Why the value of AI Search visibility cannot be reduced only to referral traffic and conversions, and what decades of marketing research can teach us about it.
I agree with Marie Haynes. Up to a point.
On October 21, Marie Haynes published an article provocatively titled GEO is a distraction. Knowing Marie and having followed her work for years, I was naturally interested in reading it. I have enormous respect for her knowledge, and for the way she approaches the evolution of Search, which makes it even more interesting when I find myself disagreeing with her.
Before explaining why, I should clarify that I certainly share Marie’s scepticism about the proliferation of GEO evangelists, self-appointed experts and platforms promising to revolutionize brand visibility through metrics whose business significance remains uncertain.
However, I think Marie’s article raises a more interesting question than whether companies should invest in GEO tools.
It concerns what we consider the purpose of SEO to be, and consequently, how we should measure its contribution to marketing.
Marie’s argument is supported by a substantial dataset. She analysed 12 months of Google Analytics 4 data from 100 websites, representing more than 145 million sessions. Across those websites, ChatGPT generated referral sessions equivalent to just 0.62% of Google organic traffic. Including other LLMs increased that figure to 0.73%.
The conversion figures were equally revealing. Among the 55 websites tracking conversions, ChatGPT visitors converted at 13.8%, compared with 3.02% for Google organic visitors. Nevertheless, Google’s much greater traffic volume meant that it generated approximately 27 times more conversions than all the LLMs combined.
These numbers provide a useful correction to the excessive enthusiasm surrounding AI Search. They also confirm something that I would expect most experienced SEO professionals to recognize: Google continues to be extraordinarily important for the acquisition of qualified organic traffic, and abandoning established SEO activities to chase ChatGPT citations would be an extremely questionable business decision.
I agree with Marie on this point.
I also agree with her that the emergence of agentic technologies deserves attention. AI systems capable of interacting with products, services, and transactional interfaces may substantially change how customers discover, evaluate, and purchase things. Preparing for these developments is a legitimate strategic concern.
My disagreement starts when referral traffic and directly attributable conversions become the principal evidence used to judge whether visibility within AI Search is worth pursuing.
To be fair, Marie does not explicitly claim that every marketing activity must result in a conversion. Her article includes considerations about user satisfaction, product quality, and the opportunities created by agentic search. But the comparison between the commercial value of Google and that of other LLMs relies heavily on measurable referrals and conversions.
And here I think we risk confusing the final economic purpose of marketing with the different mechanisms through which marketing creates value.
A marketing activity can contribute to future commercial results without being the channel through which the final transaction occurs. This is neither a radical proposition nor an attempt to justify spending money on an emerging technology.
It is one of the basic principles of marketing.
We seem to have forgotten something about marketing
One of the unintended consequences of digital marketing has been the tendency to redefine marketing according to what our analytics platforms can measure.
Website traffic became a primary objective because it could be counted. Conversions became the ultimate indicator of success because they could be associated with identifiable interactions. And as attribution technologies developed, we increasingly began treating the channel credited with a conversion as the channel responsible for creating its value.
The problem is that an observable event and the process that produced it are not necessarily the same thing.
Consider the distinction between brand building and sales activation developed by Les Binet and Peter Field in The Long and the Short of It, one of the most influential contributions to modern marketing effectiveness research.
Sales activation aims to influence people who are relatively close to making a purchase. Its effects are generally more immediate and easier to associate with observable behaviour. Brand building, meanwhile, works through the development and reinforcement of memory structures that increase the probability of a brand being considered when a relevant purchasing situation eventually arises.
Both contribute to business performance, but they operate through different mechanisms and timescales.
Binet and Field’s work is particularly important because it demonstrates why concentrating exclusively on short-term activation can undermine longer-term growth. Their research famously identified an approximate 60:40 balance between brand building and activation across their broader dataset, although the optimal proportions vary significantly between industries and business models.
I am not suggesting that companies should allocate 60% of their SEO budget to brand awareness. Such an interpretation would entirely miss the point.
What their research tells us is that marketing must address both existing demand and the conditions that influence future demand.
This connects directly with Kevin Lane Keller’s concept of customer-based brand equity, which concerns the differential effects that brand knowledge has on consumer responses. That knowledge includes awareness, but also the associations people establish between a brand and particular attributes, benefits, or situations.
