What We Believe

Discovery, research, comparison and shortlisting increasingly happen inside a single AI conversation, before a click ever takes place. Organic traffic is being redistributed across channels that existing analytics wasn’t built to measure.

This isn’t a demand problem. People are still seeking information, recommendations and advice, but where they’re doing that has changed. For many consumers, it has moved upstream into the AI conversation layer. The funnel hasn’t gone away; it has compressed.

Our priority has always been ensuring clients’ brands are present and recommendable wherever decisions are being made. That used to mean Google. Increasingly, it also means ChatGPT, Claude, and the platforms that sit between a consumer’s question and their first click. Zero-click is a visibility problem, not a demand one.

Optimising for “AI” without specifying which platform is like optimising for “social media” without distinguishing between LinkedIn and TikTok. ChatGPT, Claude, Perplexity, Google AI Mode, Google AI Overviews and emerging agentic platforms are distinct environments with different retrieval mechanisms, trust dynamics, user demographics, business models, and structural access rules.

Models add a further layer. Different models have different training data, knowledge cutoffs, reasoning behaviours, and citation tendencies. A brand prominent in pre-2024 training data will be represented very differently from one that emerged after a model’s knowledge cutoff, regardless of how well-optimised their current content is.

Platform politics directly affect visibility too. Amazon blocks all OpenAI crawlers. OpenAI began rolling out advertising in early 2026, with evidence its algorithm had been adjusted before the announcement. These decisions affect which brands appear where, and how users perceive each platform. Where you’re cited matters as much as whether you’re cited.

The way people interact with AI platforms is fundamentally different from how they interact with search engines, and conflating the two leads to bad strategy. There is no keyword planner equivalent for AI interactions, and anyone presenting prompt volume data with confidence should be challenged on their methodology.

People don’t fire off a single prompt and move on. They have multi-turn conversations with LLMs; refining, redirecting, going deeper. A single session around a purchase decision might span dozens of exchanges, with the context in which your brand might be mentioned shifting throughout.

AI prompts are natural language, highly specific, and shaped by personal context. Comprehensiveness, context-sensitivity and clear reasoning matter far more than keyword targeting. You cannot see prompts the way you can see search queries, and you cannot optimise for them the same way. Strategy has to be built on understanding audiences deeply enough to anticipate the contexts in which they’d ask about you.

We are not starting from scratch. Technical SEO fundamentals (structured data, clean semantic HTML, entity mapping, crawlability) remain essential for AI discoverability. The pages that rank well in organic search are overwhelmingly the pages that get cited in AI responses, because many AI systems ground their answers in live web searches that still hit Google and Bing indexes.

The fan-out query mechanism makes this especially clear. When an LLM breaks a prompt into sub-queries and searches for grounding information, backlinks, domain authority, and rankings are doing exactly the same work they always did.

What AI search adds rather than replaces: prompt-aware content structure, self-contained sections that work in isolation, and a focus on topical comprehensiveness over keyword targeting. The additional work is focused and deliberate, not a complete rebuild. The best foundation for AI visibility is strong SEO, strong content, and strong Digital PR. GEO is a layer on top.

AI systems are actively selecting sources they can confidently cite, and genuine E-E-A-T signals are a primary filter. Trustworthiness, expertise, experience, and authority matter more in an AI search world, not less, and they need to be present throughout the customer journey, not just on a homepage.

The issue we see most often is brands nodding to E-E-A-T on their homepage or blog while leaving product pages, category pages, and service pages thin and generic. E-E-A-T needs to be present wherever a customer or an AI agent shortlisting options might evaluate you.

Digital PR plays a direct role here. E-E-A-T signals built through earned media coverage, expert commentary in licensed publications, and authoritative third-party endorsements directly influence AI citation likelihood. These signals feed both the knowledge banks that LLMs draw on and the web search results they retrieve in real time. Demonstrate expertise and trustworthiness everywhere you might be evaluated.

