How to Find Content Gaps Using Clarity and Google Search Console

Most content decisions are guesses. You write about topics you think matter, but without data, you don’t know if visitors actually engage with your posts, or what they’re searching for but not finding. This guide teaches you to combine two powerful data sources — Microsoft Clarity (real user behaviour) and Google Search Console (search intent) — to build a data-backed content strategy. Claude automates the analysis, surfacing gaps, quick wins, and problems worth fixing.

What you need before starting

• A website with Clarity installed (Microsoft Clarity MCP connected to Claude)
• Google Search Console access to your site (optional but recommended)
• Access to Claude (via claude.ai with Clarity MCP enabled, or Claude Desktop if using alternative data sources)
• 30 minutes to set up; 15 minutes per analysis run
• A Google Sheet or spreadsheet to track content recommendations over time

Platform note: Clarity MCP is cloud-only and works on claude.ai. If you’re using Claude Desktop, you can still run this workflow using exported data from Clarity + GSC CSVs, though the analysis will be manual rather than automated.

Why these two data sources matter

Clarity shows user behaviour: Which pages people visit, how far they scroll, where they click, what frustrates them (rage clicks, dead clicks). It answers: “Are people actually engaging?”

Google Search Console shows search intent: What people search for to find you, which searches bring impressions but no clicks (title/meta problems), and which queries you’re missing entirely. It answers: “What are people looking for?”

Cross-reference these and you find:

• Posts with high traffic but low engagement (fix first)
• Search queries that get impressions but no clicks (meta description or title issue)
• Search intent you’re capturing traffic for but failing to serve well
• Topics people search for that you don’t have content for yet (write next)

The clever part: instead of reading raw data for hours, Claude synthesises both sources and surfaces actionable recommendations ranked by impact.

The workflow: three data streams

Step 1: Connect Clarity and Google Search Console

Clarity setup:

1. Install Clarity on your website (copy the tracking code into your header)
2. Give Claude access to Clarity MCP in your Claude.ai integrations
3. Verify it’s working: you should see session data within 1–2 hours of traffic

Clarity MCP pulls:

• Top pages by session count
• Scroll depth (how far visitors scroll on each page)
• Rage clicks and dead clicks (frustration signals)
• Traffic sources (direct, organic, referral, etc.)

Google Search Console setup:

1. Connect your site (verify ownership via DNS, HTML tag, or Google Analytics)
2. Wait 24–48 hours for search data to appear
3. Give Claude access via the Google Search Console tool

GSC pulls:

• Top 20 queries by impressions and clicks
• Click-through rate (CTR) by query
• Pages with high impressions but low CTR (<3%) — quick wins for title/meta fixes
• Search position (average ranking for each query)

Data freshness: Clarity updates every few hours. GSC data lags by 1–3 days. Run this analysis weekly or fortnightly for best results.

Step 2: Filter for signal, not noise

Both tools can be overwhelming. Claude automatically filters to signal:

From Clarity:

• Sessions ≥20 seconds (longer = more engaged; shorter = accidental clicks)
• Pages with >100 sessions (enough data for patterns, not noise)
• Scroll depth <40% on high-traffic pages (engagement problem, worth investigating)
• Rage click rate >5% or dead click rate >10% (UX problem signal)

From GSC:

• Top 20 queries by impressions (what people search for most)
• Top 20 queries by clicks (what actually drives traffic)
• Impressions >10 but CTR <3% (obvious title/meta improvement opportunity)
• Queries with no matching page slug (content gap signal)

This filtering cuts the noise by ~80%, leaving only insights worth acting on.

Step 3: Cross-reference against your content library

This is where it gets clever. Claude doesn’t just surface data — it cross-references the signals against your existing content to avoid duplicate recommendations.

Claude maintains a content library of:

• All published posts (title, slug, keywords, publication date, category)
• All pages, guides, case studies, portfolios (everything on your site)
• Archive of what’s been recommended before (avoid repeating suggestions)

When Claude finds a GSC query with no matching post, it checks: “Do we already have a post on this topic under a different slug? Is it too similar to something published last month?” This prevents wasted recommendations and duplicate work.

Three types of recommendations

Fix First: Posts with traffic but low engagement

These are quick wins. You have visitor attention — your post just needs restructuring or better framing.

Example from placeholder data:

Post	Issue	Signal	Suggested fix
“ERC Starting Grants 2026 Quick Reference”	Low scroll depth	847 sessions, 28% avg scroll	Rewrite intro as TL;DR bullets. Visitors land on the page but don’t read past the first section.
“Horizon Europe for Early-Career Researchers”	Rage clicks on CTA	623 sessions, 12 rage clicks	CTA button likely not working or unclear. Test button text + placement.
“Materials Science Funding Landscape”	High bounce on section 2	521 sessions, 34% scroll to section 2, then drop	Restructure: move key findings earlier. Readers are leaving at the same point.

Why fix first: You already have the audience. A restructure or UX fix takes 30 minutes and could improve engagement by 40–50%.

Write Next: Content gaps with data backing

These are topics people search for but you don’t rank for yet, or adjacent topics to your high-engagement posts.