It also connects with Byron Sharp and Jenni Romaniuk’s work on mental availability and Category Entry Points: the circumstances, motivations and needs that bring a category, and potentially specific brands within it, to someone’s mind.
Let’s consider something as ordinary as buying running shoes.
Someone decides to start running and asks ChatGPT for advice about preparing for their first five-kilometre race. The conversation initially concerns training frequency, injury prevention and how to build endurance. Eventually, the person asks whether specialized running shoes are necessary and which characteristics they should consider.
The AI assistant explains cushioning, stability, and the differences between running on roads and trails. It may also introduce examples of brands associated with running needs.
At this stage, the person has no immediate intention of purchasing anything. They are simply learning about an activity they have decided to pursue.
A few weeks later, however, they visit a sporting goods store. Looking at the available shoes, they recognize one of the brands previously mentioned and remember its association with the type of running they intend to do.
Perhaps they buy that brand. Perhaps they don’t.
Either way, if the earlier interaction influenced which products they noticed or considered, it would have performed a recognizable marketing function without generating a referral or an immediately attributable conversion.
This is exactly why the concept of Category Entry Points matters.
The situation that activates a purchasing category may arise long before the purchase itself, and the associations that influence choice may be constructed through encounters distributed across different environments.
Of course, the running-shoes scenario is illustrative, not evidence that an AI-generated recommendation actually caused a purchase. Establishing that would require appropriate research.
But its underlying marketing mechanism is well established.
If a brand becomes associated with a relevant need, it may increase its probability of being considered when that need becomes commercially actionable.
And this brings me to a deliberately provocative question.
If conversions are the only metric, should we cancel the podcasts too?
Suppose we decided that every digital marketing activity should demonstrate its value primarily through the volume of directly attributable traffic and conversions it generates.
Following that reasoning consistently would have some rather uncomfortable consequences.
Should a B2B company stop publishing a podcast because its listeners rarely visit the corporate website immediately after an episode? Should it abandon expert interviews on YouTube because the videos produce few identifiable leads? Should its executives stop sharing their knowledge on LinkedIn because most people reading their posts never click a link or request a demonstration?
We might even have to question the value of certain digital PR initiatives, educational campaigns and influencer collaborations, particularly when their objectives involve positioning, credibility or recognition rather than immediate sales.
Of course, I am intentionally pushing the argument to its logical extreme. Marie is not suggesting that businesses should abandon these activities, and I would not attribute such a position to her.
But the thought experiment illustrates the problem with treating directly attributable conversions as the principal criterion of marketing value.
More importantly, we have empirical research demonstrating that some of these activities can produce measurable effects beyond immediate transactions.
In Media in Focus, published in 2017, Binet and Field identified video advertising, both online and offline, as particularly effective for brand building, while paid search and email were especially effective for sales activation.
This does not mean that every educational video published on YouTube generates measurable brand equity. It means that using immediate conversions as the only measure of video effectiveness would ignore one of the principal mechanisms through which the format can work.
Podcasting provides another interesting example.
Nielsen’s research on emerging media, published in 2023, found that podcast advertising could increase brand awareness by 13 percentage points. Aided brand recall among people exposed to podcast advertising reached 71%, compared with 50% among non-exposed audiences in the reported comparison.
There is an important distinction between advertising within podcasts and producing a branded podcast or conducting expert interviews. We cannot automatically extend findings about the former to the latter.
Nevertheless, the research demonstrates that podcast-related marketing can produce effects that are relevant to brand strategy without those effects necessarily appearing as immediate website conversions.
The evidence becomes even more interesting when we consider social media and influencer marketing.
In October 2025, the Institute of Practitioners in Advertising published research based on 220 influencer campaigns from 144 brands across 36 sectors and 28 markets.
Its analysis showed that influencer marketing delivered short-term ROI broadly comparable to the all-channel average. However, in a smaller UK long-term analysis involving 18 campaigns, influencer marketing achieved an ROI index of 151, compared with 77 for standard paid social.
The researchers also found a substantial long-term multiplier for influencer activity.
We should be cautious about generalizing those results, especially given the relatively small sample used for the long-term comparison. But they illustrate how an activity’s apparent effectiveness changes when we examine a different timescale.
