Almost every keyword research guide on the internet teaches the same process, and it’s the process from 2019: open a tool, sort by search volume, filter by difficulty, pick the winners, write the content.
That process still works. It’s just no longer sufficient — because a growing share of your buyers never type a keyword at all. They type a sentence into ChatGPT, or ask Perplexity a question in plain English, and get back three recommendations without ever seeing a search results page.
Those two behaviors need two different research processes. You need a keyword list for Google and a prompt bank for AI engines, and you need to understand where they overlap. Here’s how we run keyword research at OptiSEOn in 2026.
TL;DR — keyword research in 2026, in one paragraph
Modern keyword research has two halves. The first is traditional: identify search terms by volume, difficulty, and intent, then cluster them into topics. The second is new: build a prompt bank of conversational questions your buyers ask AI tools, which look nothing like keywords (“best CRM for a 12-person sales team that already uses HubSpot” vs. “best crm”). The two overlap more than they differ — both reward content that directly answers specific questions — but they’re discovered differently, measured differently, and optimized differently. Skip the second half and you’re optimizing for a shrinking share of discovery.
What is keyword research, actually?
Keyword research is the process of identifying the words, phrases, and questions your potential customers use when looking for what you offer — and assessing which of them you can realistically win. It has two outputs: a prioritized list of terms to target, and an understanding of what kind of content each term demands.
That second output matters more than most people realize. Ranking for “SEO audit” and ranking for “how much does an SEO audit cost” require completely different pages. The keyword tells you the topic; the intent tells you the format.
The four types of search intent
Before volume or difficulty, classify intent. Getting this wrong is the most common reason good content doesn’t rank.
- Informational — the searcher wants to learn. “What is schema markup.” Content type: guides, explainers, tutorials.
- Navigational — the searcher wants a specific site or brand. “OptiSEOn pricing.” Content type: your actual pages.
- Commercial investigation — the searcher is comparing before buying. “Best SEO agency Dallas,” “Ahrefs vs Semrush.” Content type: comparisons, roundups, alternatives pages.
- Transactional — the searcher is ready to act. “Hire SEO consultant,” “book SEO audit.” Content type: service pages, landing pages.
The fastest way to verify intent is to search the term yourself and look at what Google is already rewarding. If page one is all blog posts, a service page won’t rank there no matter how well optimized. Google has already told you what format wins — listen to it.
Step 1: Build your seed list
Start with 10–20 seed terms describing what you do, in the language your customers actually use. Not your internal jargon.
Sources for seeds:
- Your own service and product pages — what do you actually sell?
- Sales call notes and support tickets — the exact phrasing prospects use. This is the single most underused keyword source in most businesses.
- Google Search Console — Performance → Queries shows terms you already get impressions for. Frequently the highest-value starting point, because these are terms Google already associates with you.
- Competitor sites — their navigation, service names, and blog categories.
At OptiSEOn we start almost every client engagement by mining their existing Search Console data before touching a keyword tool. Businesses are routinely already ranking on page 2 for terms they never intentionally targeted — and moving a page-2 term to page 1 is dramatically cheaper than building a new ranking from zero.
Step 2: Expand with tools
Take your seeds and expand them into a full universe of terms.
Free tools:
- Google Search Console — your existing queries, impressions, positions
- Google Keyword Planner — volume ranges (requires an Ads account)
- Google autocomplete and “People Also Ask” — real question phrasing, free
- “Searches related to” at the bottom of results pages
- AnswerThePublic (limited free tier) — question-format expansion
- Reddit and industry forums — how real people phrase problems
Paid tools:
- Ahrefs or Semrush — the two standards; volume, difficulty, SERP analysis, competitor gaps
- Moz Keyword Explorer — good difficulty modeling
- Keywords Everywhere — inexpensive browser overlay
You don’t need a paid tool to do competent keyword research. You do need one to do it efficiently at scale.
