How to choose keywords for a brand new website

Almost all keyword research advice is written for a website that already exists in Google's index with some history behind it. Find terms with decent volume and low difficulty, write something better than whatever ranks now, wait. On a domain that went live three weeks ago, that advice quietly fails. The missing ingredient is not content quality. It is time.

There is no formal sandbox, and Google has denied running a deliberate holding pen for new sites. What is observable anyway is that new domains take months to rank for anything competitive, and often longer than that. So the useful question for a new site is not which keywords are best. It is which keywords you can win while nobody, including Google, knows who you are yet.

The two-minute version

1. 1Accept that your industry's head term is a twelve-month-plus project. Do not build the site around it.
2. 2Build a first list of roughly ten terms, not two thousand.
3. 3Compose it deliberately: 3-4 service plus location terms, 3-4 specific-situation terms taken from questions real customers ask, 2-3 comparison or decision terms, and exactly one ambitious term you accept as a two-year project.
4. 4Put every candidate through two filters. Does someone searching this have a problem you are paid to solve, and does the current first page contain at least one site no bigger than yours.
5. 5Group the survivors into pages before writing a word. One strong page beats four thin ones aimed at near-identical phrases, and it avoids keyword cannibalization later.
6. 6The moment the site gets any impressions at all, switch your source of truth to Search Console and rebuild the list from what Google already shows you for.

Why a new site cannot simply outrank the incumbents

Ranking is comparative. For a query with real commercial value, the pages currently on page one have usually been there for years, have picked up links and mentions in that time, and have accumulated a history of people clicking them and not bouncing straight back. Your new page has none of that. It has words on it. Words are necessary and not sufficient.

The practical consequence is that your first wins will come from queries where the competing set is weak, not from queries where your content is strong. Those are two different things, and confusing them is the most common reason a new site publishes forty articles in three months and gets nothing. If you have already published and nothing is moving, why your website is not ranking on Google covers the other likely causes.

Filter one: is there a paying problem behind the search?

Volume is the wrong first question. The first question is whether the person typing this phrase has a problem you get paid to solve. A lot of high-volume terms in any industry are searched by students, competitors, and people doing it themselves for free. Ranking for them produces traffic and no revenue, which is its own kind of frustration.

• Good sign: the phrase names a service, a decision, a price, a location, a deadline, or a symptom of something breaking.
• Bad sign: the phrase names a definition, a template, a free download, or a school-assignment topic.
• Bad sign that looks good: broad industry terms with the highest volume on your list. They are usually researched by people nowhere near a purchase, and they are the hardest to reach.
• Test it out loud. If a customer said this sentence to you on the phone, would you be able to quote them? If yes, keep it.

Filter two: is anyone on page one your size?

Open a private window, search the phrase, and look at who is actually there. Not a score. The actual ten results. Then classify them: national brand, marketplace or directory, forum thread, large publisher, or a small business roughly like yours. If at least one result is a small independent site, the query is contestable. If all ten are aggregators and national brands, park it and move on, no matter how good the numbers looked in a tool.

This manual look beats any difficulty number, because difficulty is a vendor metric rather than a Google one and every tool computes it differently. It is worth understanding what keyword difficulty actually means before you let a two-digit score decide your content plan. And if you want the underlying method for spotting soft SERPs, how to find low competition keywords goes deeper than this section does.

What the first list actually looks like

Ten terms. Four slots. This composition exists because each slot does a different job: two of them can win this quarter, one builds trust with people who are close to buying, and one is a long bet you fund with the others.

Service plus location

This is where a new business gets its first real enquiries. The searcher has already decided to hire someone and is choosing between the options in front of them. Build one genuine page per service and per place you actually serve, with real details: the areas covered, what the work involves, what it costs or how pricing is structured, who did it before. Do not generate twenty city pages for cities you have never worked in. That pattern reads as thin doorway content and it does not hold up.

Specific situation

Take the questions customers ask you in the first ten minutes of a call and write them down verbatim, before you tidy the language. Those sentences are the closest thing you have to query data. A phrase like why does my site look fine on desktop and broken on my phone is unglamorous, has modest volume, and can be won by a two-month-old site because the people ranking for it wrote something generic.

Comparison and decision terms

People about to spend money compare. Cheap versus expensive, DIY versus hired, platform A versus platform B, do it now versus wait. These pages convert unusually well and are usually poorly served, because the sites ranking for them have an obvious stake in one answer. Publish the version that tells the reader when not to buy from you. It costs you a few unqualified leads and wins the qualified ones.

