Across an analysis of 306 million Google keywords published by Backlinko in December 2020, the median search term gets ten searches a month. Not ten thousand. Ten. The right keywords for a small business are the specific, buyer-intent phrases your customers already use, not the highest-volume terms in your industry. Start with words from real customer enquiries, expand them into longer local and commercial variations, then check each one against competitor pages before assigning it to a single page on your site.
We run this process on client sites every week, and on our own. What follows is the working method rather than the theory: where to find your starting words, how to extend them, which ones still earn a click now that AI answers sit above the results, and how to assign them so two of your pages never end up competing for the same term.
Why your own service words make poor keywords
The words you use to describe your service are unlikely to be the words your customers type into Google. That’s not a hunch about marketing. It’s a measured property of human language.
In November 1987, four researchers at Bell Communications Research published a paper in Communications of the ACM called The vocabulary problem in human-system communication. George Furnas, Thomas Landauer, Louis Gomez and Susan Dumais ran a series of studies asking people to name the same object or action across five different domains. The probability that any two people chose the same word came out below 0.20. Their conclusion was that no single term, however carefully selected, covers more than a small fraction of the words real users will try.
That paper predates Google by eleven years, which is exactly why it holds up. Search engines got much better at bridging the gap through synonym matching and semantic understanding. The gap itself never closed. It moved.
Every business owner who writes their own website copy is running that 1987 experiment on themselves, and losing at roughly the same rate. You call it ‘bespoke financial planning’. Your customer types ‘help sorting out my pension’. Both descriptions are correct. Only one of them exists in search demand.
Susan Dumais, incidentally, went on to co-invent latent semantic indexing, the technique that taught machines to connect words with related meanings. The people who identified the problem spent the next thirty years building the tools that partially solve it. Partially is the operative word.
What are seed keywords and where do they come from
A seed keyword is a short, obvious phrase that describes what you sell, used as the starting point for generating longer and more specific variations. That’s the standard definition, and it’s where most guides leave it.
Here’s the more useful version: a seed keyword isn’t a word you invent, it’s a word you collect. The difference decides whether the rest of your SEO work pays off, because everything downstream inherits the phrasing you start with. Collect from three places: the subject lines of enquiry emails you’ve received, the phrases customers use on the phone before they know the technical term, and the words your competitors’ customers leave in reviews.
Where your best keyword ideas already exist for free
Every small business owns a keyword dataset it has never opened. Your inbox holds it.
The usual advice is to open a keyword tool first. Try reversing that. A tool tells you what a large, anonymous population searched for last month. Your enquiry emails tell you what somebody wrote when they had a problem and money to spend on solving it. One is a sample of the market; the other is a sample of your market. Start with yours, then use a tool to check volume afterwards rather than to generate the ideas in the first place.
Google itself hands over the rest for nothing. The same Backlinko study of 306 million keywords found that People Also Ask boxes appear in 19.5% of search results, and only 2.4% of results contain no SERP feature at all. Those boxes are a live question dataset that Google regenerates for every query. The same research found that 14.1% of all searches are questions, with ‘how’ leading at 8.07%.
You are looking at free customer research that a market research firm would charge four figures to assemble.
How to do keyword research without paying for tools
Four sources cover most of what a paid subscription generates, and each one fixes a weakness in the others.
- Google Search Console shows the exact queries your site already appears for, including phrasing you’d never have guessed at
- Google autocomplete reveals what people commonly type next, which surfaces intent your own vocabulary would miss
- People Also Ask boxes give you the question format that your FAQ and H3 headings can answer directly
- The H2 headings on pages currently ranking for your target term show you what Google has already decided that query means
Work through all four for a single seed term and you’ll finish with thirty to fifty candidate phrases. That’s a keyword list, and it cost you an afternoon.
How to read Search Console for hidden keyword wins
Most people open Search Console and look at the queries where they rank first. Those are the terms you’ve already won, so there’s limited work to do there.
Filter instead for queries where your average position sits between eight and twenty. Those are searches where Google already considers your page relevant enough to include, but not strong enough to promote. Improving a page for a term it nearly ranks for is faster and cheaper than building a new page for a term it doesn’t rank for at all. A clearer heading, a paragraph that answers the query directly, and a tighter page title will often move a position fourteen to a position six.
Having said that, Search Console withholds a share of rare queries to protect user privacy, so the report you’re reading is a floor rather than a complete picture. If a term you’d expect to see is missing, check autocomplete before concluding nobody searches for it.
How long-tail keywords work for smaller businesses
Long-tail keywords aren’t a consolation prize for businesses too small to compete for the big terms. They’re the overwhelming majority of search, and they’re the portion of it you can realistically own.