What might appear relatively modest through a short-term performance lens can contribute significantly to longer-term business outcomes.
And then there is thought leadership, which is particularly relevant to B2B businesses.
The 2025 Edelman–LinkedIn B2B Thought Leadership Impact Report, based on research involving nearly 2,000 professionals, examined the influence of people who participate in purchasing decisions without necessarily being the visible buyers.
Among these so-called hidden buyers, 81% reported that high-quality thought leadership helped them recognize challenges or opportunities they had not previously considered. Furthermore, 79% said they would be more likely to advocate for a vendor during an RFP process if that vendor consistently produced high-quality thought leadership.
These are self-reported perceptions and intentions, not experimental proof of additional sales. That distinction matters.
However, the findings highlight something that anyone involved in complex B2B sales should recognize not everybody influencing a purchasing decision becomes an identifiable lead.
Someone may read an article, watch an interview or listen to a podcast and subsequently introduce an idea into an internal discussion. They may recommend a company to colleagues or support its inclusion in a shortlist without ever appearing in the company’s CRM.
The commercial influence of those interactions is difficult to observe through traditional web analytics, but that does not make it irrelevant.
The broader question is whether these effects translate into actual business results. Research using econometric approaches suggests that they can.
For example, a Meta-commissioned study conducted by Nielsen, GfK and Nepa examined more than 3,500 advertising campaigns using marketing mix modelling. Across the analysed studies, longer-term effects accounted for approximately 60% of total modelled advertising ROI.
The estimates depend on the models employed, the specific industries examined and the assumptions underlying the analyses. Meta also has an obvious commercial interest in demonstrating the long-term effectiveness of digital advertising.
Even acknowledging those limitations, the findings reinforce a well-established lesson: measuring advertising exclusively through short-term outcomes can underestimate its contribution.
Now, does any of this prove that a brand mentioned in ChatGPT generates the same effects as a well-designed video campaign or a memorable podcast advertisement?
Absolutely not.
The available evidence supports the broader marketing principle, not a direct equivalence between formats.
We still need to investigate whether people notice brand references within AI-generated answers, whether those references create meaningful associations, and whether these associations influence future behaviour.
But the absence of that evidence should lead us to formulate better research questions, rather than dismiss a communication environment because it produces relatively few direct referrals.
If we accepted that reasoning for AI Search, consistency would require us to apply it to other activities whose value partly depends on delayed, indirect or difficult-to-attribute effects.
And I suspect very few experienced marketers would consider that a sensible approach.
Search has never been exclusively about capturing demand
There is another reason why I find a conversion-centred interpretation of SEO inadequate.
It assumes, at least implicitly, that Search primarily serves people whose information needs are already sufficiently developed to result in an identifiable action.
This is certainly true for some searches, especially transactional ones. Someone searching for “buy running shoes online” or “CRM software pricing” may already be relatively close to making a commercial decision.
But a substantial part of Search consists of exploration, learning, and progressive problem definition.
People investigate subjects they only partially understand. They compare possibilities before knowing which criteria matter. Sometimes, they discover that the problem they initially intended to solve is not the problem they have.
Google’s research on the Messy Middle described purchasing behaviour as an iterative movement between exploration and evaluation rather than a strictly linear journey.
In that environment, Search does more than capture existing demand. It can participate in shaping how demand develops.
Consider someone thinking about a summer holiday with their family.
Their initial interaction with an AI assistant may be a relatively generic question about Mediterranean destinations suitable for young children. They may not yet have decided where to travel, which kind of accommodation to choose, or even whether the holiday should take place in June or September.
During the conversation, Sardinia might emerge as an attractive possibility because of its beaches, family-oriented accommodation, and accessibility from their departure airport.
The user then investigates particular areas of the island, compares resorts with holiday rentals and begins evaluating specific hotels.
At this stage, the AI system may introduce travel brands or individual properties that the person had never previously considered.
The family may eventually book through an OTA, visit a tour operator’s website after a branded Google search, or contact a hotel directly.
The original AI interaction might never appear in the attribution data associated with that booking, even though it may have contributed to the choice of destination, accommodation type, or supplier.
Again, this is a hypothetical journey rather than an established causal relationship. But it illustrates how search environments can participate in the formation of preferences before the purchasing decision becomes explicit.
