Skip to content
NakodaAI

AI FIELD MANUALMARKETERS

Marketing lead needs to find what competitors rank for that this site doesn't - without a week of manual spreadsheet work

A real content-gap workflow: export competitor keyword data from an SEO tool, then use an LLM only to cluster and prioritize it - never to invent the ranking data itself.

Last reviewed 1 September 2026

THE PROBLEM

A marketing lead at a mid-sized SaaS company needs a prioritized content plan: what topics do three named competitors rank for that this site has no page on at all. The manual version of this task - exporting keyword lists, cross-referencing three competitor domains against your own, clustering by topic - is real spreadsheet work that eats a full day.

The shortcut some teams reach for is asking an LLM directly: "what keywords does [competitor] rank for." A general-purpose chat model has no live index of search rankings - it will produce a plausible-sounding list of keywords a company in that space probably targets, not the keywords they actually rank for today. Building a content plan on invented ranking data wastes the quarter it takes to find out the plan was based on nothing.

The actual gap isn't in the clustering and prioritization - that part is genuinely faster with an LLM. It's in where the raw ranking data comes from.

THE APPROACH

Pull real ranking data from an SEO tool that actually crawls and indexes search results (Ahrefs or Semrush), export the competitor keyword gap report those tools already build for exactly this purpose, and only then use an LLM to cluster the resulting list into content themes and help prioritize by estimated effort versus opportunity.

The LLM never sees or guesses at ranking data - it receives a spreadsheet of real keywords, real search volumes, and real competitor rank positions, and its job is organizing and prioritizing that real data, which is a task it's genuinely good at.

WHY IT WORKS

SEO tools maintain their own crawled index of search results and rank tracking - this is fundamentally different from an LLM's training data, which has no live connection to current search rankings at all. Real data in, real prioritization out; the moment invented data enters the pipeline, everything downstream is unreliable regardless of how good the clustering looks.

Clustering hundreds of exported keywords into coherent content themes by hand is genuinely tedious and exactly the kind of pattern-finding-in-structured-data task an LLM handles well and fast, once it's working from real numbers rather than being asked to conjure them.

STEP BY STEP

  1. Run the competitor gap report in a real SEO tool

    In Ahrefs (Content Gap) or Semrush (Keyword Gap), enter your domain and up to a few competitor domains. Export the list of keywords the competitors rank for (top 10-20 positions) where your site has no ranking at all.

  2. Clean the export

    Filter out branded terms (competitor's own name), clearly irrelevant terms the tool's automated matching sometimes includes, and anything with negligible search volume for your market.

  3. Hand the cleaned list to an LLM for clustering

    Paste the keyword list (with volume and competitor rank columns) into Claude or ChatGPT and ask it to group the keywords into 8-12 coherent content themes, each with a suggested page or article angle.

  4. Prioritize by a defined formula, not gut feel

    Ask the model to rank the clusters by a stated formula you provide - e.g., total search volume in the cluster, divided by how competitively the top-ranking competitor pages seem to be covering it (rough proxy: how many competitors rank for it, and at what position). This keeps prioritization traceable rather than a vibe.

  5. Sanity-check the top five clusters manually

    For the top five prioritized clusters, actually look at what's currently ranking for the anchor keyword in each. If it's dominated by pages you have no realistic way to outrank soon (major publishers, massive domain authority), deprioritize regardless of what the volume math said.

  6. Brief writers from the cluster, not the keyword list

    The content brief should be built from the theme and angle the LLM proposed for the cluster, refined by a human editor - not a keyword-stuffed brief built directly from the raw export.

TOOLS

LIMITATIONS

  • SEO tools' keyword-to-rank mapping is itself an estimate built from their own crawling and panel data, not Google's actual index - treat exported search volumes as directionally useful, not exact.

  • Content-gap analysis finds what competitors rank for; it does not tell you whether ranking for it would actually move revenue. A high-volume keyword with no buying intent behind it can win the prioritization math and still be the wrong page to build.

  • This workflow produces a prioritized topic list, not finished content. Writing AI-generated articles directly from these briefs without real editorial judgment and subject-matter accuracy checking is a separate, much riskier workflow this entry does not cover or recommend.

EXAMPLE

A B2B SaaS company needs a Q3 content plan built from a gap analysis against three named competitors.

  1. Ahrefs Content Gap run against the three competitor domains: 340 keywords exported where the company's own site ranks nowhere in the top 100.

  2. Cleaned to 210 after removing branded terms and near-zero-volume keywords.

  3. Claude clustered the 210 into 11 themes, each with a proposed angle (e.g. "integration setup guides," "comparison pages," "industry-specific use cases").

  4. Prioritization formula applied: total cluster volume ÷ average competitor rank position. Top cluster: integration setup guides, high volume, competitors ranking 4-8 (beatable, not dominated by a major publisher).

  5. Manual check on the top cluster confirmed the ranking pages were competitor help-center articles, not deeply authoritative sources - a realistic target.

An 11-theme, prioritized Q3 content plan built from real competitor ranking data in an afternoon, with the top-priority cluster manually verified as actually winnable before a single brief was written.

RELATED

Nakoda editorial · last reviewed

This entry describes a workflow Nakoda recommends - it is not a claim about how any named tool behaves in every case, and it is not paid placement. Spotted something out of date? Tell us.

Back to the Field ManualTalk to Nakoda