Keyword Research: The 2026 Guide for Google and AI Search

Keyword research is the process of finding the words and questions people type into Google and AI assistants, then choosing the ones your site can rank for and that lead to customers. Done well, it tells you which pages to create, what each page must cover and which page to write first.

Key takeaways

• Start from your customers' problems, then expand with free Google tools; relevance comes before volume.
• Judge every keyword on four checks: search volume, difficulty, search intent and business value.
• Give each search intent one page, and group close keywords into topic clusters so your pages don't compete.
• AI search breaks one question into several sub-searches, so long-tail questions matter more in 2026.
• Before you write, reverse-engineer the terms the top-ranking pages use and cover the same ground.

What is keyword research?

Keyword research is the work of discovering what your audience searches for, measuring the demand and competition behind each search, and deciding which searches your pages should answer. The result is not a list of words but a plan: each row is a topic with one primary keyword, its secondary keywords, the intent behind it and the page that will target it.

Google's SEO Starter Guide puts the idea in one line: "Think about the words that a user might search for to find a piece of your content." It adds that people who know a topic well search differently from newcomers, and gives the example of one person typing "charcuterie" while another types "cheese board". Keyword research is how you find out which words your readers actually use, instead of guessing from your own vocabulary.

A keyword is any search, short or long. "Keyword research", "how to do keyword research for a new website" and "is keyword research still worth it with ChatGPT" are all keywords. The longer ones usually have fewer searches and a clearer intent.

Why is keyword research important for SEO?

Keyword research for SEO matters because it decides what you build before you spend hours building it. Three things go wrong without it:

1. You write pages nobody searches for. A well-written page on a topic with no demand gets no search traffic.
2. You pick fights you can't win. A new site that targets the same broad term as large, established publishers can wait a long time for page one, while a more specific version of the topic is within reach.
3. You build the wrong kind of page. If searchers want a comparison and you publish a product page, Google has little reason to rank it.

There is also more room in the long tail than most people assume. In its BERT announcement (October 2019), Google said that 15 percent of the queries it sees every day are ones it hasn't seen before. Ahrefs reports that keywords with fewer than 10 searches per month make up almost 93% of its U.S. keyword database (Ahrefs, updated May 2026). Most of the keywords people search are specific, and specific searches are where smaller sites win.

How does keyword research change for AI search?

Keyword research for AI search starts in the same place, because AI answers are built from search results. What changes is that one question now triggers several searches, so the sub-questions around your topic matter as much as the head term.

Google says AI Overviews and AI Mode "may use a 'query fan-out' technique — issuing multiple related searches across subtopics and data sources — to develop a response." Gemini works the same way when it searches: Google's grounding documentation says the model "automatically generates one or multiple search queries and executes them" before it writes the answer. ChatGPT works from the web too: OpenAI says its OAI-SearchBot crawler "is used to surface websites in search results in ChatGPT's search features."

The effect shows in citation data. In an Ahrefs study of 863,000 keyword results pages, about 38% of AI Overview citations came from pages ranking in the top 10 for the original query (Ahrefs, March 2026). Ahrefs suggests that more of the rest now come from the fan-out searches. In practice, a page can be cited for a question it doesn't rank in the top 10 for, if it ranks well for the sub-questions behind it.

Google also states there are "no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." There is no separate list of "AI keywords" to find. You do the same research, with a wider net.

How to do keyword research in 10 steps

Here is how to do keyword research from a blank page to a prioritized content plan. The table gives the overview; each step is explained below.

Step 1: Start with seed keywords from your business

Seed keywords are the broad topics your business is about, written in your customers' words. Start here rather than in a tool, because a tool can only expand what you give it.

Good sources for seeds:

• Your products and services, named the way a buyer would name them, not the way your team does internally.
• Customer conversations: sales calls, support tickets, onboarding questions and reviews. The exact phrases people use are often the keywords.
• Google Search Console: if your site already gets traffic, the Performance report lists the queries that showed your pages, with clicks, impressions and average position.
• Communities: Reddit threads, niche forums and Q&A sites show how people describe a problem before they know what the solution is called.
• Competitors' menus and category pages: they show how others in your market organize the topic.

Aim for 10 to 30 seeds. A bakery-supply shop might start with "sourdough starter", "proofing basket", "bread flour" and "stand mixer for bread".

