Doing keyword research without paid tools

In short

Usable keyword research comes from five free sources: search suggestions, the questions box in the results, Search Console, competitor pages and your own customer conversations.
For small companies, search volume is the least important number. Intent and reachability are what decide.
Judgement happens through three yes-no questions rather than columns of figures — faster and just as sound for the decision.
Paid tools pay off from roughly 30 pieces a year, or when advertising budget has to be allocated. Not before.

The usual entry into search optimisation goes: subscribe to a tool, type in a term, sort the list by volume, start at the top. The result is a list of terms everyone else also starts at the top of — and where the volume consists of people who never buy anything.

The process below reverses the order. It starts with what your customers actually ask, and only checks reachability at the end.

Five faint streams of light converge from the edges into one dense bright channel — Five sources, one result. None of them costs anything.

Step 1: the raw list from five sources

The goal here is 80 to 200 terms and questions, unsorted. No judging yet.

Source 1 – Your own customer conversations

How do customers describe their problem before they know your technical term? These phrasings are the most valuable on the whole list, because no competitor can pull them out of a tool.

Source: notes, emails, enquiries

Source 2 – Search suggestions

Type the term into search and note the suggestions. Then run the same term prefixed with "how", "why", "what does it cost", "which" — and with every letter from a to z.

Source: Google, Bing, YouTube – each gives different results

Source 3 – Questions in the search results

The "people also ask" box is a free question database. Every expanded question produces new ones. Twenty minutes of clicking yields 40 to 60 real questions.

Source: the results page itself

Source 4 – Your own Search Console

Under "Performance" sit the queries you already appear for, with position. Anything at position 5 to 20 is reachable ground and usually the fastest gain available.

Source: Google Search Console, free

Source 5 – Competitor pages

The three to five providers ranking well in your field: which headings do they use, which questions do they answer? Do not copy — look for the gaps.

Source: competitors, trade media, forums

Worth knowing

A substantial share of all search queries is asked for the very first time. Those queries appear in no tool with a volume figure — they simply do not have one yet.

At the same time they are particularly valuable: someone phrasing that precisely is usually closer to a decision than someone typing a short generic term. That is exactly why fixating on search volume costs small providers disproportionately.

Step 2: sort by intent

Now every term on the raw list gets exactly one of four marks. Assigning takes two seconds per term.

Intent	Recognised by	Example	Value to you
Knowledge	what, why, meaning	"what is a lead magnet"	low to medium
Method	how, guide, template	"keyword research yourself"	high
Comparison	or, vs, alternative, review	"spreadsheet or CRM"	very high
Purchase	buy, price, cost, provider	"marketing automation cost"	very high

If time is short, delete everything under "knowledge" outright and start with "method" and "comparison". That is uncomfortable, because knowledge questions make the easiest texts — which is exactly why nearly everyone produces them.

Step 3: check reachability — without a tool

A scattered cloud of points resolves into a column whose topmost elements glow brightly — Sorting is not by volume but by what is actually reachable.

For each remaining term, open the results page and answer three questions. Two minutes per term.

Who occupies the top five? If it is only large portals, trade media and market leaders, the term is out of reach for years. If at least one provider your size is there, it is reachable.
How good are the results really? Shallow texts in the top spots are an invitation. Three thorough, current pieces are a warning.
Does the search engine already answer the question itself? If a direct answer box sits at the top and settles the matter, there are few clicks to be had. The term then only pays off if you want to be the cited source — which is a different goal.

Tip Run this check in a private window without being signed in. Signed in, you see a results list shaped by your own behaviour — and are then judging a reality that exists for nobody else.

Step 4: cluster rather than split

The most common mistake after research: planning a separate piece for every term. That produces 40 thin texts competing with each other.

The opposite is right. Terms that ask the same question in different words belong in one piece. "keyword research free", "keyword research without tools", "keyword research yourself" and "how do I find the right search terms" are one page, not four.

Rule of thumb: if you would write the same text twice, it was a cluster. Twelve clustered topics are worth more than forty separate ones.

From practice

After a few months, Search Console delivers the most honest part of the research: the queries where you sit at position 6 to 15. For those terms the search engine already considers you roughly relevant — the way onto page one is shorter there than for any new topic.

In practice, reworking five existing pieces almost always beats writing five new ones. It just does not feel like progress, because nothing new appears on the list afterwards.

Step 5: set the order

The clustered topics become a sorted list. At the top sits whatever meets three conditions: a reachable results page, intent of "comparison" or "purchase", and something of your own to say about it.

The third condition is the most frequently skipped and the most important. A topic where all you can do is summarise what exists elsewhere wins neither with search engines nor with AI answers.

Prompt

I have collected a raw list of search terms from my industry.

My situation:
- What we offer: [offering]
- Audience: [as narrow as possible]
- Region/language: [e.g. Switzerland, German]

Here is the raw list:
[paste list, one term per line]

Tasks:
1. Assign each term exactly one intent:
   knowledge / method / comparison / purchase.
2. Group terms that ask the same question in different words into
   clusters. For each cluster, name a main term and the variants.
3. For each cluster, write a heading that names the result, not
   the topic.
4. Sort the clusters by proximity to a buying decision, not by
   assumed search volume.

Important: do not invent search volumes or difficulty scores.
If you do not know a number, write "not measured".

Careful Language models will readily hand you search volumes on request. Those numbers are invented — no model has access to a search engine's query data. That is exactly why the instruction not to invent them sits in the prompt. Cut it, and you plan your year on figures nobody has measured.

When paid tools start to pay off

There are three clear triggers. Before those, the expense is not justified.

When advertising budget is being allocated. Once real money is distributed across search terms, volume and cost data stop being a luxury and become a precondition.

From roughly 30 pieces a year. At that volume, manual research costs more working time than the subscription.

When several languages come into play. Compiling one raw list per language by hand is doable. Ten languages are not.

In closing

Sound keyword research is legwork, not a tooling problem. The five sources deliver a list in three to four hours that will carry a year of content.

What is missing is the number in the "search volume" column. What you get instead is a list of real phrasings from people who actually need something — and that is precisely what is missing from most tool-driven research.

Common questions

How do you do keyword research without a paid tool?

Through five free sources: your own customer conversations, search engine suggestions, the questions box in the results, your own Google Search Console, and competitor pages. The raw list is then sorted by search intent, reachability is judged directly in the results, and similar terms are grouped into topic clusters.

Do you really need search volume data?

For content planning in small and mid-sized companies, rarely. What decides is search intent and the reachability of the results page — both can be judged without a tool. Volume data becomes necessary once paid advertising is running or more than roughly 30 pieces a year are planned.

How long does keyword research take?

Three to four hours is realistic for a full first pass: one hour for the raw list, half an hour for intent assignment, one to two hours for the reachability check and half an hour for clustering. The result usually carries for a year.

Can you use a language model for keyword research?

For sorting, clustering and phrasing, yes — it is fast and reliable there. Not for search volumes or difficulty scores: models have no access to query data and will produce plausible-looking but invented numbers on request. That is why the prompt should always instruct them to mark unknown figures as "not measured".

How many keywords does one page need?

One main term plus the variants that ask the same question differently — typically three to eight. Creating a separate page for each variant makes your own pages compete with each other, and none of them gets strong enough.

Marketing that sets itself up

The Studio Engine beta is live. Claim your spot and help shape it from the start.

Join the beta →

← Back to overview