Backlinko’s keyword analysis found that 91.8% of search queries are long tail, defined in that study as terms attracting under 100 searches a month. The same dataset put the mean monthly volume at 989 and the median at 10, with the average keyword running to just 1.9 words.
Most articles quote the 91.8% figure and stop there, which produces genuinely bad advice. The honest half of that finding is the other number: all those long-tail terms combined account for only 3.3% of total search volume. No single specific phrase will carry your business. Thirty of them, working together across a properly structured site, will.
That reframes the whole exercise. You aren’t hunting for a keyword. You’re building a list.
Are low-volume keywords actually worth targeting
Yes, when the intent behind them is strong.
Use the median as your reference point whenever a tool shows you a discouraging number. If the typical keyword in a 306 million term dataset gets ten searches a month, then a term showing forty searches isn’t a failure; it’s four times more common than average. The volume column in your tool is calibrated to the expectations of enterprise marketers, not to the reality of a service business with a twelve-page website. Our essential SEO guide for new businesses works through how those early priorities stack up.
If the term you want is searched five thousand times a month, and nine of those searchers are ready to buy, which number were you chasing?
Which keywords still earn clicks in AI search
Keyword selection now needs a second filter. A term can rank perfectly well and still send you nothing, because the answer sits above the results in an AI summary. The keywords that survive are the ones an AI answer can’t resolve on its own.
The Pew Research Center published the clearest measurement of this on 22 July 2025. Researchers tracked 68,879 real Google searches made by 900 US adults during March 2025 using browsing data from a representative panel. When an AI summary appeared, users clicked through to a traditional search result on 8% of visits, against 15% when no summary was present. Just 1% clicked a link inside the summary itself.
The same study found something that looks, at first, like an argument against everything in the previous section. AI summaries appeared for 8% of one and two word searches, rising to 53% of searches ten words or longer. Question-shaped queries triggered a summary 60% of the time. Longer queries invite AI answers.
Then a randomised field experiment resolved the tension. Saharsh Agarwal of the Indian School of Business and Ananya Sen of Carnegie Mellon’s Heinz College recruited 1,065 US Chrome users and randomly hid AI Overviews for some of them using a browser extension, observing 68,089 searches between 7 January and 10 February 2026. Outbound organic clicks fell 39.8% when an overview was shown. Critically, informational queries made up 71% of the sample and absorbed the entire effect. Transactional and navigational queries showed no statistically significant impact at all.
So length was never the variable. Intent was.
A ten-word question about how something works gets answered on the page, because the answer is a fact. A ten-word phrase containing a location, a price, or an urgent job gets no useful AI answer, because the answer isn’t a fact; it’s a supplier who can be at the property on Thursday. Small businesses are better positioned in this environment than most publishers, and very few have noticed.
Google publicly disputed Pew’s methodology when the study was released, and the Agarwal and Sen paper is a working paper rather than a peer-reviewed publication. Both remain the strongest evidence available, and both come from researchers with no product to sell.
Should you target keywords that trigger AI overviews
Target them for visibility, not for traffic. Being cited in an AI answer builds recognition and reaches people who’d never have scrolled to your listing, which is worth having.
Just don’t build your enquiry pipeline on it. Balance every informational term in your list with a local or commercial one, so that the pages carrying your revenue sit in the query types the 2026 experiment found unaffected. We covered the citation side of this in detail in our piece on winning visibility in AI search engines.
How to judge keyword competition before you commit
Competition isn’t a number in a tool. It’s whatever currently occupies page one, and you can assess it by reading.
Run the search and count what you find. If four of the ten results are directories, aggregator sites, and forum threads, the term is winnable by a well-built page. If eight are national brands with dedicated landing pages and decade-old domains, it isn’t, whatever a difficulty score tells you. Five minutes of reading beats a subscription here, because you’re evaluating the actual competitors rather than a model’s estimate of them.
The difficulty scores do carry useful information about the trade-off, though. Backlinko’s analysis found that each doubling of monthly search volume raises Ahrefs keyword difficulty by around 1.63 points, so moving from 100 to 3,200 monthly searches adds roughly ten difficulty points. Reach and winnability move against each other in a fairly predictable way.
Worth knowing who’s counting, mind. Difficulty scores are vendor estimates built largely on backlink models, and Backlinko is owned by Semrush, which sells keyword tools. The raw dataset from that study is published openly, which is more transparency than most vendor research offers, and the findings hold up. It’s still information you should hold with the source attached.
