What keywords should I target for my business?

Most keyword selection starts with a volume-sorted export and works downward until the numbers stop looking impressive. That method reliably produces a list of terms the site will never rank for, targeted by people who searched for something adjacent to what you sell.

The right keyword is not the one with the most searches. It is the one you can win, where winning produces revenue, and where winning makes the next one cheaper. Four filters, applied in order, get you there.

Why volume is the wrong start

Search volume tells you how many people search a term. It tells you nothing about whether those people buy anything, whether you could ever outrank the sites currently there, or whether the traffic would arrive expecting something you do not offer.

It is also less reliable than it appears. Every tool models volume from clickstream samples, figures vary substantially between providers for the same term, and conversational search has dispersed a great deal of demand into phrasings too long and varied to be reported at all. A term showing zero volume is frequently not a term nobody searches — it is one nobody has measured.

Treat volume as an order-of-magnitude signal, not a forecast. The difference between 50 and 500 matters. The difference between 480 and 520 does not.

Filter 1: relevance to revenue

For each candidate term, answer plainly: could a person searching this become a customer, and how soon?

Sort candidates into three groups. Ready — they are choosing a solution now: “best X for Y”, “X pricing”, “X alternatives”, “X vs competitor”. Researching — they have the problem and are learning: “how to solve Y”, “why does Y happen”. Adjacent — related to your field, unrelated to your business: definitions, student research, general curiosity.

The third group is where most keyword strategies quietly go wrong, because it is where the volume lives. A term can be topically perfect and commercially worthless. Ranking first for a definitional query aimed at people writing coursework produces a traffic chart that goes up and a pipeline that does not move.

This does not mean ignoring informational terms — they build authority and they capture people before your competitors reach them. It means knowing which group each term is in, and not mistaking group three volume for a business case.

Filter 2: winnability

Could you realistically reach the top five within two quarters?

Judge this per query by looking at what actually ranks, not by a site-wide difficulty score. Difficulty scores average across an entire term and hide the specific thing you need to know: who is in positions one to five, and what holds them there?

• If the top five are national publishers, marketplaces or Wikipedia, the term is informational and gated by authority you do not have. Not winnable this year.
• If the top five are companies like yours with far more history, it is winnable eventually, not soon.
• If the top five include forums, thin affiliate pages or outdated posts, it is winnable now. This is the pattern worth hunting.
• If the top five are genuinely excellent and well-backed, go more specific. The term you can win is usually two or three qualifiers deeper.

Label every term now, with effort, or not yet, and park the last group explicitly rather than carrying it silently. Most keyword lists have never been through this filter, which is why most content plans contain a quiet majority of terms that were never realistic.

Filter 3: intent match

Can you produce the kind of page the results page is already rewarding?

Search the term and look at the format of the top ten. If they are all comparison articles and you intend to publish a product page, you will not rank however good the product page is. Google has published its opinion about what the query means, and the format is part of that opinion.

Three mismatches are worth checking for specifically. Format: list, guide, tool, product page, or video — the dominant format is not optional. Depth: if the ranking pages are 3,000-word buying guides, a 600-word answer will not compete, and vice versa. Perspective: if every ranking page is a neutral third-party comparison, a vendor page arguing for its own product faces a structural disadvantage no amount of optimisation fixes.

Where you cannot or will not match the format, the honest conclusion is that this is not your term. That is a useful conclusion — it saves a quarter.

Filter 4: compounding

Does ranking for this make the next term easier?

This is the filter most often skipped and it matters most over a year. Authority is topical, not just site-wide. A site known for one subject can extend into adjacent subjects far more cheaply than into unrelated ones, which means the shape of your keyword portfolio determines how expensive next quarter is.

Twenty related terms covered thoroughly build a topical footprint that makes the twenty-first almost free. Fifty scattered terms across unrelated subjects build nothing, compete with each other for internal linking, and leave you starting from zero every time.

In practice: pick a cluster you could plausibly become the best resource on, cover it completely, then move to the adjacent cluster. Resist the pull toward breadth, which always feels like progress and rarely is.

Building a portfolio, not a list

A finished selection is not twenty terms. It is a small portfolio with deliberate proportions.

The proportions matter more than the exact numbers. A portfolio that is entirely commercial terms stalls, because those are the hardest and you have no authority to attack them with. One that is entirely informational produces a beautiful traffic chart and no pipeline.

The keywords that are not keywords

One category now sits outside the keyword list entirely and belongs in the same planning exercise.

When a buyer asks an assistant “what’s the best X for Y”, no search is recorded on a results page and no keyword tool sees it. But a shortlist gets made, and if you are not on it you never enter the funnel. These are prompts rather than keywords, and the ones where competitors are named and you are not are prompt gaps.

Building that list is straightforward: write the twenty questions a real buyer would ask about your category, run them through ChatGPT, Gemini and Perplexity, and record who is named and what is cited. The questions where you are absent become content briefs, exactly like keyword gaps do. Our guide to increasing AI citations covers what to do with them.

Validating the shortlist before you commit

Before a term earns a quarter of writing capacity, spend ten minutes on three checks. They catch most of the selections that would have failed.

Search it and read the top five properly. Not the titles — the pages. You are looking for whether a reader would consider the question answered. If they would, you need something genuinely additional, and you should know what it is before you start rather than discovering mid-draft that you have nothing.

Check whether you already rank for it. Filter Search Console for the term. If a page is already at position 24, you have an improvement task rather than a publishing task — a completely different piece of work with a much shorter payback. Teams routinely publish a second page onto a term they already half-own, and then wonder why neither ranks.

Ask an assistant the same question. If ChatGPT answers it completely and names three competitors, you have learned two things: the query is genuinely commercial, and the competition for it now includes a surface where you may not appear at all. Both change how you write the page.

Five selection mistakes

Choosing terms because you like them. The words you use internally are frequently not the words customers search. Check the actual phrasing before committing.

Targeting the term your competitor ranks for, because they rank for it. Their authority, age and link profile may make it viable for them and hopeless for you. Winnability is relative to your site, not theirs.

Ignoring terms with low reported volume. Especially long, specific, conversational phrasings. These convert best, face the least competition, and are systematically under-reported by every tool.

Building the list once. Rankings drift, competitors publish, demand disperses. A selection older than about ninety days should be re-validated before you commit a quarter to it.

Confusing a keyword list with a content plan. Twenty terms are usually six to eight pages. Clustering by intent — verified by checking whether the ranking URLs overlap — is the step between the two, and skipping it produces a site full of near-duplicate pages competing with each other.

Sources

• Google Search Console — the only first-party source of the queries you actually appear for.
• Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande — “GEO: Generative Engine Optimization”, ACM SIGKDD 2024 (arXiv:2311.09735).
• SERP feature tracking, 2026 — AI Overview presence averaging ~32.5% of monitored queries within a 31.02%–34.40% band.

Frequently asked questions