You want the short list of words that keyword research starts from, and you are probably looking for the tool that produces it. That is the wrong end of the job. Your seed keywords are already written down, in your own material: the sales emails you send, the quotes that came back rejected, and the three questions every customer asks before they say yes. A seed keyword is simply a plain word or phrase for the thing you sell, the one a person would type if they wanted it. Write your own list first, out of that material. Then open a tool and use it to check the list, not to write it. Google’s own help page for Keyword Planner is plain about which job the tool does: you arrive with “words and/or websites related to your business”, and it gives you ideas from there. A tool expands what you give it. Give it no words and it wants a website instead, which is Google’s own second path: “Start with a website: Enter any website and Google will look for keywords related to the content on that site.” Point that at your own site and it reads your own language back to you, and your own language is rarely your customers’ language.
The rest of this page is that order, step by step, with what each check needs. We have also put our own Search Console figures on the page as the worked example, because they happen to show exactly what the method is for. Search Console is Google’s own report on what people searched for before your site appeared in the results. In July 2026, ours recorded 603 impressions across 22 different search rows carrying the word seed, and not one click.
Our Search Console figures are from the July 2026 export, pulled on 14 August 2026. Google’s own pages were checked on 20 September 2026.
What a seed keyword actually is, and what it is not
A seed keyword is an input. It is a short, plain phrase naming a thing you sell or a problem you fix, and its whole job is to be fed into something else so that more phrases come back. That is different from the keyword a page is written around, which is usually longer and more specific, and which you only get to after the seed has done its work.
The two are easy to blur together, and the cost of blurring them is real: you end up trying to write one page for “web design”, which is nobody’s actual question, instead of the page that answers what somebody typed.
| Seed keyword | Target keyword | |
|---|---|---|
| What it is | A short plain phrase for the thing itself | The full phrase one page is written to answer |
| What you do with it | Type it into a tool or a search box to get more phrases back | Write one page around it, and only one |
| How many you want | Enough to cover everything you sell, which for most small firms is a short list rather than a spreadsheet | One per page |
| Example | website maintenance | what a website maintenance service costs in Singapore |
The tool expands what you bring it. It does not choose it for you. Google says so on its own help page, which describes the tool’s job as helping you “discover new keywords related to your business and view estimates of the searches they receive and the cost to target them”. The flow starts with an instruction to the user: “Enter words and/or websites related to your business to view keyword ideas.” You are expected to arrive with something of your own, either words or your site. Words you chose beat words a tool reads back off your own site. Everything in the next two sections is about where your words come from.
Note the other word in Google’s sentence: estimates. The monthly search figure beside a keyword is the tool’s estimate for that word, not a property the word carries around with it. Hold on to that, because it is a short slide from “this word has a big number beside it” to “this is a good seed for me”, and the second sentence does not follow from the first. What makes a seed good for you is in the section on wrong seeds, further down.
Where your seeds are already written down
This part needs no tool and no budget, only an hour of your attention. A small firm that has been trading for even a year has already collected the words its customers use. Nobody has typed them into one list yet, that is all. Set aside an hour and do these five things in order.
- Open your sent folder and read the last twenty emails you wrote to somebody who was not yet a customer. Write down the words they used for the thing you do, not the words you used back at them.
- Pull the last ten quotes that were rejected or went quiet. Note what each one was for, in the wording of the enquiry that started it.
- Write down the three questions you answer on every first call. The ones you are tired of answering are the ones people are looking up before they call.
- List what you call your own services on your website. Beside each one, write what customers call it out loud. Where the two differ, the customer’s word is the one to keep.
- Add the thing people keep asking you for that you do not actually sell. It is still a seed for a page, even when it will never be a line on an invoice.
What comes out of that hour is a short list of plain phrases, in the words of people who were about to spend money. It is not tidy and it is not ranked. That is fine. It is the only list you will build that nobody else has.
The alternative, copying the keywords a competitor ranks for, imports more than words. Their list was built around their prices, their service area and the size of client they chase. If they sell to fifty-person companies in Sydney and you sell to five-person companies in Singapore, their words will quietly aim your whole site at the wrong reader, and the mistake will not show up for six months.
There is a second reason to start with your own material, and it comes from Google’s own published guidance on what makes a page helpful. That page gives site owners a list of questions to ask themselves, and two of them bear on this. “Does the content provide original information, reporting, research, or analysis?” And: “Does the content provide substantial value when compared to other pages in search results?” (Creating helpful, reliable, people-first content, page stamped as last updated 10 December 2025.) That guidance is about pages, not about keyword lists. But a page planned from a list you copied off a rival is starting the work of answering both questions with somebody else’s material, which is a hard place to start from.