In traditional Search, such exploration has been distributed across articles, videos, comparison pages, forums, reviews and other resources.
Conversational AI systems can bring several of these activities into the same interface. A user can ask for an explanation, refine the question, explore alternatives, and request comparisons without necessarily leaving the conversation.
This changes the structure of the discovery experience, even if the fundamental informational needs remain recognizable.
The important distinction is between the ability of a brand to be discovered and its ability to influence the subsequent decision.
The former can be investigated through visibility measurements. The latter requires a different form of evidence.
But if we acknowledge that traditional Search can contribute to awareness, consideration, and positioning, there is no conceptual reason to assume that conversational Search cannot participate in those processes.
The real question is how frequently it does so, under which circumstances, and with what measurable effects.
AI Search visibility is not the objective. Relevant brand associations are.
At this point, I should clarify a potential misunderstanding, because defending the strategic relevance of AI Search does not mean defending the metrics currently promoted by many GEO platforms.
In fact, I think much of the current conversation around AI visibility suffers from a problem remarkably like the old obsession with rankings.
A company measures how often its brand appears in answers to a predefined collection of prompts, compares that frequency with competitors and treats any increase as evidence of progress.
But presence alone tells us very little.
We need to know whether the selected prompts represent meaningful information needs, whether relevant audiences actually encounter those answers, and how the brand is described when it appears.
This last point is particularly important.
In classical marketing, positioning concerns the place a brand seeks to occupy in the minds of its intended audiences relative to competing alternatives.
For example, a company may want to be recognized for specialized expertise, reliability, affordability, innovation or the ability to solve a particular type of problem.
These associations are strategically different, and greater visibility does not necessarily improve them.
A premium travel operator does not benefit from being repeatedly represented as a budget supplier. A software company specializing in regulated industries may derive little strategic value from being described merely as a generic solution with a large feature set.
AI Search introduces a new environment in which such associations can be communicated, reinforced, or distorted.
Unlike a traditional list of search results, an AI-generated response may explicitly characterize a brand, compare it with alternatives and explain why it is relevant to a particular situation.
Whether that characterization affects a person’s perception remains a matter for empirical research. But the representation itself is certainly something we can analyse.
This is why I consider the relationship between SEO, Brand Marketing, PR, Content Marketing and Social Media increasingly important.
A brand cannot assume that its preferred positioning will automatically be reflected in AI-generated answers simply because it describes itself in those terms on its website.
Its presence in the broader information environment matters.
Independent journalism, original research, professional commentary, customer experiences, specialist publications and community discussions can all contribute evidence about what an organization does and how it is perceived.
That does not mean that obtaining more mentions automatically improves AI visibility, or that independent references are guaranteed to influence model responses. The mechanisms vary between systems, and causal relationships are far from fully understood.
It does mean that the information environment in which a brand exists extends beyond its own domain.
The role of SEO is not simply to receive content from Brand Marketing or benefit passively from PR.
It can actively contribute to both disciplines.
Search research can reveal emerging needs, changing terminology, unanswered questions, recurring comparisons and the problems through which people approach a category. These insights can inform positioning, product communication, editorial planning and digital PR.
Conversely, the research, stories and independent evidence generated by these activities can strengthen the broader information environment from which search systems retrieve and synthesize answers.
This reciprocal relationship is important because it reflects how I believe SEO should operate, independently of the arrival of generative AI.
Its function has been evolving beyond isolated website optimization for years.
AI Search makes that evolution more visible, but it did not invent it.
What this changes in practical SEO strategy
None of this means we should start recommending significant investments in AI Search visibility to every business.
Marie’s findings should make us cautious, particularly when budgets are limited and existing organic acquisition channels remain commercially important.
The appropriate strategy depends on the business model, the audience and the role Search plays in the customer journey.
For a local business that primarily needs immediate enquiries, investing in traditional local SEO and improving its Google Business Profile may be substantially more valuable than purchasing sophisticated AI citation-monitoring software.
For a publisher dependent on advertising revenue generated by website visits, AI-generated answers that satisfy information needs without referring traffic may represent a real economic threat.
For a B2B company operating in a complex category with long purchasing cycles, however, there may be greater value in understanding how potential customers encounter and evaluate its brand across different search experiences.
