How to Use Claude for Keyword Research? [MCP Included]

Using Claude for keyword research is faster and more capable than most SEO practitioners realise – not because Claude is a keyword tool, but because Claude connected to Ahrefs and SEMrush via MCP becomes one.

The Model Context Protocol (MCP) lets Claude call your keyword tools directly, pull live volume and difficulty data, and run the full research workflow inside a single conversation. No tab-switching. No copy-pasting lists between tools.

The failure mode most people hit is treating Claude as a brainstorming layer and nothing more. They prompt it for keyword ideas, get a generic list with no data attached, and conclude Claude is not useful for research.

That conclusion is wrong – the setup is wrong.

Once the MCP connections are in place, Claude handles seed expansion, intent classification, cluster mapping, and brief generation with live data at every step.

This guide covers the practical workflow: how to connect the tools, what to prompt at each stage, and how to close the session with a brief the model can actually write against.

No conceptual overview of what keywords are – this assumes you already do keyword research and want to do it faster.

How Do You Connect Claude to Ahrefs and SEMrush via MCP?

MCP – Model Context Protocol – is the layer that lets Claude call external tools directly rather than waiting for you to copy data between tabs. For keyword research, it means Claude can query Ahrefs or SEMrush mid-conversation, retrieve live metrics, and factor them into the next step without you touching either platform.

The setup takes under 15 minutes per tool and the research workflow is fundamentally different once it is in place.

Ahrefs MCP setup: Ahrefs exposes an MCP server that Claude can connect to via the Claude settings panel under Integrations. You need an active Ahrefs subscription with API access – available on Standard plans and above. Once connected, Claude can pull keyword volume, keyword difficulty, SERP overview, and top-ranking pages for any query directly inside the conversation.

SEMrush MCP setup: SEMrush runs a separate MCP server, also connectable via Claude’s Integrations panel. The SEMrush connection gives Claude access to keyword volume by geography (critical for Australian market data), intent classification signals, CPC data, and keyword variations. For AU-specific research, SEMrush tends to return more granular local volume data than Ahrefs.

Run this prompt first to verify both connections are active before starting any research session:

// Test both MCP connections
Using Ahrefs, pull keyword data for "SEO agency Australia".
Using SEMrush, pull AU volume and intent data for the same keyword.
Return both results side by side.

If both tools return data, the connections are working. If one returns an error, check the API key in your tool’s account settings and re-authenticate in Claude’s Integrations panel.

How Do You Build a Seed Keyword List Using Claude and Live Data?

The standard approach – prompt Claude for keyword ideas, then open Ahrefs in a separate tab to check volume – works, but it misses the point of having MCP connected. The better workflow runs seed generation and data validation inside the same prompt chain. Claude generates the candidates, immediately pulls live metrics, and filters the list based on the data it just retrieved.

A practical seed expansion prompt looks like this:

// Seed expansion with live validation
Industry: [e.g. accounting firms]
Market: Australia
Audience: [e.g. small business owners looking for an accountant]
Intent stage: [e.g. awareness / consideration / decision]
Generate 20 seed keyword topics for this audience.
Then use SEMrush to pull AU search volume and keyword difficulty for each.
Return as a table: keyword | monthly volume (AU) | KD | intent type.
Flag any keywords where AU volume differs significantly from global volume.

The AU volume flag is important. Many professional services keywords in Australia carry 80-95% lower search volume than the same terms in the US. A keyword that looks viable at global volume may be near-zero locally. Claude will surface this discrepancy automatically if you ask for it – but it will not flag it unprompted.

How Do You Classify Search Intent Across a Large Keyword List?

Manually classifying intent across 100 keywords takes 30-60 minutes of careful SERP review. Claude with MCP connected does it in one prompt, with live SERP data attached to each classification rather than just text inference. The output is a structured table that feeds directly into content planning without a secondary review step.

The batch intent classification prompt:

// Batch intent classification with SERP context
Here is a list of [X] keywords: [paste list]
For each keyword:
1. Use Ahrefs to check what page types dominate the top 5 results
 (agency pages / how-to guides / directories / forums)
2. Classify intent as one of:
 - Informational - DIY
 - Informational - research
 - Commercial - comparison
 - Transactional - buy/hire
3. Recommend article type: pillar guide / how-to / comparison / local page
4. Assign priority: high / medium / low based on intent-to-conversion distance
Return as a structured table. Flag any keywords where SERP results
conflict with the keyword phrase (mixed intent signals).

The SERP conflict flag is the most valuable part of this output. Keywords where the phrase implies one intent but the ranking pages serve another are high-risk targets – write the wrong article type and it will not rank regardless of quality.

How Do You Map a Topic Cluster with Claude?

Topic cluster mapping is where Claude with MCP access produces output that would take a senior SEO strategist several hours to replicate manually. The workflow runs in three stages inside one session: cluster generation, live difficulty validation for every cluster URL candidate, and competitor gap analysis against a real competitor’s ranking profile.