Step 3: Assess difficulty honestly
Keyword difficulty scores are estimates, not measurements. Use them as a first filter, then verify manually.
The manual check that actually matters: search the term and look at who’s ranking. If page one is Wikipedia, HubSpot, Semrush, and three national publishers, a new site is not going to rank there this year regardless of what the difficulty score says. If page one includes a couple of small niche sites and some thin content, there’s an opening.
Practical guidance by site authority:
- New site (DR under 20): target long-tail, low-competition, question-format terms. Accept low volume. Volume compounds later.
- Established site (DR 20–50): mid-tail terms, comparison content, and local/vertical modifiers.
- Authoritative site (DR 50+): head terms become realistic, but competition is fierce and content quality has to be genuinely best-in-class.
This is where a lot of SEO advice fails people: telling a brand-new site to target “SEO services” is setting them up to produce content that will never rank. Start where you can win.
Step 4: Cluster into topics, not keywords
Modern Google doesn’t rank pages for single keywords — it ranks them for topics. A well-written page about schema markup will rank for hundreds of related variations without individually targeting each one.
So group your keyword universe into clusters where every term shares the same underlying intent and would be satisfied by the same page. “What is schema markup,” “schema markup explained,” “how does schema markup work,” and “structured data definition” are one cluster, one page — not four.
Practical approach: sort your keyword list by the top-ranking URLs for each term. If two keywords return substantially the same page-one results, they belong to the same cluster.
This clustering approach is what produced the internal structure of this blog — you’ll see it in our breakdown of AEO vs SEO vs GEO vs LLM Optimization, which serves as a hub that dozens of related terms funnel into.
Step 5: Build your prompt bank (the 2026 addition)
Here’s the half that traditional guides skip entirely.
When someone asks ChatGPT or Perplexity a question, they don’t type keywords. They type sentences. Compare:
| Google query | AI prompt |
|---|---|
| dallas seo agency | Who’s a good SEO agency in Dallas for a small B2B company? |
| schema markup types | What schema markup should I add to my site if I run a dental practice? |
| ecommerce seo | My Shopify store gets traffic but no sales — is it an SEO problem? |
| b2b saas seo | How should a Series A SaaS company prioritize SEO with a small team? |
The AI prompts are longer, more specific, more contextual, and often contain constraints (“small B2B,” “Shopify,” “Series A”). They also frequently ask for a recommendation rather than information — which is why brand visibility in AI answers matters so much.
How to build a prompt bank:
- Start from your buyer’s actual situation. For each customer segment, write 10–15 questions they’d ask an AI assistant, in full sentences, including their constraints.
- Mine your sales calls. The questions prospects ask on discovery calls are, almost verbatim, the questions they ask AI tools.
- Include the three prompt types: branded (“Is OptiSEOn any good?”), category (“Best SEO agency in Dallas”), and problem-first (“How do I show up in ChatGPT results?”).
- Run them. Actually type them into ChatGPT, Perplexity, Gemini, and Claude. Record whether you appear, whether competitors appear, and what sources get cited.
- Track weekly. Answers shift as models update.
That last step is measurement, not research — we covered the full methodology in how to measure AI search traffic from ChatGPT, Perplexity & Gemini. The research output is knowing which prompts matter; the measurement tells you whether you’re winning them.
How keyword lists and prompt banks interact
The good news: they’re not two separate content strategies. They’re two lenses on the same content.
A page that thoroughly answers “what schema markup should a dental practice add” will:
- Rank for the Google cluster around “schema markup for dentists”
- Get extracted as a featured snippet if structured correctly
- Get cited by Perplexity when someone asks the conversational version
- Feed ChatGPT’s answer when it browses the web
One page, four surfaces. That’s the entire thesis behind how we structure content — and it’s why the tactics in how to get cited by ChatGPT, Perplexity & Gemini overlap so heavily with classic on-page SEO.
