Most keyword research ends as a spreadsheet nobody opens.
Two thousand rows. No decisions.
Good keyword research ends as a list of pages: which ones to fix, which ones to build, and the one keyword each page targets.
How do you do keyword research? Start from what the business sells and where. Give every existing page one target keyword, expand each service into the full set of searches around it, sort those searches by intent, check what's missing and who ranks today, then turn it all into a page-by-page plan. Treat every number from a keyword tool as directional, and verify the ones that matter.
Here's the process we use for client sites, step by step, including how the Website Quality Audit in Agency OS does the heavy lifting.
Here's the short version:
- Start with services and locations, not a seed keyword. The business decides the research, not the tool.
- Every page gets one target keyword. Most pages on a real site rank for nothing, so you have to research them, not read them off Search Console.
- Build hubs and spokes. Each service is a hub. Every search around it is a spoke with a job: money page, supporting page, cost page, near-me page or question.
- Intent decides the page type. Awareness, consideration or decision.
- Treat tool data as directional. Traffic estimates and search volumes are now wrong often enough to hurt campaigns.
- Prioritize by intent and customer value, not volume. Then verify the money keywords against real data.
What Is Keyword Research?
Keyword research is finding the searches a business's customers make, how often they make them, and which page on the site should answer each one. The last part is the one most agencies skip.
Keyword research that dies
- ✕2,000-row spreadsheet
- ✕Starts from a seed keyword
- ✕No page assigned to anything
- ✕Numbers taken on faith
Keyword research that ships
- ✓A decision for every page
- ✓Starts from services and markets
- ✓One target keyword per page
- ✓Numbers checked twice
For an agency, it does three jobs:
- Picks the target keyword for every existing page.
- Finds the pages that should exist and don't.
- Prices the opportunity, so you can forecast what the work is worth.
Why You Can't Trust Keyword Tool Data in 2026
Keyword tool numbers are estimates, and right now they're wrong often enough to hurt campaigns. Plan with them. Don't forecast with them.
Search Console and GA4 are the free checks on those numbers. See where each tool fits in SEO tools for agencies.
| What breaks | Why | What it does to you |
|---|---|---|
| Traffic estimates | Assume click rates from before AI Overviews | Overstate traffic, show growth that isn’t there |
| Search volume | Some keywords show spikes no person made | You rank and get nothing |
| Difficulty and cost per click | Built on the same inputs | Wrong priorities, wrong forecasts |
I'll be honest: this is the biggest change in how we do keyword research. We've watched tools show a client's traffic surging while their real traffic fell. We've seen keyword volumes nobody could have searched.
Three things are going wrong at once.
Traffic Estimates Assume Clicks That No Longer Happen
Ahrefs and Semrush estimate traffic by multiplying a keyword's search volume by the share of clicks each ranking position usually gets. Those click shares were built for a results page without AI answers.
| Google search results page | Visits with a click on a regular result |
|---|---|
| Without an AI summary | 15% |
| With an AI summary | 8% |
| Click on a link inside the AI summary | 1% |
According to Pew Research Center, Google users clicked a regular search result on 8% of visits when an AI summary appeared, versus 15% when it didn't, based on March 2025 browsing data from 900 US adults.
Read that again. Nearly half the clicks gone on those searches. A tool that assumes the old click rate overstates traffic for every keyword with an AI Overview.
Search Volume Is Inflated on Some Keywords
Some keywords show search volumes that can't be real. The trend line spikes and the other numbers stop making sense.

Here's "car accident lawyer" in Ahrefs on Nov 8, 2026: 382,000 searches a month, a difficulty of 4 ("Easy") and a cost per click of 45 cents.
Anyone who has bought that click in Google Ads knows it isn't 45 cents, and no one ranks a personal injury firm for it with five links. The trend line jumps sharply in the last year or two, and a spike like that is a reason to check before you trust it.
Rank for an inflated keyword and you get the position without the traffic.