The one ambitious term

Keep one. Write it at the top of the document with a date twenty-four months out next to it. Its purpose is psychological: naming it as a long project stops it from silently becoming the benchmark you fail against every month, and it tells you which of your smaller pages should eventually point at it.

Where candidates come from before you have any data

With no Search Console history, you are not doing analysis yet. You are collecting raw language, and the best sources are places where people describe problems in their own words.

1. 1Customer and sales conversations. Sent messages, WhatsApp threads, enquiry form submissions, the questions asked before someone paid. Highest quality source you will ever have, and you already own it.
2. 2Competitor sitemaps and navigation. A competitor's sitemap is a list of every topic they thought was worth a page, and their menu shows which ones they consider commercially important.
3. 3Google autocomplete. Type your service and go through the alphabet after it. Type it as a question. These suggestions come from real searches.
4. 4People Also Ask. Expand a few boxes and it keeps generating related questions. Collect the wording, not just the topic.
5. 5Forums, subreddits and industry groups. Look for posts where someone is stuck and describing it badly. That bad description is the query.
6. 6Review language. Read one-star and five-star reviews of your competitors. What people praise and complain about is what they search for next time.

Group the list into pages, not a page per phrase

Ten terms does not mean ten pages. Several of your phrases will be the same question worded differently, and Google treats them as the same intent. Cluster them first, then decide how many pages you need. Usually it is fewer than you expected, and each one gets to be substantial instead of thin. Which pages a small business website actually needs covers the structure this tends to settle into.

What to expect, month by month

These windows are shapes, not promises, and they assume the site is technically sound and genuinely useful. A slow, badly indexed site adds months to every row.

What not to do

• Do not buy a 2000-keyword list. It is an export, not a strategy, and 95 percent of it will be terms you cannot reach or would not want to. Someone still has to apply the two filters by hand, and that someone is you.
• Do not build one thin page per phrase. It splits your own signals across near-duplicate pages, and at scale it starts to resemble what Google's spam policies call scaled content abuse: many pages made mainly to catch rankings rather than help anyone. The policy is method-agnostic, so hand-written thin pages count too.
• Do not chase your biggest competitor's head terms. Their rankings are the output of years of history you do not have. Copying their target list is copying the answer key to a different exam.
• Do not tune keyword density. There is no density target, Google does not use a meta keywords tag, and time spent counting repetitions is time not spent answering the question.
• Do not judge a term by volume alone. A phrase with modest monthly searches and buying intent is worth more than a broad term you will never reach.
• Do not rewrite the list every fortnight because nothing ranked yet. Three months is not a result.

Switch to Search Console as soon as you have data

Every source above is a substitute for data you do not have yet. Once the site gets impressions, Search Console (opens in a new tab) becomes the best keyword tool available to you, because it reports the queries Google already associates with your pages rather than someone's model of them. Look for queries where you appear on page two or three, and for queries you never targeted at all. Those are the cheapest wins on the site.

One caveat worth knowing: the AI search performance reports added to Search Console in June 2026 show impressions, pages, countries, devices and dates, but no click data. So you can see that AI surfaces are showing your pages. You cannot see what that traffic is worth yet.

On a new site, the right keyword is not the best one. It is the best one you can actually reach this year.

Questions people actually ask

How many keywords should a new website target in total?

Around ten to start with, mapped onto a handful of pages. Not because ten is magic, but because it is the number one person can genuinely serve with substantial pages in the first few months. Adding the eleventh before the first four pages are good is how sites end up with fifty pages and no rankings. How many keywords per page covers the mapping side of this.

How long before a new site ranks for anything?

For very narrow, specific phrases with weak competition, weeks to a couple of months once the page is indexed. For local service terms, usually a few months. For anything genuinely competitive, plan in quarters and expect the first year to be about building the case rather than collecting the reward. Anyone promising faster on a competitive term is either buying ads or guessing.

Should I ignore high-volume keywords completely?

No, record them and do not build the site around them. Write them in a second list dated a year out. When your smaller pages start ranking and the site has some history, those terms stop being fantasy and become the obvious next project, with the earlier pages available to link to the new one.

Do I need a paid keyword tool to do this?

Not for the first list. Customer conversations, autocomplete, People Also Ask, competitor sitemaps and looking at page one yourself will get you ten good candidates. Paid tools earn their money later, when you have enough pages that prioritising between hundreds of options is genuinely hard.

What if my business has no location angle at all?

Replace the location slot with a narrower audience or use case: the service for a specific industry, a specific platform, a specific company size, a specific regulation. The mechanism is the same. You are trading breadth for a first page where somebody your size already survives.