How do you do keyword research without paid tools?

TL;DR

• A small site with a few hundred pages does not have a keyword problem at the scale Ahrefs or SEMrush were built for. The free tools surface the same top fifty queries that actually drive your traffic, plus the language your customers actually use.
• Google Search Console is the single richest free keyword data source a small site has. Sixteen months of real query history. The Performance report sorted by impressions descending shows the keywords you already rank for between positions three and twenty. Those are the keywords one new page or one rewrite can move.
• Google AutoComplete and the People Also Ask box give you the queries Google itself thinks are real. Both surfaces are free, both update daily, and together they expand a seed list better than any chat prompt without a validation step.
• A free Claude or ChatGPT prompt expands fifteen seeds into a hundred variants in under a minute. The trap is that the same prompt will confidently invent search volumes for every variant. Use the chat surface for the language. Never trust it for the numbers.
• The next page to write is the cluster of three to seven related queries where you already rank between positions three and twenty. Pick that cluster — it sits one rewrite away from page one. Write that page first. The next page is the next cluster down.

A new client asks how often you research keywords. The honest answer for a freelancer or a small business is once a quarter, on a Sunday afternoon, with a notebook and three browser tabs. There is no monthly subscription.

There is no spreadsheet with twenty thousand rows. The next page to write is almost always sitting in the data Google already gives you for free.

The paid keyword tools are not bad. They are built for a different job. An agency tracking five hundred client sites needs the dashboards and the API access.

A small site with eighty pages does not. The same site can run keyword research that drives the next quarter’s content calendar in under two hours, with tools that cost zero dollars.

Keyword research without paid tools is like cooking from a small kitchen garden. You work with what is already in front of you, instead of paying a delivery service for the same vegetables marked up four times. Your garden holds the queries Google already shows you, the questions your customers already ask, and the language your own pages already use.

This piece is the seven-step free workflow. It uses Google Search Console, AutoComplete, People Also Ask, a free chat surface, and one cross-validation step.

Why does free keyword research still beat paid for a small site?

A small site with a few hundred pages does not have a keyword problem at the scale paid tools were built for.

The expensive tools surface keywords nobody on a small site has the time to write for anyway. They are built for the job of finding ten thousand keywords across a thousand pages, then prioritising the top hundred. A small site has eighty pages and three to write next quarter. The top fifty queries that drive your traffic are sitting in Search Console for free.

The free tools also give you the language your customers actually use. The autocomplete suggestions are real queries enough people typed for Google to remember the pattern. The People Also Ask box is curated from real questions.

The chat-surface prompt over your seed list pulls from billions of real conversations. Paid tools layer their own database on top. The free stack reads from the same underlying data Google already exposes.

Spending two hundred dollars a month for the rest is the wrong tradeoff at small scale. The signal-to-cost ratio drops once you are past the first thousand keywords, and most small sites do not need the second thousand.

Where do your seed keywords actually come from?

Three places, all free.

The pages your customers email you about. Look at the last twenty support emails or sales enquiries. The phrases the customer used to describe what they wanted are the seed phrases that mean money to your business.

The questions you answer in the first call with a new client. Write down the three to five questions you find yourself answering on every kickoff call. Those are the questions everyone in your category asks. The chances are good that nobody on your site has written a clean answer to them yet.

The words your business actually uses on its product or service pages. Pull the H1 and H2 lines from your top ten pages. Strip out the marketing-deck phrases and keep the plain ones. Those are the words your site already ranks for, even if you never planned to.

A list of fifteen seed terms drawn from those three sources beats any auto-generated seed list a paid tool ever exported. The seed list is not the bottleneck. The page that gets written from it is.

What does Google Search Console give you that no paid tool can?

The actual queries that landed real visitors on your site in the last sixteen months.

Search Console is free, hooked up in five minutes, and stores more historical query data than most paid tools surface in their default views. The Performance report shows you, for any period in the last sixteen months, every query that drove an impression to your site. It also shows impressions, clicks, average position, and click-through rate per query.

Sort by impressions descending. The first fifty rows are the queries your site already ranks for. They are the queries with money attached, because they are the queries Google has decided your site is genuinely an answer for.

Then sort by position descending, filtered to positions three through twenty. The first thirty rows are the queries you are close to ranking on page one for. A new page targeting one of those queries, or a rewrite, moves the position from twelve to four faster than any other content move. This is the single most powerful free keyword move a small site has, and most owners never use it.

The site article on classifying search intent at scale covers the next step. It labels each query by intent so you know what page format to write.

How does Google’s own AutoComplete and People Also Ask fill the gaps?

Type your seed keyword into Google with the cursor at the end. Watch what the autocomplete suggests.

Each suggestion is a real query enough people typed for Google to remember the pattern. AutoComplete will not show you fabricated queries. The list updates as you type, and adding a single letter at the end ("photographer m" then "photographer ma") branches the suggestion tree.

Then scroll the SERP and read the People Also Ask box. Four to eight question-shaped queries Google attached to your seed. Click one and the box expands with three to five more. Click one of those and you are at twenty real queries about your seed without paying anything.

Both surfaces are free. Both update daily. Together they give you better seed expansion than any chat prompt without a real-data validation step.