Now check your list against Google’s own free tools
The tools come second, and none of the three charges you anything. Autocomplete needs a search box. Search Console reports on a site you have added as your own property. Keyword Planner sits inside a Google Ads account, and Google’s help page says you must finish the account setup by entering your billing details before the keyword ideas feature opens. Its own answer to why that feature is locked: “You must enter your billing information and complete your account setup to access basic features like keyword suggestions.” Nothing is charged for using it, and the card details are still asked for first. You are not asking them what to sell. You are asking them whether anybody outside your office says it the way you just wrote it down.
Autocomplete, for the wording
Type each phrase from your list into a Google search box and stop before you press enter. The list that drops down is not a list of ideas Google thinks you might enjoy. Google is explicit about this on its own post explaining how autocomplete works: it calls the entries “predictions” rather than “suggestions”, and says why. “Autocomplete is designed to help people complete a search they were intending to do, not to suggest new types of searches to be performed.” Where the predictions come from is the useful part: “We look at the real searches that happen on Google and show common and trending ones relevant to the characters that are entered and also related to your location and previous searches.” That post carries the date 20 April 2018, so the wording is old and the explanation is still Google’s own.
So a prediction is a record of searches people actually finished typing. If your phrase brings back a drop-down full of related wordings, people say it roughly the way you wrote it. If it brings back almost nothing, that is worth knowing before you plan a page around it.
The same sentence carries the catch, and it is the one most likely to waste your afternoon: the predictions are shaped by your location and your own previous searches. Your drop-down is partly a mirror. Check from Singapore, in a browser window that is not signed in to you. Then the drop-down is closer to what a Singapore customer sees, and your own week of searching stays out of it. And a screenshot of somebody else’s autocomplete, in somebody else’s country, is not evidence about yours.
Search Console, for what you already appear for
If your site has been live for a while, Google is already telling you which searches it shows you for, in the Performance report in Search Console. Google’s own description of the report is that it lets you see “what search queries are most likely to show your site” and “which queries are bringing traffic to your site”.
Two of its columns matter here. Impressions are, in Google’s wording, “How many times your site appeared in Search results”. Average position is “The average position of the topmost result from your site”. A query with impressions and no clicks means Google is showing you for something, and nobody is picking you. That is a seed with evidence attached, and the next section is what ours looked like.
The obvious limit: this only works if your site already shows up for something. A site that went live last month has an empty report, and the section on having no traffic and no budget is written for exactly that.
Keyword Planner, last, for scale
Now feed the list in. Keyword Planner takes your words and returns ideas and estimates, which is the job it says it does. Treat the output as a spelling check on your list and a source of longer phrasings, not as the list itself. For the wider set of tools around this job, we have been through them in our notes on SEO tools.
What our own Search Console showed us
We ran step two of our own advice on this topic before writing this page, and the result is worth showing rather than describing. In July 2026 our site appeared in Google’s results for 22 different search rows carrying the word seed. Together they drew 603 impressions and zero clicks. We had never written a page about it.
| What people typed | Impressions |
|---|---|
| how to find seed keywords | 251 impressions |
| seed keywords | 107 impressions |
| service-led seed keywords | 67 impressions |
| seed keywords tool | 43 impressions |
| seed keywords list | 12 impressions |
| what are seed keywords | 11 impressions |
The largest single phrasing was “how to find seed keywords” at 251 impressions, sitting at an average position of 40.3, which is around the fortieth result. The bare two-word version, “seed keyword”, drew 10 impressions at position 23.9. Both numbers say the same thing in different ways: Google had decided our site was somewhere in the neighbourhood of this question, and no reader had any reason to pick us out of the list. For context, the property averaged position 35.8 across all queries that month, on 28,411 impressions and 44 clicks in total.
That is what a seed with evidence attached looks like: a question your own site is already being shown for and losing. A big estimate beside a word tells you none of that.
One other thing in the same export, worth a paragraph and no more
Seven of the query rows in that July export are not searches a person typed. Each one carries an assistant’s own instructions in front of the question, in the form “context: location: united states (not for language). do not include location references in your response. question:” followed by the actual question. The questions behind them are things like how to find which new products are gaining popularity, and which keyword research tools help spot seasonal demand.
Their average positions were 3, 4, 6, 6, 7, 8 and 8. The property as a whole sat at 35.8 that month. So the pages a machine found for us were near the top of the result set, while the same site sat in the twenties, thirties and forties for the people asking about seed keywords in their own few words.
