generative engine Optimisation is no longer a side project—it sits at the centre of how our team plans every client campaign. This week’s batch of industry research confirms that AI Search visibility demands sharper landing pages, cleaner keyword maps, and a fundamentally different understanding of how large language models choose which brands to cite. Here is what matters right now and what we are doing about it.

Key Takeaways

• Landing pages built around a single conversion action consistently outperform multi-purpose pages in both traditional and AI-driven search.
• Keyword cannibalisation now hurts AI citations as well as organic rankings, making regular audits non-negotiable.
• Google’s latest spam research uses pattern-level clustering to catch AI-generated video spam at scale.
• Ecommerce stores must restructure product content specifically for LLM consumption to remain visible in agentic commerce.
• ChatGPT’s Thinking mode and Instant mode cite almost entirely different sources—only 25% overlap—so a single GEO strategy is not enough.

Single-Action Landing Pages Still Win the Conversion Race

Every landing page we build for clients follows one rule: one page, one job. That principle is reinforced by Semrush’s updated guide to landing page best practices, which outlines seven core tactics including matched messaging, clear CTAs, and minimal navigation distractions. In a GEO context, pages with tightly scoped intent also tend to be the ones LLMs pull structured answers from. We are now auditing every client landing page to ensure the primary heading, supporting copy, and schema markup all reinforce a single query intent.

Keyword Cannibalisation Now Threatens AI Citations Too

When two pages on the same domain compete for the same keyword, Google has always struggled to pick a winner. The damage now extends further. As detailed in Semrush’s cannibalisation guide, competing pages can confuse AI models that rely on consistent topical authority signals when deciding which source to cite.

Our standard workflow now includes quarterly cannibalisation audits using a combination of Search Console query overlap reports and site-level clustering tools. Fixes typically involve:

• Merging thin competing pages into a single authoritative resource.
• Adding canonical tags where consolidation is not practical.
• Rewriting title tags and H1s to differentiate intent clearly.

Google Targets AI Video Spam With Pattern-Level Detection

Google’s research team has moved beyond flagging individual pieces of AI-generated spam. According to reporting on Google’s new pattern-level AI spam detection, the system now clusters accounts exhibiting coordinated behaviour rather than analysing videos in isolation. For our clients producing legitimate video content, this is a positive signal: authentic, well-structured video assets should gain relative visibility as spam networks get suppressed. We are advising every ecommerce and service-based client to invest in original video with proper metadata and channel authority signals.

Ecommerce Stores Must Optimise Directly for LLMs

Agentic commerce—where AI assistants research, compare, and even purchase products on behalf of users—is arriving faster than most retailers expected. Semrush’s ecommerce AI SEO playbook lays out the structural changes needed: detailed product specifications in machine-readable formats, consistent use of product schema, and content that answers comparison queries head-on.

We have already started restructuring product pages for several retail clients. Key actions include adding FAQ blocks that mirror natural language queries, enriching structured data with GTIN and brand-level markup, and writing product descriptions that read as factual summaries rather than marketing copy.

ChatGPT’s Two Reasoning Modes Cite Different Sources Entirely

This is the finding with the most immediate strategic impact. New research from Semrush reveals that only 25% of cited sources overlap between ChatGPT’s Thinking mode and Instant mode. Thinking mode favours in-depth, technically rigorous content. Instant mode leans towards concise, well-structured pages that deliver quick answers. Treating ChatGPT as a monolithic system is a mistake we see competitors making constantly.

Our response: we now create two content tiers for priority topics. A long-form, data-rich pillar page targets Thinking mode citations, while a shorter, direct-answer companion page is designed for Instant mode. Both are interlinked and share consistent entity references so the domain’s topical authority remains unified.

The direction is clear. GEO requires granular, mode-aware content strategies, cleaner site architectures, and landing pages that do one thing exceptionally well. Every tactic above is already part of our active client delivery—these are not future plans, they are current practice.

Frequently Asked Questions

What is Generative Engine Optimisation and why does it matter for UK businesses?

Generative Engine Optimisation (GEO) is the practice of structuring website content so AI-powered search tools like ChatGPT and Google’s AI Overviews cite and recommend it. UK businesses that ignore GEO risk losing visibility as more consumers use AI assistants to find products and services.

How do web designers prevent keyword cannibalisation on large sites?

Regular audits using Search Console data and clustering tools identify pages competing for the same queries. Fixes include merging overlapping content, applying canonical tags, and rewriting page titles to target distinct intents.

Why does ChatGPT cite different sources in Thinking mode versus Instant mode?

Each reasoning mode processes and prioritises content differently—Thinking mode favours depth and technical detail, while Instant mode prefers concise, structured answers. Marketers need separate content assets optimised for each mode to maximise AI citation coverage.

How do ecommerce stores optimise product pages for AI search?

Stores should use comprehensive product schema markup, write specification-rich descriptions in plain language, and add FAQ sections that mirror natural language queries. These structural changes help LLMs accurately parse and recommend products during agentic commerce interactions.