What Bad Data Does to a Campaign
When the tool is wrong, you target the wrong keywords, report the wrong wins and forecast the wrong results.

This is a real client site. Ahrefs shows organic traffic more than doubling from May to October 2026. The site's actual organic sessions fell more than 25% over the same months.
Report the Ahrefs line and you'd be celebrating while the client loses leads. Here's why Ahrefs and Semrush traffic estimates are wrong, and how to check your own clients.
Step 1: Write Down What the Client Sells and Where
Before you open a keyword tool, list three things: the client's services, the markets they serve and the competitors they actually lose to.
| List | Example (a roofing company) | Why it matters |
|---|---|---|
| Services | Roof repair, roof replacement, gutters, storm damage | Each one becomes a hub |
| Markets | The three cities they serve | Each service × city is a page to check |
| Competitors | Two local roofers they lose jobs to | Directories and big-box stores aren’t a model |
This is the step that keeps the research about the business. A tool can only expand what you give it.
Two traps:
- Services the site doesn't mention. If you read services off the website, you'll never find the one they sell but never wrote a page for. Ask.
- Directories as competitors. For a law firm, the sites that rank for every term are Avvo, Justia and Best Lawyers. They're no model for a firm's content. Name real competitors.
In Agency OS, these three lists are saved against the client, so every later audit reads the same context. More on picking rivals in our guide on how to run an SEO competitive analysis.
Step 2: Give Every Existing Page One Target Keyword
Every page on the site needs one target keyword before you plan anything new. That's how you find pages to fix, merge or retarget.
| Page | Where the keyword comes from |
|---|---|
| Ranks already | Its best-ranking keyword, if it fits the page |
| Ranks for nothing | Its title or slug, run through a keyword tool: highest-volume match |
| Never | A Search Console query with six impressions |
Here's the catch. Tools only show keywords for pages that already rank. On one real audit, 36 of 156 pages ranked for anything at all.
The lazy fallback is Search Console. That's how a page ends up "targeting" a query with six impressions, because that's the only query it ever showed up for.
What to do instead:
- Pages that rank: use the keyword they rank best for, if it matches the page's purpose.
- Pages that don't: take the page's title or URL slug, run it through a keyword tool, and pick the highest-volume term that matches.
- Money pages first. They're the ones that produce customers.
The audit in Agency OS does step 2 automatically. It seeds Ahrefs Keywords Explorer from each page's title, money pages first, and writes the keyword onto the page with its volume, difficulty, cost per click and current position.
Step 3: Build the Keyword Universe With Hubs and Spokes
A keyword universe is every search around each service, grouped under that service. Each service is a hub. Each search around it is a spoke.
| Spoke | Job | US searches/mo |
|---|---|---|
| roof repair | Hub: the money page | 54K |
| roof leak repair | Supporting page | 24K |
| storm damage roof repair | Supporting page | 14K |
| roof repair near me | Near-me: money page + Business Profile | 76K |
| emergency roof repair near me | Near-me: money page + Business Profile | 16K |
| how much does it cost to repair a roof | Cost page | 700 |
| how to repair a damaged roof | Question: blog post | 400 |
Every spoke gets a job:
- Money page: the main commercial term. One page.
- Supporting page: a specific version of the service, like "roof leak repair."
- Cost page: "how much does it cost to..."
- Near-me page: usually answered by the money page and the Business Profile, not a separate page.
- Question: a blog post or an FAQ.
Filing spokes under their hub is the point. A flat list of keywords is how you end up with 20 pages competing for the same search.
For a local business, repeat it for each market: "roof repair" is one hub, and "roof repair in each city they serve" is a hub per city.
Pull the Spokes From a Keyword Tool
Run each hub through a keyword tool and keep the terms with real volume. We set the floor at 20 searches a month and keep up to 40 spokes per hub.

The matching terms list is where most spokes come from. The questions list fills the blog. Ignore "also talk about" for planning: it's related words, not related searches.