The practical implication: use your keyword research to pick topics, and your prompt bank to shape the structure within each topic. Keywords tell you what to write about. Prompts tell you which specific questions to answer with H2s and FAQ entries.
Step 6: Prioritize (the part most people rush)
You’ll end up with more opportunities than capacity. Prioritize on four factors:
- Business value — does ranking for this actually produce revenue? A high-volume informational term that never converts is worth less than a low-volume term with buying intent.
- Achievability — can you realistically rank given your current authority?
- Existing position — terms where you’re already on page 2 are the cheapest wins available.
- Cluster leverage — does this page unlock a cluster, or is it a one-off?
A simple prioritization that works: start with terms where you rank positions 5–20 and have commercial intent. Those are existing assets that need refinement, not new assets that need building. We’ve seen clients get more organic revenue lift from three weeks of optimizing existing page-2 content than from six months of new content production.
Common keyword research mistakes
- Chasing volume over intent. 10,000 monthly searches with zero buying intent is worth less than 100 searches from ready buyers.
- Ignoring zero-volume keywords. Tools report “0” for many long-tail and emerging terms that get real searches. Especially true for AI-era conversational queries. If your customers ask it, it matters, regardless of what the tool says.
- One keyword, one page. Leads to thin, cannibalized content. Cluster instead.
- Skipping SERP analysis. The single fastest way to waste a quarter is writing a blog post for a query where Google only ranks product pages.
- Never revisiting. Keyword landscapes shift. Re-audit quarterly.
- Building a prompt bank and never running it. Research without measurement is a document nobody reads.
Where this fits in OptiSEOn’s process
Keyword and prompt research is the first deliverable in every OptiSEOn engagement, because everything downstream depends on it — content planning, technical SEO priorities, and LLM optimization targets all follow from knowing what your buyers actually search and ask.
For vertical-specific applications, see our B2B SaaS SEO playbook and eCommerce SEO playbook — the research process is the same, but the highest-value keyword types differ significantly by business model.
Frequently Asked Questions
How do I do keyword research for free? Google Search Console (your existing queries), Google autocomplete, People Also Ask boxes, “searches related to” at the bottom of results, and Google Keyword Planner cover most of what a small business needs. Paid tools like Ahrefs and Semrush make the process faster and more thorough, but they aren’t strictly required.
What is a good search volume to target? There’s no universal number — it depends on your site’s authority and the term’s commercial value. A new site is usually better served targeting terms with 50–500 monthly searches and low competition than fighting for 10,000-volume head terms it can’t win. Low-volume, high-intent terms often produce more revenue than high-volume informational ones.
Does keyword research still matter with AI search? Yes, and it now has a second half. Traditional keyword research still governs Google rankings, which remain the largest single traffic source for most businesses. But you also need a prompt bank — the conversational, full-sentence questions people ask ChatGPT and Perplexity — because those queries look and behave differently from Google keywords.
What’s the difference between a keyword and a prompt? A keyword is the compressed phrase people type into a search engine (“dallas seo agency”). A prompt is the full-sentence question people ask an AI tool (“who’s a good SEO agency in Dallas for a small B2B company?”). Prompts are longer, contain more context and constraints, and often ask for a recommendation rather than information.
How often should I redo keyword research? A full refresh quarterly, with lighter monthly reviews of Search Console data for emerging queries. AI prompt banks should be run weekly (to track visibility) even though the bank itself only needs quarterly expansion.
Should I target keywords I have zero chance of ranking for? Not as primary targets. But high-difficulty head terms can be worth including as secondary keywords within a page targeting achievable long-tail terms — you’ll pick up incidental impressions and occasionally surprise yourself as authority grows.
Want a keyword and prompt research audit for your business — the Google half and the AI half? OptiSEOn is a Dallas-based SEO, AEO, GEO, and LLM optimization agency. Book a free audit and we’ll show you the terms you’re already close to winning, the prompts your buyers are asking AI tools, and where the gap is.
Leave a Reply