For most of my career in SEO, success has been measured in a remarkably simple way: did you rank? Whether I was speaking to a business owner, an in-house marketing manager or another agency, the conversation usually returned to position. Had the website moved above its competitors? Was its share of search visibility growing? Were those rankings translating into traffic, enquiries and revenue? It was a logical way to evaluate performance because the relationship between rankings and commercial opportunity was relatively easy to understand. Google organised webpages, people selected which results to visit, and the businesses occupying the strongest positions generally received a larger share of the attention.
That model has not disappeared, nor have the principles behind it become obsolete. Technical SEO, relevant content, backlinks, user experience and authority remain fundamental to being discovered online. Google has made clear that its core SEO and quality systems continue to underpin its generative search features, while pages still need to be crawlable, indexed and eligible to appear in Search before they can participate in experiences such as AI Overviews and AI Mode. There is no sensible argument for abandoning SEO when the systems producing AI answers still depend heavily on the infrastructure, content and quality signals that SEO helps businesses improve. [1]
What has changed is the point at which the competition ends. For years, appearing prominently within a list of results was treated as the finish line, but AI search introduces another decision after retrieval. A website can be discovered, understood and considered relevant without its information being used in the final response. Another source may then be chosen because it provides a clearer explanation, contributes first-hand evidence, carries stronger signals of trust or offers an original insight that improves the answer. Rankings still influence whether content enters the conversation, but they no longer fully explain which sources shape that conversation.
This distinction sits at the heart of what I call Search Everywhere Optimisation. When I wrote Found, Trusted, Chosen, one of my central arguments was that search no longer happens exclusively inside Google. Customers move between search engines, AI assistants, social platforms, marketplaces, review websites, online communities and industry publications as they research a problem and decide who deserves their trust. They may discover a company through ChatGPT, validate it through Google, review the founder’s experience on LinkedIn, read customer opinions and return directly several days later. Trying to compress that journey into a single channel ignores how modern decisions are actually made.
AI has accelerated this shift because it increasingly takes on part of the evaluation that people previously performed themselves. A traditional search engine presents a range of possible sources and asks the user to choose between them. An AI assistant attempts to retrieve information, interpret it, compare it with other information and construct a response before the user visits a website. Although the underlying systems vary considerably between Google, ChatGPT, Perplexity, Gemini and Microsoft Copilot, the commercial consequence is similar. Businesses are no longer competing only for the click. They are competing to influence the answer that comes before it.
That change is already commercially meaningful. Over the past 12 months, AI assistants have generated or influenced more than £100,000 in new business for Fly High Media and contributed to a sales pipeline worth more than £500,000. Those figures include prospects who arrived through identifiable AI referral traffic, people who explicitly told us they had used ChatGPT or another assistant during their research, and opportunities where AI influenced the discovery process before a different channel completed the conversion. I would not claim that every pound can be attributed perfectly, because customer journeys rarely provide that level of certainty, but the scale and quality of the opportunities make one thing clear: AI visibility is no longer an experimental metric sitting on the edge of an SEO report. It is already influencing commercially valuable decisions.
The challenge for marketers is therefore not to choose between traditional SEO and AI search. It is to understand how they fit together. SEO helps information become eligible, accessible and discoverable. AI systems then make additional judgements about which of the available sources are useful enough to include. That means the question businesses need to ask is changing from “How do we rank first?” to “If an AI system had to answer this question, would it confidently use our work as evidence?”
Ranking gets you retrieved. Evidence gets you chosen.
Traditional SEO helps information enter the consideration set. AI search introduces another decision by determining whether that information is sufficiently useful, trustworthy and relevant to become part of the answer.
Why rankings are no longer the finish line
For much of Google’s history, the division of responsibility between the search engine and the user was relatively clear. Google attempted to organise the web, evaluate relevance and rank the pages it believed were most likely to satisfy a query. The individual then opened one or more results, compared different views and decided what to believe. Even when featured snippets and other enriched results began answering questions directly within the search interface, users still understood that the underlying experience was built around a ranked collection of documents.
Generative search changes that relationship because the system is increasingly expected to synthesise an answer rather than simply identify where one might be found. This requires more than relevance. The system may need to interpret an ambiguous request, break it into related questions, retrieve information from several sources and combine the results into a response that is both coherent and defensible. Google describes using Retrieval-Augmented Generation (RAG)* to ground responses in relevant and current pages from its Search index. It also describes query fan-out*, where a model generates several related searches to explore different aspects of a request before producing an answer. [1]
Consider someone asking for the best customer relationship management platform for a growing professional services company. A traditional results page can present product websites, review platforms, comparison articles and forum discussions without resolving the disagreements between them. The user retains responsibility for deciding which criteria matter and which publisher deserves attention. An AI response is expected to do more of that work. It may need to understand the size of the company, the complexity of its sales process, the importance of integrations, its likely budget and whether ease of use matters more than customisation. It then has to select information that supports a useful recommendation without pretending there is one universally correct answer.
