the list.

Factor	Question to answer
Business value	If we won this query, would it produce enquiries or revenue, or only traffic?
Evidence of demand	Did a real buyer ask it, or did a tool suggest it?
Winnability	Can we make a better page than the current top three, with the authority we have today?
Effort	Can we publish this in a week, or does it need original data and a month?
Fit with existing pages	New page, or an improvement to something we already have?

Pair the list down to one page per query group. One strong page covering a question and its close variants beats five thin pages splitting the same intent.

Reading volume figures correctly

Every tool's volume number is a modelled estimate, grouped across close variants and averaged over months. Use it to rank candidates against each other, which is what it is good for, and use your own Search Console impressions as the measured figure for queries you already appear on. For a local business or a narrow trade category, a query with a small reported volume can be the most valuable one on your list, because the intent is exact.

This page covers finding and qualifying keywords. For turning the chosen list into pages, the next job is a content brief per page that names the query, the intent, the format that is ranking and the questions the page must answer.

The find SEO keywords questions worth an answer

Can I find SEO keywords without paying for a tool?

Yes. Search Console, your site search log, autocomplete, People Also Ask, related searches and your own sales and support records cover the discovery job completely, and they are more accurate about your buyers than any third party dataset. Paid tools speed up competitor gap analysis and give you comparable volume estimates, which saves time rather than replacing the work above.

How many keywords should one page target?

One primary query and the close variants that share the same intent, which is usually a cluster of five to thirty phrasings. Write for the question, not for the string. Splitting near identical queries across separate pages makes both weaker.

Are long tail keywords still worth targeting?

They are where most sites win. Longer queries carry clearer intent, face less competition and convert better, and they are the phrasings people use when they talk to an assistant. Build the specific pages first and the broader terms become reachable as the topic strengthens.

What is a good keyword difficulty number to aim for?

Difficulty scores differ by tool and are a rough proxy. Use them to sort a long list quickly, then make the real decision by opening the results page and comparing the top three pages against what you can produce. The read of the actual page beats the score.

How often should I redo keyword research?

Re export Search Console monthly to refresh the striking distance list, and do a full discovery pass each quarter or whenever you launch something, enter a new market or see the results page format change for your main queries. Keep one running list rather than starting again each time.

One source of question wording most lists miss

Ask your category question to ChatGPT, Gemini, Perplexity, Claude and Google's AI Overviews, and read both halves of what comes back: the follow up questions the assistant suggests, which are a clean map of buyer intent, and the sources it cites, which tell you which pages are being treated as the authority on your topic.

Those five answer differently, name different companies and cite different sources, and they share no common ranking, so each one gives you a different view of the same demand. Most keyword workflows do not look at this yet, which is why it is a cheap edge. Keep your own brand name out of the prompt so the answer is not steered. AI Knows Us tracks those answers and the sources behind them.