Step 2: Expand the seeds into keyword ideas

Keyword ideas come from letting search engines and tools show you what people type around each seed. Work through the sources below and paste everything into one sheet; you will filter later.

• Google autocomplete: type a seed and note the suggestions. Add letters ("sourdough starter a…") and question words ("how", "why", "can", "vs").
• People also ask and related searches: these boxes on the results page are questions Google has connected to your topic.
• Google Keyword Planner: its "Discover new keywords" option takes words or a website and returns related ideas (Google Ads Help).
• Search Console: filter for queries that contain your seed. Queries with many impressions but few clicks are often topics you touch but don't answer well.
• AI assistants: ask ChatGPT or Gemini which questions a beginner has about your topic. Treat the answer as brainstorming, not data, and check each idea against real search results.

Modifiers multiply the list: "best", "for beginners", "vs", "cost", "near me", "template", "example" and the year. Each one points to a different intent, which you will sort out in Step 5.

Step 3: Check search volume, and know its limits

Search volume is the estimated number of times a keyword is searched per month. Use it to compare keywords with each other, not as a traffic forecast.

The standard free source is Google Keyword Planner. Google defines its "Avg. monthly searches" as the average number of searches "for a keyword and its close variants", which by default "is averaged over a 12-month period" (About Keyword Planner forecasts). Two things follow. Seasonal keywords look flatter than they are, so check Google Trends for anything tied to holidays or weather. And close variants are pooled, so "keyword research tool" and "keyword research tools" won't give you two independent numbers.

Keep three more limits in mind:

• Volume is an estimate, and different tools show different numbers for the same keyword.
• Ranking first doesn't win every search: ads, AI Overviews and other results also take clicks.
• A 50-search keyword with buying intent can be worth more than a 5,000-search keyword with none.

Our guide to Google Keyword Planner for SEO walks through the setup and filters. One note for RankTera users: RankTera doesn't show search volume, so take volumes from Keyword Planner and use RankTera for the later steps.

Step 4: Judge keyword difficulty

Keyword difficulty is how hard it will be to reach the first page for a keyword. Third-party tools give it a score from 0 to 100, but each tool calculates it differently, so treat the score as a first filter and confirm it by looking at the results page yourself.

Don't confuse difficulty with the "Competition" column in Keyword Planner. Google defines that column as "the number of advertisers that showed on each keyword relative to all keywords across Google" (Google Ads Help). It measures competition for ads, not how hard the organic results are.

A manual difficulty check takes about two minutes per keyword. Search the keyword in the country you target and look at the first page:

• Who ranks? Household-name brands and big publishers in every slot mean a long road. Forum threads, small blogs or thin pages mean an opening.
• How good are the pages? Outdated, shallow or off-topic pages are beatable.
• Do the results match the page you would write? If every result is a product category page and you plan a blog post, the keyword isn't a fit for that post.

Sort keywords into three bands: winnable now, winnable in 6–12 months with supporting content, and long-term. Our keyword difficulty guide explains the scores and how to find low-competition keywords.

Step 5: Identify the search intent behind each keyword

Search intent is what the searcher wants to get done. Intent decides the page type, and a page that doesn't match it rarely ranks, however well it is written.

The most reliable way to read intent is the results page itself, because it shows what Google already serves for the query. If the first page is all how-to guides, write a guide. If it's mixed, the keyword has more than one intent; pick the one your page serves best. Our guide to search intent covers the types and edge cases in more depth.

"Keyword research" itself is a mixed-intent keyword, as our data further down shows: some searchers want a tool, others want to learn the method.

Step 6: Find long-tail keywords and questions

Long-tail keywords are specific, lower-volume searches such as "keyword research for a new website with no backlinks". They are easier to rank for, and their intent is usually clearer because the searcher has already said what they want.

Long-tail searches matter twice in 2026. They make up most of the distinct keywords people search (see the Ahrefs figure above), and they are the kind of sub-questions AI search runs when it fans out a broad question. A page that answers "how long does keyword research take" and "how often should you redo keyword research" in clearly headed sections can be found for both.

Where to find them: People also ask, autocomplete with question words, forum threads, your support inbox and Search Console queries of four words or more. Then decide for each one whether it deserves its own page or a section inside a bigger page. If the results for the long-tail term show the same pages as the head term, it's a section. Our long-tail keywords guide has more ways to find them.