That is why AI citation selection should not be treated as a simple replacement for ranking. Retrieval establishes relevance, but generation introduces questions of usefulness, compatibility and confidence. An article may be relevant to the original prompt yet add little to the final answer because it repeats information already supplied by stronger sources. Another may be authoritative but too broad to support the specific detail being discussed. A third may introduce a useful statistic, comparison or first-hand observation that gives the response something it could not obtain elsewhere. Selection can therefore occur at the level of ideas and passages, not only at the level of whole webpages.
The original research that introduced Retrieval-Augmented Generation described a model combining information stored during training with external, retrievable knowledge. One of its aims was to improve performance on knowledge-intensive tasks while making it easier to update information and provide provenance for answers. The systems used by commercial search products have advanced significantly since that 2020 paper, and it would be wrong to assume they all work in exactly the same way, but the basic principle remains important: retrieving external evidence gives a language model access to information that may be more current, specific or verifiable than its internal knowledge alone. [2]
ChatGPT Search similarly uses web search to provide timely answers with links to relevant sources, while Google’s generative experiences surface clickable links supporting their responses. Microsoft’s reporting now exposes which pages are being cited across Copilot, Bing’s AI-generated summaries and selected partner integrations. These products are not identical, but each reflects the same broader development: the web is no longer only a destination to which users are sent. It is also an evidence layer used to construct answers inside other interfaces. [3][4]
This is also why I am cautious when people claim AI citation behaviour is random. There is undoubtedly variability. Results can change as prompts, users, locations, product versions and available information change, while generative systems may not produce identical answers every time. Yet variability is not the same as meaninglessness. The presence of retrieval systems, search indexes, quality controls and citation reporting suggests there are processes governing which information becomes available and which sources are displayed. We may not know every weighting, but we can still identify characteristics that make content more useful within those processes.
Equally, I would avoid claiming that AI simply rewards extractability instead of quality. Poorly written but conveniently formatted content is not automatically better evidence than a deeply researched source. Google explicitly says there is no requirement to divide content into tiny chunks, rewrite it in a special style, create an llms.txt file or add a unique form of structured data to appear in its generative search experiences. Its guidance remains focused on technical accessibility, people-first usefulness and non-commodity content informed by real expertise. [1]
The more defensible interpretation is that quality and clarity reinforce one another. An original piece of research becomes more useful when the method and findings are explained clearly. An experienced practitioner becomes more citable when their conclusion can be understood without guesswork. A comparison becomes easier to reuse when the criteria are consistent and the data is transparent. Structure does not replace substance, but it can make substance easier to recognise.
In traditional SEO, you are competing for clicks. In AI search, you are competing to be believed.
How AI decides what is useful enough to repeat
Nobody outside the companies developing these systems can provide a complete formula for citation selection. Google, Microsoft, OpenAI and other providers disclose useful elements of their approaches, but they do not publish every model, weighting or quality threshold. Any article claiming to reveal a universal AI citation algorithm should therefore be treated cautiously. What we can do is combine official guidance, information retrieval research and repeated practical observation to develop a more useful way of thinking about the problem.
The idea I keep returning to is confidence. I do not mean a visible percentage showing that a system is 87 per cent certain a claim is correct. I mean the broader process of reducing uncertainty until information becomes useful enough to include. A source does not need to be perfect, and different queries demand different standards, but an AI-generated response needs some basis for preferring one claim over another, reconciling disagreement or deciding which detail adds value.
Google’s own helpful content guidance provides an instructive parallel. It explains that its automated systems use a mixture of factors to identify content demonstrating experience, expertise, authoritativeness and trustworthiness, commonly abbreviated to E-E-A-T, while emphasising that trust is the most important aspect. Google is also careful to explain that E-E-A-T is not one specific ranking factor and that quality rater scores do not directly determine rankings. The value of the framework lies in helping creators understand the kinds of characteristics associated with helpful and trustworthy information, not in providing a score that can be manipulated. [5]
I believe the same distinction is useful when thinking about AI search. Confidence should not be reduced to a new technical score or marketed as a guaranteed ranking factor. It is better understood as the outcome produced when several qualities reinforce one another. A recognised author can improve confidence because their experience gives the claim context. Independent references can improve confidence because the expertise is not being asserted solely by the person who benefits from it. Original evidence can improve confidence because readers can examine how a conclusion was reached. Clear language can improve confidence because the meaning is less likely to be misinterpreted.