The autocomplete output is the head-term cluster. The People Also Ask output is the long-tail question cluster. The two together cover most of what a paid keyword tool would surface for the same seed.

The mining trick that paid tools cannot replicate is reading Reddit threads and Quora pages for your seed. The phrases people use in a forum to describe their problem are the phrases that show up as long-tail queries six months later. A small site that pulls fifteen long-tail queries from one Reddit thread is doing the same work the chat surfaces do. The source data is the same forum thread.

Can a free chat reply expand your seed list — and where does it lie?

A free Claude or ChatGPT prompt can take your fifteen seed terms and produce a hundred candidate variants in under a minute.

The prompt is short. "Here are fifteen seed keywords for my small business in (your category). Generate a hundred candidate variants grouped into clusters by intent.

For each variant, label intent as informational, commercial-investigational, or transactional. Return as a CSV with five columns: keyword, intent, cluster, variant rationale, and a confidence score from 0 to 100."

The variants come back fast. The clustering is usually decent. The intent labels are reliable on clear transactional and informational cases. They are brittle on commercial-investigational queries and on local-intent queries — those are the rows that need a spot-check.

The trap is the column the prompt did not ask for and the chat surface will sometimes add anyway. Search volume. The chat surfaces will confidently invent search volumes that look plausible and are entirely fabricated.

Hallucinated volume is the dominant failure mode of free chat-based keyword research. The numbers look like they came from a database — they did not.

Use the chat surface for the language and the variant generation. Never trust it for the numbers. The site article on expanding seed keywords with language models covers the deeper how-to. It includes the prompt patterns that reduce the hallucination rate.

How do you validate volumes without buying a tool?

Cross-reference the candidate variants against your real Search Console impressions.

A variant Google has actually shown you impressions for is a real query. A variant the chat surface invented but Search Console has never registered impressions for is suspect. It might be a real query nobody at your scale searches for.

Or it might be a fabrication. Either way, do not write a page targeting it before you check.

The free Google Keyword Planner inside Google Ads gives ranged volumes for real queries. Open Google Ads. Set up a free account. Open the Keyword Planner.

Paste your top thirty validated variants. The Planner returns a volume range for each (something like "100-1K monthly searches") and a competition signal. The ranges are wide, but they are still better than fabricated numbers.

Together those two free sources catch most of what a hundred-dollar-a-month tool would also catch. Search Console tells you what your site has actually been shown for. Keyword Planner tells you the volume range for queries you have not yet been shown for. The combination is enough to plan a quarter’s content for a small site.

The validation step is the difference between free keyword research that works and free keyword research that wastes a Sunday. The wasted Sundays go on pages nobody will ever search for.

Other questions worth answering

When should a freelancer move from the zero-cost stack to a subscription dashboard?

Once your domain crosses roughly 500 indexed URLs and your monthly impressions clear 100,000, the zero-cost stack starts leaking signal. Below those rough thresholds, the Search Console plus AutoComplete plus free chat workflow returns the same top-ranking queries a subscription dashboard would. Above them, premium filters earn back the spend in saved hours.

How do Reddit threads compare to forum mining as raw input?

Reddit gives you dated comments, vote counts, and a thread-level upvote signal that filters weak phrasings out before you read them. Older niche forums skew narrower but carry deeper jargon from long-tenured members. Pull from both. About 2026 of every 10 long-tail variants in my own notebooks came from Reddit first.

Does a brand-new domain with zero impressions get anything from this workflow?

Yes, but the load shifts. With zero impressions, Search Console gives you nothing for the first 60 to 90 days. Lean harder on AutoComplete branches, the Also-Asked box, and customer-email phrasings during that window. Treat the first quarter as a hypothesis quarter, then validate against real impressions once they arrive.

Why does intent labelling break on commercial-investigational queries?

Because the same phrase serves two readers at once. "Best CRM for solo consultants" looks transactional to a buyer comparing four products and informational to a reader who has never used a CRM. Per the AI in keyword and topic-cluster workflows brief dated 2026, that ambiguity is where free chat classifiers misfire most often. Spot-check the low-confidence rows by hand.

How often should a quarterly refresh truly run?

Once every 90 days for most freelancers, with one mid-quarter check around day 45. The 90-day cadence matches Search Console’s rolling-data rhythm. The mid-quarter check catches any cluster that jumped from position 14 to position 6 between refreshes. Anything tighter than 30 days burns Sunday afternoons without changing the next-page decision.

Which page should you write next?

Sort your validated keyword list by current Google position.

Find the cluster of three to seven related queries where you already rank between positions three and twenty. That cluster is one new page or one rewrite away from page one. Google has already decided your site is a relevant answer. The position just needs a stronger page to push it.

Write that page first. Use the cluster’s queries to shape the H1, the H2s, and the body. One H2 per query if the cluster has five or fewer queries.

One H2 per sub-cluster if the cluster has more. Answer each H2 in the first paragraph below it, in three or four short sentences a chat surface can lift cleanly.

The next page after that is the next cluster down the list. The page after that is the next. Most small sites can produce three pages a quarter on this rhythm and see the cumulative result inside two quarters.

If you have run through the workflow and are stuck on which cluster to pick, you can contact me here. Send your Search Console export of impressions for the last ninety days and the rough category of your business. I will name the one cluster to write next and the one query to put in the H1. No pitch.