Technology changes faster than human behaviour. Audience understanding (motivations, anxieties, decision triggers) is what produces content strategies and brand positions that AI systems find worth citing. You cannot reverse-engineer this from prompt data you cannot see.

Most industry commentary on AI search is written by practitioners who understand the implications of model updates and capability benchmarks. But most consumers just know they’re “using ChatGPT.” Strategy needs to reflect where audiences actually are, not where we are.

Consumers may be comfortable using AI for low-involvement tasks (renewing car insurance, comparing broadband) but will likely retain a strong preference for human involvement in higher-stakes or emotionally charged decisions. Understanding where a client’s product sits on that spectrum is essential. Significant segments of the population are also AI-sceptical or actively opposed, varying by age, sector, and geography. Start with who is using AI, how, and for what – not the technology itself.

Being cited is only valuable if you’re being cited positively. An AI system that mentions your brand alongside poor reviews or unresolved complaints is doing active damage – and because the consideration phase increasingly happens inside the AI conversation, that negative sentiment lands before a potential customer ever reaches your site.

This extends to hallucinations. AI systems can and do misrepresent brands, products, and prices. Structured, complete, accurate product and brand data is the best defence, because AI systems are less likely to fabricate what has been explicitly provided.

Trust dynamics are shifting too. AI systems prioritise authentic community content (Reddit threads, independent reviews, forum discussions) over brand-owned marketing copy. Post-purchase experience and customer service quality now directly influence which brands get recommended. A brand’s online reputation; the sum of everything being said about them across the web, is their most important AI marketing asset.

Standard SEO metrics do not capture whether a brand is being cited, how it’s being characterised, or which AI platforms are including or excluding it. The metrics available for AI visibility are inadequate, inconsistently defined, and frequently misrepresented by vendors. There is no equivalent of Search Console for LLM interactions.

We have developed a confidence framework that tiers AI search metrics by reliability and actionable value. High-confidence metrics (fan-out query rankings, LLM bot crawl frequency, branded search demand) form the core of reporting and strategy. Mid-confidence metrics provide quarterly directional context. No-confidence metrics (proprietary “AI visibility scores,” citation position rankings, exact keyword matching) we ignore entirely.

We can measure more than nothing but less than everything. Building the right framework now matters because if and when standardised tooling arrives, those already tracking the right signals will be ahead. Track what is reliable, use what is directional with appropriate context, and ignore what misleads, regardless of how impressive the vendor dashboard looks.

The current AI search landscape (AI Overviews, citations, conversational discovery) is the intermediate state. Agentic AI, where systems autonomously research, shortlist, and complete purchases on behalf of users, is the step-change ahead.

We are firmly in an advisory-before-autonomous phase. Consumers increasingly use AI to assist with research, but very few trust it to buy without oversight. The gap between “help me decide” and “buy for me” will persist for most categories for several years. The immediate opportunity is ensuring your brand is well-positioned at the point where AI-assisted research hands off to a human final decision.

Google’s agentic checkout, announced in January 2026, already lets US shoppers complete purchases within AI Mode without leaving the interface. Zero-click commerce has started. The infrastructure for an agentic world; structured product data, accurate brand information across platforms, community reputation, machine-readable content, is being built now. Brands that treat this as a future concern will find competitors have already laid the foundations.

AI search is moving faster than any framework can fully capture. The only honest response is to keep testing, keep publishing what we find, and resist the pull of confident-sounding claims that aren’t grounded in evidence. For an agency advising clients in this space, experimentation isn’t optional… it’s the thing that stops us blindly buying into hype.

Experimentation is active across our SEO, Content, and Digital PR teams, with Paid Media joining as channel access develops. The work spans controlled technical experiments, content structure testing, and Digital PR approaches designed to test citation and reputational outcomes across AI platforms.

We publish what we find – not selectively, and not only when the results are flattering. The measurement gap will only close if practitioners share real findings, and we would rather contribute to that conversation than wait for someone else to run the research we needed. When we make a recommendation, we want it to be traceable to something we have actually tested, not to a vendor claim or an industry consensus that formed before the evidence did.