This leads to what I call the Confidence Model. It is not an official framework used by Google or any other platform. It is my practical interpretation of the qualities that repeatedly appear when I study sources used across AI search and compare them with the principles described in official documentation.
The first quality is authority, although authority should not be confused with a third-party domain metric. Tools such as Ahrefs and Moz provide useful comparative measurements, but an AI platform is not citing a source because it has achieved a particular commercial SEO score. Authority is better understood as the extent to which a person or organisation is recognised as a credible source for the subject being discussed. That recognition may be demonstrated through first-hand experience, specialist qualifications, original research, consistent publishing, relevant links, editorial coverage, professional profiles, customer feedback and other independent evidence across the web.
This is where the Experience element of E-E-A-T becomes especially significant. Artificial intelligence can summarise publicly available information at extraordinary speed, which means another generic summary adds very little to the supply of knowledge. A person who has implemented the strategy, tested the product, treated the patient, managed the campaign or built the company can contribute observations that are not available from surface-level synthesis. Their advantage is not that they are human by default, but that they possess evidence created through action. Google’s current guidance for generative search explicitly recommends unique, expert-led and non-commodity content, including first-hand reviews and points of view based on genuine experience. [1]
The second quality is clarity, which includes what some people describe as extractability. The information needs to be expressed in a form that can be interpreted accurately and used in the context of the question. That does not mean every paragraph must be short or every answer must appear inside a table. Google specifically warns against mechanically dividing content into tiny chunks for AI, and there is no ideal page length. Clarity is a property of the thinking rather than a formatting trick. A 4,000-word essay can be clear when the argument progresses logically, terminology is explained and each section fulfils a defined purpose. A 400-word page can be unclear when it uses vague claims, inconsistent terminology and headings that fail to match the content beneath them. [1]
When I review a page with AI visibility in mind, I ask whether a relevant passage would still make sense if it were retrieved without the entire article surrounding it. Does it define important terms? Does it identify the subject unambiguously? Does it explain where a figure came from? Does it distinguish between proven fact and professional interpretation? These questions improve the content for people as much as they improve its usefulness to retrieval systems. Someone should not need to decode an article before they can trust it.
The third quality is corroboration. In a world overflowing with confident claims, independent support matters. If a company says it is the leading specialist in its field, that statement tells us little on its own. If trade publications, customers, conference organisers, recognised experts and reputable directories independently associate that company with the same field, the claim becomes easier to validate. This does not mean every source needs to agree or that consensus is always correct. It means important claims become more defensible when readers can understand the evidence behind them and compare them with other credible perspectives.
Corroboration also explains why PR, brand building, reviews, podcasts and expert contributions cannot be separated neatly from SEO anymore. A podcast appearance may not always provide a followed backlink, but it creates another public connection between a person and their expertise. A journalist quoting proprietary research helps validate both the data and the organisation that produced it. A detailed customer review contributes first-hand evidence of the experience a business provides. These signals have value beyond referral traffic because they help create a consistent account of who the organisation is and what it should be known for.
The fourth quality is originality. This may be the most important commercial opportunity created by the growth of generative AI. Average content has become cheaper and faster to produce, which means the internet will not suffer from a shortage of competent explanations. What remains scarce is information that could only have been created by a particular person or organisation. Proprietary data, original experiments, transparent case studies, unusual comparisons, first-hand experience and distinctive frameworks add something to the available knowledge rather than merely restating it.
Originality should not be confused with disagreeing for the sake of attention. A contrarian claim unsupported by evidence is not automatically more valuable than a well-established conclusion. The strongest original work usually begins with a real observation, demonstrates how the evidence was gathered and explains where uncertainty remains. It gives other people something they can inspect, challenge and build upon. That is how an idea becomes citable rather than merely provocative.
These four qualities work together. Authority without clarity can produce information that is credible but difficult to use. Clarity without authority can make an unsupported claim sound convincing. Corroboration without originality can create another summary of what everyone already knows. Originality without evidence can become speculation presented as insight. Confidence grows when credibility, communication, validation and contribution reinforce one another.
AI can make content sound informed without making it more trustworthy.