Already have a page for a spoke? Count it, don't propose it again. The audit checks the site first so the plan never asks for a page that exists.
Step 4: Sort Every Keyword by Intent
Search intent is what the person wants to do: learn, compare or hire. It decides the page type, so getting it wrong means building the wrong page.
| Intent | What they want | Page type |
|---|---|---|
| Awareness | To learn | Blog post, FAQ |
| Consideration | To compare and price | Cost page, comparison, “best” page |
| Decision | To hire | Service page, location page |
- Awareness: questions and explainers. Blog posts and FAQs.
- Consideration: comparing, pricing, shortlisting. Cost pages, comparisons, "best" pages.
- Decision: about to hire. Service pages and location pages.
You can guess intent from the shape of the phrase, and the audit does. But check the results page: if Google shows service pages for a term, write a service page, whatever the phrase looks like.
Step 5: Find What's Missing
The content gap is the pages the site should have and doesn't. There are two kinds.
| Gap | How you find it | Floor |
|---|---|---|
| Structural | Every service × every market the client serves | 30 searches/mo |
| Competitive | Terms the real competitors rank for, with no page on the site | 30 searches/mo |
| Never a gap | Branded terms |
- Structural gaps: every service crossed with every market. A business selling four services in three cities should be checked for all twelve pages.
- Competitive gaps: terms the real competitors rank for that the site has no page for.
Set a volume floor. We use 30 searches a month, so a services-by-cities grid doesn't turn into 48 thin doorway pages. Branded terms are never a gap.
Each gap becomes a row on the plan with no URL yet. That's the content you'll write. Our content brief template turns each one into an assignment.
Step 6: Check Who Ranks Today
Before you commit to a keyword, open the page ranking first for it. That's the page you have to beat, not a domain or a difficulty score.
- ?Page typeService page, blog post, directory?
- ?DepthHow much does it cover?
- ?Site strengthAuthority, reviews, brand
- ?Can you beat it?Realistically, and how soon
Ask four questions:
- What type of page is it? Service page, blog post, directory, video.
- How deep is it?
- How strong is the site? Authority, reviews, brand.
- Could your client's page realistically be better?
If the top results are all directories and national brands, the keyword may still be worth a page. Just don't forecast page one for it in month three.
Step 7: Verify the Keywords That Matter
Before a keyword goes in the plan or the forecast, check it against a second source. At least for the money keywords.
| Check | Cost | How reliable |
|---|---|---|
| Google Keyword Planner | Free with an Ads account | Second opinion on volume |
| Trend line | Free | Flags spikes |
| Read the results page | Free | Real difficulty, AI Overviews |
| Search Console impressions | Free, after access | Real for the client’s site |
| Small Google Ads test | A few days of spend | Real impressions for any term |
How to check, from cheapest to most reliable:
- Google Keyword Planner. A second opinion on volume, straight from Google's ad data.
- Look at the trend line. A sudden spike deserves suspicion.
- Read the results page. If page one is all national brands with huge link profiles, "Easy" is wrong. If there's an AI Overview, expect fewer clicks.
- Search Console impressions, once you have access. That's how often the client actually showed up.
- A small Google Ads test. A few days of ads on a term shows real impressions. It's the fastest way to settle a big decision.
For paid search, price every forecast off Keyword Planner's top-of-page bid range, not a tool's cost per click. Here's how we price a PPC proposal.
Step 8: Turn the Research Into a Page Plan
The output of keyword research is a decision for every page. Not a spreadsheet.
| Action | When |
|---|---|
| Leave | Ranks for its keyword and converts |
| Expand or refresh | Ranks 4 to 20, thin or dated |
| Retarget or rewrite the title | Aimed at the wrong keyword |
| Add links | Good page, no internal links |
| Merge or redirect | Two pages competing for one search |
| Build | A gap with real volume |
For each existing page, pick one or more actions. For each gap, add a page to build. Then order the work by how fast it produces leads.
Rank by intent and customer value, not volume. For a local business, a 50-search decision keyword beats a 5,000-search question every time.
