Keyword research for a local client with AI

For a local business, keyword research answers one question: what do the people who would hire this client type into a search box, and does the site have a page that answers each of those? An assistant is a fast way to get a first list. It is a poor way to get numbers.

Never ask for search volume

An assistant has no live search data. If you ask how many people search a phrase each month, it will give you a number anyway, and the number will look believable. Do not use it. For real data, use Google Search Console, which is free: once a site has a few weeks of impressions it shows the phrases people actually found the site with. A keyword tool that reports volumes works too. Use the assistant for the wording, and real data for the size.

Start from the client brief

Paste the client brief from the last chapter first. Then ask for seed phrases, grouped by what the searcher wants to do. Customers rarely use the trade's own words, so ask for theirs.

Using the client brief above, list 40 phrases a {customer type} in {city} might type when they need {service}. Group them under these headings: ready to hire, comparing options, asking a question, emergency. Use the words customers use, not the trade's words.

Do not give search volumes or difficulty. You don't have that data. I will check the phrases against Search Console.

Delete the ones that are wrong for the business, then add phrases you know from the client's own calls and emails. Those are the best ones, and the assistant cannot see them.

Cluster by intent, one page per cluster

Two phrases belong on the same page when one page could answer both. "emergency plumber Springfield" and "burst pipe plumber Springfield" want the same page. "How much does a water heater cost" wants a different one. Giving each cluster one page stops a site competing against itself, and shows you which pages to write.

Here is my list of phrases:

{paste the list}

Group them into clusters where one page could answer every phrase in the cluster. For each cluster give me:
- a page name
- the main phrase, and the other phrases
- the intent: informational, local service, comparison or ready to hire
Flag any phrase that needs its own page. Flag any phrase that two clusters would compete for.

The output should be a keyword map you can scan in a minute. For a plumber in Springfield it might look like this:

This is an example, not data. The point is one row per page, with the phrases that page has to answer. A phrase with no row has no home, which is a page to build or a phrase to drop.

Find the gaps

Now set the clusters against what the site already has. Export the page list with titles from your audit or crawler and paste it next to the clusters.

Below are the keyword clusters, and then the pages that exist on the client's site today with their titles.

Clusters:
{paste clusters}

Existing pages:
{URL | title, one per line}

Tell me which clusters already have a page, which have a page that is a poor match (say why), and which have no page. Give me the missing pages in the order I should build them, and name the existing page each new one should be linked from.

The output is a short keyword map: cluster, target page, main phrase. Keep it with the client's record. The next two chapters use it to write titles, headings and page content, and in the monthly report you can show which pages you built.

Check it against real data

Once the site has been live a month or more, open Search Console, go to Performance, and look at the queries. Impressions without clicks mean you appear but the title is not earning the click, which is chapter 4. Queries you did not expect are phrases to add to the map. A phrase from the assistant's list that never appears is not proof it is useless, but it is a prompt to ask whether anyone really searches it.

Don't bother with

• Hundreds of long-tail phrases. A local business needs a few dozen clusters at most. A thousand generated phrases is a thousand things to ignore.
• Keyword density. There is no ideal percentage. Write for the reader and use the phrase where it belongs.
• One page per suburb with the name swapped. Pages that differ only by place name are near duplicates and add little. Build one where you have something real to say about the place.

Next

• Chapter 4: Titles and meta descriptions, in bulk
• Keep the keyword map in the client record
• Google's SEO Starter Guide