How to map keywords to pages that convert
One page, one primary keyword. Most small-business ranking problems aren’t keyword problems at all; they’re mapping problems, where three pages chase the same term and Google can’t decide which one to rank, so it under-ranks all three.
The full process runs to five steps.
First, collect your seed terms from customer language rather than your own service names. Second, extend each seed using Search Console, autocomplete and People Also Ask until you have thirty to fifty candidates. Third, filter by intent, separating the informational terms from the commercial and local ones. Fourth, read page one for every remaining candidate and cut anything dominated by national brands. Fifth, assign exactly one primary keyword to each page, with two or three close variations supporting it.
That fifth step is where the discipline pays. It also tells you when you need a new page rather than another paragraph.
How many keywords should a small business target
There’s no authoritative figure, and anyone quoting one is giving you a preference rather than evidence. Recommendations circulating online range from five starter keywords to two hundred across a site, with nothing underneath them.
The number is set by your site, not by best practice. One primary keyword per page, plus two or three variations, means a ten-page site works with around ten primary terms. If you want to target thirty, you need thirty pages worth writing, and each one has to be good enough to keep updated. Our content packages exist for exactly that reason, because the constraint is rarely the research; it’s the writing.
How to choose between local and national keywords
Take a heating engineer in Elgin. The national term is ‘plumber’. The local, transactional term is ’emergency boiler repair Elgin’. The first has vastly more volume and belongs to national directories. The second belongs to whoever builds the clearest page for it.
The intent difference is what settles the argument. Someone typing ‘plumber’ might be a student writing an essay. Someone typing ’emergency boiler repair Elgin’ has cold water, a postcode, and a decision to make today. That query also has the exact shape the 2026 field experiment found unaffected by AI Overviews, because no summary can dispatch an engineer.
Scale still favours getting this right. Ofcom’s Online Nation 2025 report, published on 10 December 2025, recorded Google Search reaching 82% of UK online adults and handling around three billion UK searches a month, with roughly 30% now returning an AI overview. A small share of three billion is a substantial number of people, and the structure of your site decides whether you’re visible to any of them. That structural question is the one we work through in our guide to building your small business website once.
If you had to delete every page on your site except the three that answer a customer’s most specific question, which three would survive?
The keywords your competitors keep ignoring
The 1987 vocabulary research and the 2026 field experiment describe the same problem forty years apart. There’s a distance between how a business talks about itself and how a customer asks for help, and every technique here is a method for measuring or closing that distance. The researchers found people agree on wording less than a fifth of the time; the field experiment found the queries that still send clicks are the ones phrased around a real, immediate need. Same finding, different decades.
Ofcom’s annual report also recorded ChatGPT drawing 252 million UK visits in August 2025, up 156% year on year. Search demand is spreading across more surfaces than Google alone, and the businesses described in specific, transactional, human language are the ones those surfaces can recommend with any confidence. Vague positioning was survivable when there was one search box. It’s expensive now.
So before you open a keyword tool, open your last twenty enquiry emails and write down the words your customers used.
Let Us Handle Your Digital Marketing
Common questions about small business keyword research
How many keywords should a small business target?
Target one primary keyword per page, plus two or three close variations. A ten-page website works with roughly ten primary terms. No authoritative research sets a universal number, and sources quoting fixed figures between five and two hundred are stating a preference rather than a finding. The practical limit is how many pages you can write well and keep genuinely up to date.
What are seed keywords?
Seed keywords are the short, obvious phrases describing what you sell, used to generate longer and more specific variations. The most useful ones come from customer language rather than your own service names. Look at enquiry emails, notes from sales calls, and the words people use to describe their problem before they know what the industry calls the solution.
Are low-volume keywords worth targeting?
Yes, when the intent is strong. Backlinko’s analysis of 306 million keywords found the median term gets ten searches a month, so low volume is the normal condition of search rather than a sign of failure. Ten searches from people ready to book beats five thousand from people browsing definitions. Build a portfolio of thirty specific terms instead of chasing one large one.
How do I do keyword research without paying for a tool?
Use Google Search Console to find queries where you already rank between positions eight and twenty, then extend them with Google autocomplete and People Also Ask boxes, which appear in 19.5% of Google results. Read the headings on pages currently ranking for your target term. Those four free sources cover most of what a paid subscription generates.
Should I target keywords that trigger AI overviews?
Target them for visibility rather than traffic. Pew Research found that when an AI summary appeared, users clicked a search result on 8% of visits against 15% without one, and only 1% clicked a link inside the summary. Balance those terms with local and transactional keywords, which a randomised 2026 field experiment found were not significantly affected.