Step 7: Find keyword gaps against your competitors

A keyword gap is a search your competitors rank for and you don't. Gap analysis is one of the quickest ways to find proven topics, because someone else has already shown there is demand.

Start by naming your real search competitors: the sites that keep appearing for your seed keywords. They aren't always your business rivals; a software company's search competitors are often publishers and review sites. Then:

1. List the topics your competitors' blogs, guides and category pages cover.
2. Search the topics that fit your business and note who ranks.
3. Mark every relevant topic where competitors appear and you don't.
4. Look for the reverse too: topics no competitor covers well, where a detailed, honest page can stand out.

Paid SEO suites can export a competitor's ranking keywords in bulk; the manual version works well for a small site. Our walkthrough on how to find competitor keywords covers both methods. If you track rankings in RankTera, it discovers competitors automatically from the Google results of your tracked keywords, which shows who you are actually up against.

Step 8: Group keywords into topic clusters

Keyword clustering means grouping the keywords one page can satisfy, so you build one strong page per topic instead of several thin ones. A topic cluster then connects a broad pillar page with the narrower pages around it.

The practical test is results overlap. Search two keywords. If several of the same URLs rank for both, Google treats them as one topic, and they belong on one page. If the results are different, they need separate pages. "Keyword research tool" and "keyword research tools" will usually share results; "keyword research" and "keyword difficulty" won't.

Clustering also prevents keyword cannibalization, where two of your pages target the same intent and compete for the same search. Assign every keyword to exactly one URL in your sheet and you will catch most cases before they happen. Our guide to topic clusters shows how to plan the pillar and supporting pages and link them together.

Step 9: Prioritize with a simple scoring model

Prioritization turns a long list into an order of work. Score each cluster from 1 to 3 on three questions, add the scores, and use search volume only to break ties.

A cluster that scores 8 or 9 goes first, even with modest volume. A 3 or 4 waits, however big its volume. This keeps a young site from spending months on terms it can't reach while easy, valuable topics sit unwritten.

Then map each prioritized cluster to a URL, new or existing, and track the ranking of each primary keyword so you can see which bets pay off and which pages need work.

Step 10: Reverse-engineer the terms the first results use

Reverse-engineering the first results means reading the pages that already rank for your keyword and listing the subtopics and phrases they share. Those shared terms show what a complete answer looks like for that search, and your page should cover them too, in your own words and with something the other pages lack.

To do it by hand, open the top-ranking pages for the keyword in your target country and note:

• the H2 and H3 headings each page uses, and which subtopics appear on most of them;
• recurring phrases and related terms (for "keyword research": search volume, keyword difficulty, search intent);
• the rough length and format of each page (guide, list, tool, video).

These related terms are what many people call LSI keywords; our guide to LSI keywords explains why the label is misleading and what actually helps. Cover the terms naturally and don't stuff them: Google's spam policies list "repeating the same words or phrases so often that it sounds unnatural" as keyword stuffing.

How to do this in RankTera

RankTera's KW Research automates this step. Type a keyword (or paste a URL), pick the country, and choose which of the top-ranking Google pages to learn from in the Select SERP Results picker; you can also add your own URL, file or text as a source. RankTera extracts the phrases those pages use under their H2–H6 headings and shows each one with its average count across the pages, plus the pages' average word count and heading structure. AI Clean-up removes junk keywords, and you can export each research as an XLSX file.

RankTera also reverse-engineers the keywords the first results use in AI search, in the pages ChatGPT and Gemini cite, so your article covers the same terms for Google and for AI answers. The research can then feed the AI Writer if you want a full draft built on those terms.

What we found: the terms ranking guides use for "keyword research"

We ran "keyword research" through RankTera's KW Research on October 15, 2026, using US Google results. The five ranking guides it analysed (semrush.com, backlinko.com, ahrefs.com, zapier.com and conductor.com) averaged 3,412 words, and the run found 83 keywords. These were the most-used terms, with their average count per page:

The guides Google ranks for this query treat volume, difficulty, intent and idea generation as the core of the topic, which is why each one gets its own step in this guide.

