Ahrefs Keyword Research, Augmented with KeywordGraph: From Matching Terms to Topic Clusters

Export Matching Terms from Ahrefs

Step 1 · 5 minutes inside Ahrefs

In Ahrefs, open Keywords Explorer and type the seed phrase. In the support documentation walkthrough, the worked example uses ai tools as the seed.

From the left menu, navigate to Keyword Ideas > Matching Terms. This report returns the queries that contain the seed phrase as a substring, with volume, difficulty, and SERP metadata. For a commercial seed, the report typically returns one to several thousand rows.

Apply a minimum-volume filter to drop the long-tail noise. The reference walkthrough uses 50 monthly searches as the cutoff, which leaves around 2,000 keyword combinations for the ai tools seed.

Export the filtered table as a CSV file with UTF-8 encoding. The columns that matter downstream are keyword, difficulty, and volume. Sample datasets are available in the InfraNodus public datasets repository at github.com/infranodus/datasets.

Import the CSV into KeywordGraph

Step 2 · 3 minutes inside KeywordGraph

In KeywordGraph, start a CSV import. On the column-mapping step, choose the keyword column as the column to analyze. This is the column the co-occurrence graph will be built from.

On the next step, mark difficulty and volume as filter columns. These do not enter the graph as nodes; they become sliders applied to the resulting graph later, used to surface the high-volume / low-competition segments in step 4.

The import takes a minute or two depending on row count. The output is a knowledge graph where each unique entity in the keyword strings is a node and each co-occurrence of two entities inside the same keyword string is an edge.

Read the topical clusters

Step 3 · 15–30 minutes of interpretation

The graph surfaces the topical clusters automatically. For the ai tools export, five clusters emerged in the walkthrough:

• AI tools (especially for marketing)
• Image generation (especially the “ghibli” style)
• Blog insights
• Video creation (especially YouTube shorts)
• Ad research (audience study)

Each cluster is a candidate page or page group. Click any node to see the specific keyword combinations inside — useful for confirming the cluster name and spotting outliers.

A useful refinement: remove the dominant seed nodes (ai, tool) from the graph and re-run the cluster detection. The clusters recompute against the surrounding context and reveal tighter topics. In the walkthrough, the refined clusters were:

• Video creation
• Content detection
• Data optimization
• Teaching excellence
• Marketing automation

Drilling into the content detection cluster, the walkthrough notes the cluster covers both AI content generation and AI content detection — the poison and the remedy in one cluster. Content planning here would split into two pages: one for the writing/generation intent, one for the detection intent, with a bridge between them.

Filter by difficulty and volume to find the overlap

Step 4 · 10 minutes

With the filter columns set up in step 2, apply both sliders to surface the high-volume / low-competition segments. The walkthrough uses these ranges:

• Difficulty: 30 to 59 — low enough to avoid the highest-competition head terms; high enough to skip uncontested noise.
• Volume: 100 to 1,200 — meaningful demand without jumping into the saturated head-term band.

Selecting the overlap of both filters reveals the keyword combinations that deserve the first wave of publishing. In the worked example, the overlap surfaces ai tools for data analysis as a target combination — meaningful volume, not the highest competition, adjacent to the team’s text-analysis positioning.

Drop the difficulty band further (to a lower-competition tier) and the surface shifts toward data analysis + market research. Raise it and content detection emerges — useful to know but too competitive for the first wave.