That plan is also your forecast. Here's how to calculate the ROI of an SEO campaign from the same keyword data.
The full audit, from crawl to page plan, is in our guide on how to run a website quality audit.
How to Do Keyword Research With AI
AI is good at the grunt work of keyword research and bad at the judgment. Use it for the first, not the second.
| Task | AI | Strategist |
|---|---|---|
| Pull matching terms for every hub | ✓ | |
| Group spokes under hubs | ✓ | |
| Guess intent | ✓ | Confirm |
| Pick services and markets | ✓ | |
| Judge whether a number is believable | ✓ | |
| Decide if a page can win | ✓ |
Connected to Ahrefs or Semrush through MCP, Claude or ChatGPT can pull matching terms for every hub, group them, guess intent and draft the plan in minutes. That's the part that used to take a day.
What AI shouldn't decide on its own:
- Which services and markets matter. That comes from the client.
- Whether a number is believable. See Step 7.
- Whether you can actually beat the page ranking first.
That's how the audit in Agency OS works: it pulls and sorts automatically, then stops for a strategist to rule on page types, keywords and plans.
Keyword Research Tools
You need one keyword tool, Search Console and Google Keyword Planner. That's it.
| Tool | Use it for | Cost |
|---|---|---|
| Ahrefs or Semrush | Volumes, matching terms, who ranks | From $129/mo |
| Google Search Console | What pages already rank for | Free |
| Google Keyword Planner | A second opinion on volume; ad pricing | Free with an Ads account |
Not sure which paid tool to pick? Here's our Ahrefs vs Semrush vs Moz comparison.
Keyword Research Mistakes
- ✕Seed keyword first
- ✕Six-impression targets
- ✕No hubs
- ✕Unchecked intent
- ✕No volume floor
- ✕Untested numbers
- ✕Spreadsheet, not a plan
- ❌ Starting from a seed keyword instead of the client's services.
- ❌ Targeting Search Console queries with a handful of impressions.
- ❌ A flat keyword list with no hubs. You'll build pages that compete with each other.
- ❌ Guessing intent without checking the results page.
- ❌ A services-by-cities grid with no volume floor. Doorway pages.
- ❌ Trusting volume, difficulty or cost per click without a second source.
- ❌ Reporting a tool's traffic estimate as the client's traffic.
- ❌ Ending with a spreadsheet instead of a page plan.
Keyword Research FAQs
How do you do keyword research for SEO? List the business's services, markets and real competitors. Give every existing page one target keyword, expand each service into hubs and spokes, sort by intent, find the missing pages, check who ranks today, sanity-check the numbers and turn it into a page plan.
How many keywords should a page target? One primary keyword, plus the close variations and questions that belong to the same search. If two keywords need different pages to answer them, they're two pages.
How accurate is keyword tool data? Less accurate than it used to be. Traffic estimates assume click rates from before AI Overviews, and some search volumes look inflated. Use tool numbers to compare keywords with each other, and check the money keywords against Keyword Planner, Search Console or a small ads test.
What's a good search volume to target? It depends on the business. For a local service business, 20 to 30 searches a month for a commercial term can be worth a page. We set the floor at 20 for spokes and 30 for gap pages.
How do I find keywords a page doesn't rank for yet? Run the page's title or URL slug through a keyword tool and pick the highest-volume matching term. Search Console only shows queries a page already appeared for.
Can AI do keyword research? AI can pull, group and sort keywords quickly when it's connected to a keyword tool. A strategist still needs to decide which services matter, whether the numbers are believable and whether a page can win.
How often should you redo keyword research? Fully, once at the start of a campaign. Then review the plan every quarter, and research new keywords whenever the client adds a service or a market.
Turn Keyword Research Into a Page Plan in One Audit
Blueprint members get Agency OS, whose Website Quality Audit researches a keyword for every page, builds the keyword universe and turns it into a page plan, plus the training on how to deliver the work.
Join The Blueprint Training →