Our separate analysis of six of the top Google results for the same keyword, on the same day, showed how mixed the intent is. Four were keyword-tool landing pages, one was a glossary entry and one was a long guide of 10,214 words; the median page had 790 words. Only one of the six had an FAQ section, and only one used a table. Someone typing "keyword research" may want a tool or a method, and Google shows both.

Worked example: keyword research for an espresso equipment shop

This example runs the process end to end. The search volumes and scores are illustrative, not real data.

An online shop sells home espresso machines, grinders and accessories, and its blog is new.

Seeds (Step 1): espresso machine, espresso grinder, descaling, milk frothing.

Expansion (Steps 2 and 6): autocomplete and People also ask produce about 200 ideas, including "best espresso machine for beginners", "espresso machine vs coffee maker", "how to descale an espresso machine", "how to descale an espresso machine with vinegar", "how to froth milk without a steam wand" and "espresso grind size".

Clustering (Step 8): "how to descale an espresso machine" and "how to descale an espresso machine with vinegar" share most of their first-page results, so they become one page. "Espresso grind size" shows different results from "best espresso grinder", so those stay separate.

Scoring (Steps 3, 4, 5 and 9):

Decision: the shop writes "espresso machine vs coffee maker" first. It scores 9: the searchers are comparing before they buy, and the first page in this example holds forum threads and an outdated list. The descaling guide comes second (score 8), with a link to the descaling kit the shop sells. The beginner-machines list has the most volume, but large review publishers hold its first page, so it waits until the cluster has supporting pages. Before writing each page, the team lists the subtopics the top-ranking pages share (Step 10), such as water hardness and descaler versus vinegar for the descaling guide, and makes sure the page covers them with its own instructions.

Which keyword research tools do you need?

A keyword research tool speeds up one or two of the steps above; none does all ten. You can do solid keyword research with free tools and add paid ones when the amount of work justifies it.

Frequently asked questions

Can you do keyword research for free?

Yes. Google Keyword Planner gives search volumes and keyword ideas at no cost once you have completed your Google Ads account setup (billing details are required, but you don't have to run ads). Search Console shows the queries your site already appears for, and autocomplete, People also ask and Google Trends cover ideas and seasonality. Paid tools save time on bulk work and competitor exports, but the method in this guide works with free tools alone.

How many keywords should one page target?

One page should target one primary keyword plus the close variants and secondary keywords that share its intent. The results page is the test: if two keywords return mostly the same URLs, one page can rank for both. If the results differ, give each keyword its own page, or you risk two of your pages competing for the same search.

Where should you use your keyword on the page?

Use the primary keyword where it describes the page naturally: the title, the main heading, the first paragraph and the URL slug, plus a subheading if it fits. Cover the secondary keywords and related terms in the sections they belong to. Write for the reader, not a keyword count; unnatural repetition is keyword stuffing under Google's spam policies.

How often should you redo keyword research?

Revisit your keyword research every three to six months, and whenever you launch a product, enter a new country or see rankings drop. Search behavior shifts with new products, seasons and AI features. Between full reviews, check Search Console each month for new queries your pages already appear for; they are often the easiest new keywords you will find.

Is keyword research still relevant with ChatGPT and AI Overviews?

Yes. AI Overviews, AI Mode and Gemini build answers by running searches, often several per question, and they cite the pages they find; ChatGPT's search features also surface websites from the web. Keyword research tells you what those searches are: the head term plus the sub-questions around it. What changes is scope. Cover the long-tail questions in clearly headed sections, because AI answers draw on pages that rank for them.

How is keyword research for Amazon different?

Keyword research for Amazon targets shoppers who are already buying, so almost every search is transactional and product-specific: brand, size, material or use case. Google Keyword Planner reports Google searches, not Amazon ones, so start from Amazon's own search-bar suggestions and competitors' listings, then work the strongest terms into product titles, bullet points and descriptions.

Turn your keyword research into pages that rank

Keyword research pays off only when the pages get written and you can see how they perform. RankTera's KW Research reverse-engineers the terms the first results use on Google, ChatGPT and Gemini, the AI Writer turns them into full articles, and the Rank Tracker checks your Google positions and ChatGPT/Gemini citations for every tracked keyword. Start your 14-day free tier — no card needed; it includes 3 keyword researches, 3 AI articles and weekly tracking for 300 keywords.