Why We Use AI as an Analyst, Not Just an Assistant

How we integrate large language models into our core SEO methodology — and what that means for the quality of insight your campaigns receive.

Most agencies that claim to "use AI" mean they've added a chatbot to their workflow. We mean something different: at Semola Digital, large language models sit at the centre of how we diagnose problems, interrogate data, and build strategy — not as a shortcut, but as a structural advantage.

This article explains exactly how that works in practice. If you're evaluating SEO partners, or simply want to understand what modern, AI-informed strategy looks like at the agency level, you're in the right place.

The Difference Between a Task and an Analysis

There's a critical distinction we draw internally between using AI for tasks and using it for analysis. It shapes everything about how we structure our work.

Most agencies operate at the task level. Our team operates at the analysis level — and that's where the real value is generated for our clients.

Three Pillars of Our AI-Informed SEO Process

Our methodology is built around three interconnected capabilities that AI makes faster, deeper, and more consistent across every client account.

Let's consider a case of cannibalisation:

Cannibalisation: Finding the Silent Traffic Leak

Keyword cannibalisation — where two or more of your own pages compete for the same search query — is one of the most damaging and least visible issues on established websites. It dilutes click-through rate, confuses search engines about which page should rank, and often explains persistent ranking plateaus that resist all other interventions.

Manually identifying cannibalisation across hundreds or thousands of queries is not feasible. Our approach uses a structured, multi-step analysis against your Search Console data to surface it systematically.

Our Detection Workflow

• Data Preparation: We export your last 90 days of Google Search Console query data — clicks, impressions, CTR, and average position — prioritised by impression volume.
• Pattern Surfacing: AI analysis identifies query clusters where multiple URLs are competing. Each case is assessed for severity: Critical (both pages ranking page one and splitting clicks), Medium, or Low.
• Case-by-Case Recommendation: For each cannibalisation pair, we drill into specific performance signals to determine the right resolution: consolidate, canonicalise, or differentiate.
• Developer-Ready Brief: Findings are translated directly into implementation briefs — canonical tag instructions, redirect maps, or content restructuring plans — ready for your development team.

Let's give you an idea on how we run this:

For cannibalisation analysis, we use the popular tool you're familiar with; the Google Search Console’s Search type: Web filter: Clicks, Impressions, CTR, and Position columns. We set date range to at least 3 months, and we export the data. We then compare with page structure: page title, meta description, all heading tags (H1–H6) in order (which we can grab by opening the page, and then DevTools, and run:

const headings = Array.from(document.querySelectorAll('h1, h2, h3'))
  .map(h => `${h.tagName}: ${h.innerText}`)
  .join('\n');

console.log(headings);

in the Console tab, and let Claude code run the show for us.

1. Claude, analyse this Search Console data and identify queries where multiple pages are competing..."
2. Drill into a specific case "For the cannibalisation between page A and B, what do you recommend — consolidate, canonicalise, or differentiate?"
3. "Develop a technical SEO brief for implementing the canonical tag fix on these pages."

Content Structure Analysis: Understanding Why Pages Underperform

When a page isn't ranking as it should — despite solid technical foundations and reasonable authority — the problem is usually structural. Either the page doesn't match the search intent for the query it's targeting, it covers too much ground and becomes ambiguous to Google, or a competing page on your own site (or a competitor's) simply does the job better.

We approach this with a direct structural comparison: extracting the heading hierarchy, intent signals, and topical coverage from two competing pages and analysing exactly where the divergence — or problematic overlap — lies.

What We Actually Analyse

Intent Alignment

Does each page match the dominant search intent — informational, commercial, or transactional — for the target query? A mismatch here overrides almost every other optimisation.

Topical Gaps and Overlap

What does Page A cover that Page B doesn't, and vice versa? Near-duplicate heading structures are a clear signal that consolidation will outperform trying to differentiate both.

Structural Recommendation

Based on the above: consolidate, differentiate, or identify the clearly superior page and invest exclusively in optimising that one. Specific headings to add, remove, or rewrite — not vague advice.

Action Plan

If consolidating: what should the merged page structure look like? If differentiating: what specific content changes remove the overlap? Every recommendation is actionable, not advisory.

Strategy Brief Generation: Senior Thinking at Speed

One of the highest-value outputs in any agency engagement is the strategic brief — the document that defines priorities, phasing, KPIs, and the logic behind every recommendation. Traditionally, this is a senior strategist's half-day effort requiring deep synthesis of technical audit findings, competitive context, resource constraints, and business goals.

We haven't replaced that thinking. We've made it available to every client at the start of their engagement, rather than weeks in.

What Goes In, What Comes Out

The output of this process is approximately 80% complete as a first draft. The remaining 20% — the decisions that require human judgment about client-specific nuance, relationship context, and market knowledge — is where our senior team adds the most value. That's the right ratio.

How We Get Reliable Reasoning from AI

Working with AI at the analysis level requires a different discipline than working with it for tasks. The quality of reasoning you get back is a direct function of how you structure the question. Our team is trained on three techniques that materially improve output reliability.

Chained Prompting: Broad to Specific

We never ask a single large question when a sequence of focused questions produces better reasoning. For cannibalisation: first surface the patterns across the full dataset, then drill into individual cases. For ranking drops: first identify which queries declined and when, then reason about cause, then isolate the most likely explanation given what changed on the site.

Explicit Reasoning Chains

We ask AI to reason through a problem step-by-step before concluding. The difference in output quality is significant — especially for decisions where the answer isn't obvious. We also explicitly ask the model to flag its assumptions and confidence level on each recommendation. This produces more honest output that we can review more critically.

Adversarial Checking

For any strategic recommendation, we ask: what's the strongest case against this? Where could this recommendation be wrong? This surfaces edge cases and risks that the initial analysis might have passed over — the equivalent of having a junior analyst present their recommendation to a sceptical senior partner.

What This Means for You as a Client

The practical effect of this methodology is measurable at every stage of an engagement.

Faster Time-to-Insight

Cannibalisation analysis that previously required multi-day manual review now arrives at your kickoff meeting. Your 90-day strategy brief is ready before the onboarding call ends.

More Coverage, Less Guesswork

We analyse the full picture of your Search Console data — not a manually sampled subset. Issues that would have been missed are now systematically surfaced.

Consistent, Explainable Recommendations

Every recommendation has a documented reasoning chain. You'll never receive a recommendation from us that can't be fully explained — because the analysis itself requires explicit reasoning as a precondition.

Senior Attention on the Right Problems

When analytical work runs faster, our senior strategists spend more time on the decisions that genuinely require experience and judgment — not pattern recognition that a model handles well.

Oladoyin Falana

Founder, Technical Analyst

Oladoyin Falana is a certified digital growth strategist and full-stack web professional with over four years of hands-on experience at the intersection of SEO, web design & development. His journey into the digital world began as a content writer — a foundation that gave him a deep, instinctive understanding of how keywords, content and intent drive organic visibility. While honing his craft in content, he simultaneously taught himself the building blocks of the modern web: HTML, CSS, and React.js — a pursuit that would eventually evolve into full-stack Web Development and a Technical SEO Analyst.

Follow me on LinkedIn →