You can do genuinely good keyword research without buying anything. Google publishes its own demand data through autocomplete, People Also Ask, and related searches; Search Console shows the queries already reaching you; your inbox and sales calls contain the exact wording customers use; and community sites show the questions people ask when no brand is listening. Paid tools make this faster and add volume estimates, but they are an accelerator, not the starting line. Here is the free method, in the order we run it for new clients.
Why tools are not step one
Keyword tools answer "how many people search this phrase". They cannot tell you which phrases your buyers use, which questions stall your sales calls, or which problems your best customers had before they found you. That knowledge lives in your business, and starting with a tool means you search for keywords you already thought of, which is the most common way a keyword list ends up describing the company instead of the customer.
Start with language, then validate with data. That order also produces content that sounds like a human, which matters more now that both readers and AI assistants reward natural phrasing.
Seven free sources, in order
1. Your own conversations
Open your sent mail, your support inbox, and your enquiry form submissions, and write down the exact phrasing people use when describing their problem. Not your category name, their words. This single hour usually produces the best keywords in the whole project, because customers describe problems while companies describe products.
2. Google autocomplete
Type your core topic into Google and read the suggestions without pressing enter. These come from real search behavior. Then repeat with modifiers in front: how, why, best, cheap, near me, vs, for. Each modifier opens a different branch of demand, and the "vs" branch in particular reveals exactly who your buyers compare you against.
3. People Also Ask
Search your topic and expand the People Also Ask box. Every expansion loads more questions, and each is a real query Google considers related. Twenty minutes here produces more genuine questions than most paid exports, and it maps directly onto the answer-first structure AI assistants extract from.
4. Related searches
Scroll to the bottom of the results page. These are the searches people run before or after this one, which makes them a map of the journey rather than a list of synonyms.
5. Google Search Console
If your site has any history, the Performance report lists actual queries that showed your pages, including hundreds you never targeted. Filter for queries with impressions but few clicks: those are topics Google already associates with you where you rank too low to earn the visit. They are usually the fastest wins available to any site.
6. Communities where your customers talk
Reddit, niche forums, Facebook groups, and question sites show people asking without a brand in the room, which produces unfiltered wording and unmet needs. This source doubles as AEO research, because assistants frequently cite exactly these discussions, as covered in our piece on what social media really does for SEO.
7. Competitors' own headings
Open the pages currently ranking for your topic and read their H1s, H2s, and FAQ sections. You are not copying; you are reading which subtopics Google has already rewarded, and noticing which questions every one of them fails to answer properly. That gap is your opening.
The AI-era addition almost nobody does
Search boxes capture short queries; assistants capture full questions, and the two vocabularies differ. Two free habits worth adopting:
- Ask the assistants what people ask. Prompt ChatGPT, Gemini, or Perplexity with "what questions do people ask before hiring a [your category]" and note the phrasing. The output reflects patterns in the text these systems were trained on, which is a reasonable proxy for how your market discusses the problem.
- Watch which sources get cited. Ask the assistants your buyers' questions and read the citation list. Those pages are your real competition for the answer, and their structure tells you what gets quoted.
Together these give you a question list rather than a keyword list, which is the correct raw material for content that has to satisfy both a results page and an answer engine. Our AI visibility self-test turns the same habit into a monthly score.
Which keywords are actually worth targeting
Volume without judgment produces a list of things you cannot win. Three free checks before committing to any keyword:
- Read the intent from the results. Search the phrase and look at what ranks. Product pages mean buying intent; guides mean learning intent. Publish the format that is already winning, because the results page is Google telling you what the query means.
- Estimate winnability by eye. If page one is entirely national brands and major publications, a young site will not break in soon. If it contains forum threads, thin pages, or off-topic results, that is an opening.
- Prefer specific over broad. "SEO" is unwinnable and converts poorly; "how much does SEO cost for a small business" is winnable and converts well. Specific queries carry the intent that turns visits into enquiries.
Turning the list into pages
One question, one page, answered completely. Group near-duplicate phrasings onto a single page rather than making a page per variation, because Google treats them as the same query and splitting them competes with yourself. Then open each page with a direct answer in the first two sentences, which serves impatient readers, featured snippets, and AI extraction with the same paragraph.
Publish in clusters around one topic rather than scattering across many, for the reasons set out in our guide to SEO timelines: authority accumulates by topic, so ten connected pieces outperform thirty unrelated ones.
Frequently asked questions
Do I need paid keyword tools eventually?
They help once you are publishing regularly and need volume estimates, difficulty scores, and competitor gap analysis at scale. Before that point, they mostly provide expensive confirmation of what the free sources already showed you.
How many keywords should one page target?
One primary question, plus the natural variations of how people phrase it. Pages built around a single clear question rank better and get quoted more often than pages stuffed with loosely related terms.
What is a good search volume to aim for?
For most small businesses, a handful of specific queries with clear buying intent beats one broad term with large volume. Ten visitors ready to buy are worth more than a thousand who are curious, and specific queries are also far easier to win.
Does keyword research still matter with AI search?
More than ever, though the unit changed from phrases to questions. You still need to know what people ask; you simply need the full question rather than the truncated keyword, because that is the form assistants receive and answer.
The bottom line
Spend an hour in your own inbox, an hour in Google's free suggestion features, and an hour reading the questions your market asks in public, and you will have a better keyword list than most businesses buy. Then judge each one by what already ranks, build one page per question, and answer it properly. If you would like that list built for you, with the winnable queries separated from the wishful ones, our free SEO and AEO audit includes exactly that analysis for your site.