We are not drawing a trend out of that. It is seven rows, each with one or two impressions, in one month, on one site. What it is useful for is the shape of the wording. Those machine-written questions are long, complete and plainly worded. So is a customer’s email. So is the thing somebody says on the phone. The odd one out is the bare two-word version, the kind of short phrase known in keyword research as a head term. The list you built in the last section is already made mostly of the long plain kind.
When a good seed is the wrong seed for you
No tool will tell you this, because no tool knows what you sell or who to. Skipping it is what sends small firms chasing words that will never pay. A seed can be genuinely popular and still be the wrong one for your business, in three ways that all look identical in a tool.
Wrong country. The searching is real and it is happening somewhere you do not sell. If you install air-conditioning in Bukit Merah, a word with a large worldwide figure beside it can have very little behind it inside Singapore. Check that the figure in front of you is set to Singapore before you let it decide anything.
Wrong budget. The people typing it are shopping at a price you do not charge. Say a phrase like “cheap website design” comes back with a healthy figure. If your smallest build runs to five figures, every one of those readers costs you a quote and none of them becomes a client.
Wrong reader. The searcher will read the whole page and was never going to buy a service in the first place, because they are doing it themselves. This is the one that is hardest to spot, because the traffic arrives and looks like a success.
So the test is not the number. The test is whether the person typing that phrase could plausibly become your customer. That is a judgment about your own business, it takes seconds per word once you think to ask, and no tool will make it for you.
By that test, the 603 impressions we showed you are a weak seed for us. We build and look after websites and set up Google Workspace and Microsoft 365. We do not sell keyword research, and a fair share of the people asking this question are doing their own SEO precisely so they do not have to hire anybody. We wrote the page anyway, because the question sits next door to work we do and we had first-party data that answers it better than a page with none. That is the trade, and it is worth naming out loud rather than pretending every high-impression word is a prize.
If you have no traffic and no budget
Two of the three checks above quietly assume something. Search Console assumes your site already appears for something. Keyword Planner assumes a Google Ads account with its setup and billing details finished. A firm in its first year, with a new site and no advertising, has neither, and that reader needs a different answer.
The answer is that the five steps at the top of this page do not need either one. They need a sent folder and an hour. If you are newer than that, and even the sent folder is thin, there is a way to borrow other people’s wording honestly: ask people to type it for you.
A free tool called Seed Keywords does exactly that. In its own words: “Simply create a search scenario and ask your contacts to type in the keywords they would use to solve it.” You write the situation, send the link to people, and collect the words they would actually have searched. It costs nothing and it was built by Red Evolution, a UK industrial marketing agency, which the site says on its own front page.
The version with no tool at all is the same idea by hand: write the situation in one sentence, send it to ten people on WhatsApp, and ask what they would type into Google. Watch particularly for the case where they name the thing and you have been naming the category, or the other way round. Ten replies is a small sample. Treat what comes back as a prompt to go and check. It is still ten more real phrasings than a blank page, and it is the first step in this piece done the hard way when your sent folder cannot do it for you.
Questions people actually ask about this
The four below come from the phrasings our own Search Console recorded for this topic in July 2026, put into question form. They are what people were looking for, not questions we made up to fill a section.
How do you choose your seed keyword?
Start with the service you most want more of over the next three months, and that you have the capacity to take on if it works. Then pick the phrase from your list that names it. Choosing by size of the estimate is how a firm ends up ranking for work it cannot deliver, or does not want.
What is the difference between seed keywords and keyword ideas?
The seed is the part you chose. The ideas are what came back. An ideas list looks valuable because it is long, but most of it was generated for a business that is not yours, and it has no memory of who you sell to. The seed is the only part of the exercise that carries your judgment in it, which is why it is worth an hour of your own time.
How long should a seed keyword list be?
As long as the number of distinct things you sell, plus the questions people ask before they buy. For a small firm that is usually a short list rather than a spreadsheet. Our own rule of thumb, which is judgment and not a measurement, is that past about thirty entries you have stopped listing seeds and started listing phrasings of the same three.
Is brainstorming seed keywords enough on its own?
Brainstorming on a blank page gives you the words your trade uses internally, which is the vocabulary least likely to match a customer. The five steps above are not brainstorming. They are reading things other people already wrote to you. That difference is the whole method, and the free checks then catch what the reading missed.
Your list is not finished until you know where each word came from
That is the one test worth keeping. Go down the list and say, out loud, where each phrase came from: this one was in a rejected quote, this one is what three customers called it on the phone, this one came back in autocomplete. A phrase with no answer is a phrase somebody else put there, and it should come off.
The next question is a different one: whether people are still searching for those words, and whether they search all year or only in March. That is what Google Trends is for, and we have written up how to read it in how to use Google Trends for SEO. Take your finished list there before you write a single page.