Example:

1. “Comparing ERC and Horizon Europe grants for materials scientists”

   • Why: GSC shows “ERC vs Horizon Europe” gets 28 impressions/month, 0 clicks. You rank but visitors don’t click your link. Likely title/meta issue + topic confusion.
   • Angle: Side-by-side comparison, decision tree (which programme for your goals?)
   • Data-backed alternative: “Your top traffic page is the ERC guide. A comparison post clusters related search intent.”

2. “How to write a compelling research impact statement”

   • Why: GSC shows “research impact statement examples” gets 34 impressions, 0 clicks. You don’t rank. Search intent is clear (people want examples). Your grant-writing posts mention impact but don’t deep-dive into this.
   • Angle: Template + examples + common mistakes
   • Avoids duplicate: Nearest content is “Grant Writing Fundamentals” (published 60 days ago). This is distinct (focus on impact specifically).

3. “Funding for interdisciplinary research in Europe”

   • Why: Clarity shows your “Materials Science Funding” post is top 5 by sessions (847) but only 28% scroll depth. Visitors interested in cross-disciplinary funding but your post is materials-specific. New post serves adjacent audience.
   • Angle: European funders accepting interdisciplinary teams (EIC, SNF, NWO, Villum)
   • Avoids duplicate: Distinct from materials-only scope.

The pattern: these recommendations come from real user behaviour + real search data, not guesswork.

Remove or Archive: Pages draining resources with no traffic

Less common but important: if a page gets <10 sessions/month, ranks nowhere in GSC, and hasn’t been updated in 12+ months, it might be worth archiving (keeping it live but de-prioritising) or redirecting to a more relevant post.

The clever architecture: approval layers

Claude doesn’t just dump recommendations. It synthesises across three layers:

1. Behaviour layer (Clarity): What’s working, what’s frustrating users, where engagement drops
2. Intent layer (GSC): What people search for, how well your titles match their queries, where you’re missing opportunities
3. Content layer: What you’ve already published, when, how similar new topics might be, what’s overdue for refresh

A recommendation is only surfaced when all three layers align. This filters out false signals and noise.

Example: “Write a post about X” only appears if:

• Clarity shows adjacent content is popular (behaviour signal)
• GSC shows people search for X (intent signal)
• Your content library shows no existing post covers X (duplication check)

Making this actionable

Once Claude generates recommendations, you triage them by effort and impact:

High impact, low effort (do first):

• Fix title/meta on high-impression, low-CTR posts (30 min per post)
• Restructure low-scroll-depth posts (1–2 hours per post)
• Create a simple comparison post when two related topics get search traffic (2–3 hours)

High impact, high effort (do next):

• Write a new guide covering a clear content gap (4–6 hours)
• Cluster and expand a category with multiple quick-reference posts (5–8 hours per cluster)

Low impact or low urgency (do later or skip):

• Archive pages with <10 sessions/month
• Update dated posts that still perform (maintenance vs. new creation)

Troubleshooting

Clarity shows sessions but GSC shows no impressions for those pages
Clarity tracks all traffic (direct, apps, internal links). GSC tracks only organic search traffic. This gap is normal. Focus GSC recommendations on search-driven pages, and Clarity recommendations on behavioural engagement regardless of traffic source.

Claude’s recommendations feel too obvious or unhelpful
Your content library might be outdated. Ask Claude to rebuild it: “Refresh the content library, then re-run the analysis.” This re-scans all published posts and catches recent work the library missed.

You disagree with a recommendation or think it duplicates something
Flag it. Ask Claude to re-check: “Does post X already cover this topic? Should we expand it instead of writing a new one?” The approval layer works both ways — you challenge recommendations, Claude reconsiders.

GSC shows high impressions but low clicks, but the title/meta looks fine
Could be positioning: maybe your snippet doesn’t answer the searcher’s intent. Try a different angle in your title or rewrite the description to match search intent more closely. Clarity + GSC alone don’t tell you intent — they tell you there’s a gap. You fill it strategically.

What you can do now

Once you’ve run your first analysis:

• You have a ranked list of “fix first” posts — pick one and restructure it
• You have 3–5 “write next” recommendations with data backing — choose one and draft it
• You know which topics are underperforming despite high traffic — you can prioritise those for refresh
• You have baseline metrics to compare against — run this again in 4 weeks to measure impact of changes

Example: Fix the top “Fix First” post (30 min). Re-run analysis in 2 weeks. If scroll depth improved and bounce rate dropped, you’ve validated the approach. This becomes your content maintenance rhythm.

Key principles for best results

1. Data beats opinion — If something contradicts your assumptions, trust the data first. Your intuition about what people want is often wrong.

2. Approval layer prevents hallucination — Claude only recommends content that aligns across behaviour, intent, and deduplication checks. If it doesn’t check all boxes, it’s not recommended.

3. Run regularly — Weekly or fortnightly analysis catches trends early. Monthly is too infrequent; daily is overkill.

4. Fix before writing — Restructuring an underperforming post takes 30 minutes and can be more impactful than writing something new. Prioritise fixes in your first run.

5. Track your changes — Keep a simple log of what you fixed, when, and how the metrics changed 2–4 weeks later. This builds your instinct over time.

6. Content clusters matter — If multiple high-traffic pages are about “funding for researchers in Europe,” consider bundling them into a series or guide. Clarity + GSC signal cluster opportunities if you’re looking for them.

---

Want more guides like this? Browse all AI Guides or get in touch →