The practical starting point should not be a list of prompts to track.
It should be the organization’s positioning and the information needs associated with it.
Consider, for instance, a company providing practical AI training for financial institutions.
Before investigating ChatGPT visibility, I would want to understand which situations make its services relevant to potential customers. That might include new regulatory requirements, concerns about responsible AI use, difficulties implementing AI across departments, or the need to develop practical capabilities among specific professional groups.
These situations can inform a map of Category Entry Points and corresponding search behaviours.
Traditional keyword research remains useful, but it needs to be complemented by an examination of the questions, comparisons and exploratory conversations through which people progressively understand these problems.
We should then examine how those needs are addressed across different environments, from Google Search and AI Mode to ChatGPT, Gemini, specialist publications and professional communities.
The objective is to understand where the company has opportunities to provide genuinely useful information and where its existing positioning is poorly represented.
Some opportunities may require better technical accessibility or clearer website architecture. Others may justify original research, expert content, case studies, industry partnerships, or more effective distribution through professional media.
This is where cooperation between SEO, Brand, PR, Product, and Social Media becomes operational rather than theoretical.
The resulting work should be evaluated according to the outcomes it is intended to influence.
For activities focused on demand capture, qualified traffic, leads and sales remain essential indicators.
For activities intended to strengthen positioning or future consideration, the measurement framework should also investigate brand recognition, relevant associations, consideration, and changes in audience behaviour.
AI visibility metrics can contribute to this analysis, provided we understand their limitations.
For example, Google’s Generative AI performance reports in Search Console provide impressions associated with URLs shown in AI Overviews, AI Mode, and generative AI features in Discover. This is useful first-party evidence of exposure opportunities, although it does not demonstrate that users noticed or remembered the brand (and it is indeed very limited information from Google, honestly)
Microsoft offers another, complementary perspective through the AI Performance report in Bing Webmaster Tools, introduced in February 2026. It reports how frequently website content is cited in AI-generated answers across Microsoft Copilot, Bing and selected partner experiences, including the pages cited and the grounding queries associated with those references.
Microsoft subsequently expanded the report with intent classifications, thematic groupings, and Citation Share. More recently, its Topics Portfolio beta, announced on October 13, 2026, introduced a broader view of thematic visibility, approximate topic size, competing cited sources, and potential query-level opportunities.
These developments are particularly interesting because they begin to connect AI visibility with the informational contexts in which content is cited. Nevertheless, the metrics require careful interpretation. Bing’s grounding queries are not the verbatim prompts submitted by users; its citation data is aggregated and sampled, and neither Citation Share nor topic-level visibility establishes that a brand is considered authoritative or preferred by its audience.
Google and Bing are also measuring different events: impressions in Google’s generative search experiences and citation activity across Microsoft’s supported AI experiences. Their figures should not be treated as directly comparable measures of audience reach.
Both reporting systems provide useful evidence about content visibility, but neither demonstrates human attention, brand memorability, or incremental commercial impact.
For platforms such as ChatGPT, third-party tools frequently depend on repeated prompts and sampled responses. These observations can help identify patterns in how brands are represented, but they are not equivalent to measuring real audience reach.
Nor should we interpret increases in branded searches, direct traffic, or sales as proof that AI visibility caused them.
These changes may result from numerous simultaneous marketing activities.
Where the potential investment is sufficiently large, brand-lift research, experimental designs or econometric analysis may help investigate incremental effects. Smaller organizations may have to rely on more limited evidence and correspondingly cautious conclusions.
The important thing is to define the hypothesis before measuring it.
In this context, I find Aleyda Solís’s three-layer framework for measuring AI Search Presence, Readiness and Business Impact particularly useful. It distinguishes between Presence, which examines whether and how a brand appears in relevant AI-generated answers; Readiness, which investigates the technical, informational, and credibility-related conditions that may explain that visibility; and Business Impact, which connects these observations with measurable outcomes while explicitly distinguishing observed data, proxy indicators, and modelled estimates. What I appreciate about this approach is that it neither treats visibility as proof of commercial value nor reduces the entire evaluation to directly attributable conversions. Instead, it provides a diagnostic framework in which different metrics answer different questions, and the strength of the conclusions must reflect the quality of the available evidence.