Start with cluster generation:

// Topic cluster mapping
Pillar keyword: [e.g. "SEO for accountants"]
Market: Australia
Goal: Build topical authority - mix of traffic-volume targets and
 zero-volume supporting articles
Map a full topic cluster:
- 1 pillar article
- 8-12 supporting articles grouped by subtopic category
- For each article: suggested keyword, article type, intent type
Then use SEMrush to pull AU volume and KD for each supporting keyword.
Flag which articles are zero-volume topical authority plays vs
genuine traffic targets.
Return as a structured table.

Once the cluster map is back, run the competitor gap prompt in the same session:

// Competitor gap analysis
Using Ahrefs, pull the top 20 ranking pages for [competitor domain]
in the [industry] category.
Compare against the cluster map above.
Identify topic areas the competitor is not covering.
Prioritise gaps by: search volume, intent-to-conversion distance,
and topical relevance to the pillar keyword.

The gap output tells you not just what to write, but what to write first. Topics the competitor has ignored in a cluster they otherwise dominate are the fastest path to ranking – less established competition, clear intent, direct topical relevance.

How Do You Find Long-Tail and Question Keywords With Real Volume Data?

Long-tail and question-based keywords are where most AI Overviews appear and where most competitor content is thinnest. Claude is particularly effective at generating question keywords for specific professional audiences because it can simulate the exact language a practitioner in that industry would use – which is often different from the generic phrasing a keyword tool surfaces.

The question keyword expansion prompt:

// Question keyword expansion for practitioner audiences
Industry: [e.g. physiotherapy]
Audience: Practice owners managing their own SEO
Market: Australia
Generate 25 questions this audience would actually type into Google -
not generic SEO questions, but questions specific to how physiotherapy
practices think about their online presence.
Then use SEMrush to pull AU volume for each question.
Separate into three groups:
- Has search volume (traffic target)
- Zero volume but high AI Overview potential (topical authority)
- Zero volume, low AI Overview potential (deprioritise)

Zero search volume does not mean zero search presence. Question-format keywords with no measurable volume consistently appear in AI Overviews and People Also Ask boxes – which means they generate impressions and clicks that GSC records but no keyword tool predicted.

The third group – zero volume, low AI Overview potential – is genuinely worth deprioritising. Not every question a practitioner might ask deserves a published article. Claude will generate more candidates than you need; the SEMrush data and the AI Overview filter help you cut the list to what is worth building.

For additional signal, paste a relevant Reddit thread or forum discussion into the same session and add this prompt:

// Extract keyword-worthy questions from forum content
Here is a forum thread from [source]: [paste content]
Extract every question being asked that could become a standalone
search query. Rewrite each as a natural Google search phrase.
Then check SEMrush for AU volume on each.

How Do You Build a Keyword Brief Claude Can Actually Execute?

The research session only produces value if it closes with a structured brief. A keyword list is not a brief. A brief specifies confirmed intent, the exact keyword target, supporting keywords with volume attached, the entity map for the vertical, sourced statistics, FAQ questions from real search data, and the structural outline. Claude can generate all of this from the session it just completed – but it needs to be prompted to do so before the conversation closes.

The brief generation prompt:

// Auto-generate article brief from research session
Based on the keyword research we have completed in this session,
generate a structured article brief for: [primary keyword]
Brief must include:
- Confirmed intent type and audience
- Primary keyword + 5 supporting keywords with AU volume
- Entity map: regulatory bodies, associations, directories, tools
- 3 verified statistics with source and year
 (use only data retrieved from Ahrefs/SEMrush in this session)
- 4 FAQ questions sourced from the question keyword list above
- Recommended H2 structure (8-10 questions)
- Word count target
- Article type
Output as JSON.

The JSON output format matters. A brief returned as plain text gets manually reformatted before it reaches the writer or the next model in the chain. JSON can be passed directly into a drafting workflow – or saved and reopened in a future session as the starting point for the article.

One step stays human regardless of how well the session ran: intent sign-off. Before any article moves from brief to draft, a human editor confirms the intent classification is accurate against the live SERP. Claude can misread intent on ambiguous or low-volume queries where ranking precedent is thin. That check takes two minutes and prevents writing a 3,000-word article at the wrong angle.

Author Profile

Pham Van Hien (pvhien)

I’m an SEO Manager with 7+ years of experience helping brands grow through data-driven strategies. Passionate about the intersection of search, content, and technology, I blend technical SEO, analytics, and creativity to drive performance and build meaningful digital experiences.

Latest entries

• AI SEOMarch 30, 2026How to Use Claude for Keyword Research? [MCP Included]
• AI SEOMarch 18, 2026AI-Executed vs AI-Generated Content: 1 Builds, 1 Kills?
• AI SEOMarch 18, 2026ChatGPT Is Crawling Your Site. Your GSC Data Shows It.
• SEOMarch 9, 2026SEO for Construction Companies: Best Practices to Win More Leads in Australia