If a company’s objective is to become more strongly associated with responsible AI training among financial institutions, measuring how often its brand appears in arbitrary AI-generated answers is insufficient.
We need to understand whether the relevant audiences increasingly recognize the company in relation to that expertise.
And if the evidence eventually suggests that the activity is not producing meaningful effects, or that its resources would create greater value elsewhere, we should be prepared to change the strategy.
This is not an argument for investing in AI Search without demanding results.
It is an argument for demanding the right kinds of results.
What about the agentic future?
There is one final point where I think Marie’s argument and mine can converge.
She is right to emphasize the opportunities associated with agentic search, including technologies such as the Universal Commerce Protocol and WebMCP.
As AI systems become capable of performing more actions on behalf of users, the processes of discovery, evaluation, and purchase may increasingly occur without the familiar sequence of search result, website visit, and transaction.
This development has substantial implications for SEO.
Products and services will need to be understandable and accessible not only to human visitors but also, where relevant, to the systems interacting with them.
Accurate structured information, product availability, commercial conditions, functional interfaces and reliable data exchange can become increasingly important.
However, I would extend the reasoning beyond transactional capability.
An agent cannot recommend or purchase an offering that it has not identified as relevant to the task.
Before a transaction can occur, the system may need to discover available alternatives, interpret their characteristics, and evaluate their suitability.
This introduces a form of machine-mediated discoverability that is conceptually related to, but technically different from, human mental availability.
We should not confuse the two. An AI system does not remember brands in the psychological sense described by marketing research, and the processes through which it retrieves and evaluates information are not equivalent to human recognition or recall.
Nevertheless, both human and machine-mediated discovery can affect which brands enter a decision process.
The eventual commercial value will depend on whether those brands are appropriate for the task, competitive in their offering, and sufficiently credible to be selected.
So, even in an agentic environment, the familiar marketing questions remain relevant: what problem does the brand solve, for whom, under which circumstances, and why should it be preferred over alternatives?
Technical accessibility and transactional integration matter, but they are not substitutes for positioning, differentiation, or product quality.
Nor does the emergence of agentic capabilities make today’s informational and exploratory interactions irrelevant.
If anything, it reinforces the need to understand how brands become discoverable and considered before a transaction takes place.
Conclusion: perhaps we should bring marketing back into SEO
I appreciate Marie’s attempt to bring empirical discipline to a conversation often dominated by enthusiasm and questionable commercial promises.
Her analysis is a useful reminder that Google remains enormously important for organic acquisition, and that we should be sceptical of anyone presenting AI citations as a guaranteed source of commercial growth.
I also agree that SEO professionals need to understand the changes introduced by agentic technologies rather than concentrating exclusively on familiar ranking and traffic metrics.
Where I disagree is with the idea that relatively low referral traffic and directly attributable conversions are sufficient grounds for treating non-Google AI Search visibility as a distraction.
They are sufficient grounds for questioning its immediate acquisition value. They are not sufficient to determine its contribution to brand positioning, recognition, consideration, or future demand.
Marketing research has spent decades investigating precisely these distinctions.
We would not dismiss video advertising, podcasting, thought leadership or other brand-building activities simply because their effects are not always visible in conventional web analytics.
We should apply the same conceptual discipline to AI Search, while being equally demanding about evidence of its actual effects.
This does not mean that every AI mention is valuable, that visibility automatically creates memorability, or that every organization should dedicate substantial resources to monitoring LLM responses.
It means understanding that different marketing activities perform different functions, and that those functions require appropriate methods of evaluation.
I continue to believe that SEO is a marketing discipline concerned with how people discover, understand and evaluate information through search environments. Its relationship with Brand Marketing, PR, Content, Social Media and Product Strategy has become progressively more important, and not simply because ChatGPT appeared.
AI Search adds new interfaces, new intermediaries and new forms of information representation to an already complex ecosystem.
But the underlying marketing principles remain remarkably familiar.
The real danger is not that businesses might invest in AI Search visibility. It is that they might do so without a coherent marketing strategy or dismiss its potential contribution because the effects they should be investigating cannot be found in a referral report.
GEO becomes a distraction when it forgets that it is marketing. But dismissing AI Search because it does not yet produce enough conversions risks making exactly the same mistake, only from the opposite direction.