Fluency is not evidence. The competitive advantage belongs to businesses that can support clear communication with first-hand experience, transparent data and a reputation that extends beyond their own website.
What businesses should change
The obvious response to a new search environment is to look for a new checklist. Every technological shift produces a market for shortcuts, and AI search has already generated discussions about special files, citation schemas, prompt manipulation, content chunking and entirely new categories of optimisation. Some of those discussions are useful, particularly when they encourage better technical accessibility or measurement, but they can also distract businesses from the more demanding work that produces durable authority.
Google’s current position is notably conservative. It says established SEO practices remain relevant, that no special optimisation is required for its AI features and that publishers should concentrate on unique, valuable and people-first content. It specifically advises that llms.txt, tiny content chunks, AI-specific rewrites, artificial mentions and excessive structured data are not necessary for visibility in Google’s generative search. This guidance applies to Google rather than every AI platform, but it provides a useful warning against assuming that every new technology requires a completely separate publishing strategy. [1]
The first practical change I would make is to stop treating the website as a publishing target and start treating it as an organised expression of what the business knows. A useful knowledge base does more than target a collection of keywords. It explains services accurately, answers commercial questions, documents methods, presents original findings and connects related subjects so that both people and machines can understand the organisation’s depth of expertise. Keyword research still helps identify demand, but it should not be the only source of ideas. Sales calls, customer objections, internal data, failed projects, successful experiments and specialist employees often contain insights no keyword tool can reveal.
The second change is to make experience visible. A sentence claiming that a company has decades of combined expertise is weaker than a case study explaining the problem, constraints, decision, implementation and result. An author page listing a job title is weaker than one connecting the writer to relevant projects, qualifications, appearances and work. An ecommerce review that paraphrases the manufacturer is weaker than one showing how the product performed in real use. Google’s people-first guidance asks creators to think about who made the content, how it was produced and why it exists, while encouraging clear authorship where readers would reasonably expect it. [5]
The third change is to invest in reputation beyond the domain. A business cannot establish independent authority entirely through claims published on its own website. Editorial coverage, trusted directories, professional bodies, customer communities, podcasts, events and expert commentary help create a wider record of the organisation’s relevance. This should not become an exercise in scattering artificial mentions across low-quality websites. Google explicitly warns that inauthentic mentions do not provide the shortcut some people imagine. The objective is to earn genuine recognition in the places where the audience and industry already pay attention. [1]
The fourth change is to create assets worth referencing. A clear statistic supported by transparent methodology is easier to cite than an unsupported prediction. A detailed comparison using consistent criteria is more useful than a list of products assembled from affiliate feeds. A framework built from years of practical experience can give a complex problem a memorable structure. A calculator, benchmark, template or dataset may become more useful than another opinion article. The point is not to add features for the sake of appearing sophisticated. It is to contribute something that improves the quality of the answer available to customers, journalists, search engines and AI systems.
None of this means abandoning good writing. One risk in the current conversation is that marketers begin stripping all personality and nuance from content because they believe machines prefer sterile answers. Google’s guidance does not support that conclusion. It recommends unique viewpoints and warns that content does not need to be rewritten specifically for AI systems. A memorable analogy, an unexpected example or a distinctive argument can improve understanding while making the work recognisably yours. The objective is not to sound like a database. It is to communicate knowledge with enough precision that it can be trusted and enough originality that it deserves to be remembered. [1]
The best AI search strategy is not to create content that looks optimised for AI. It is to create knowledge that would weaken the answer if it were left out.
Measuring AI visibility without losing sight of revenue
Until recently, much of the conversation about AI visibility depended on manual testing and screenshots. A brand would appear in an AI response, someone would share the result on LinkedIn and the mention would be celebrated without any clear understanding of its commercial effect. Manual monitoring remains useful because it can show how platforms describe a brand and which competitors appear alongside it, but measurement is beginning to move beyond anecdotal evidence.
Microsoft introduced AI Performance within Bing Webmaster Tools in public preview in February 2026. It reports total citations, average cited pages, grounding query phrases, page-level citation activity and trends across Microsoft Copilot, AI-generated Bing summaries and selected partner integrations. Microsoft is careful to explain that citation counts do not represent placement, page importance or a conventional ranking, which is an important distinction for marketers tempted to turn every new metric into another league table. [4]
Google launched a dedicated Generative AI performance report in Search Console in June 2026. The report provides impression data for supported features such as AI Overviews and AI Mode, with breakdowns by page, country, date and device. At the time of writing, Google describes it as a staged rollout to a subset of website owners, so not every property will have access immediately. The data also focuses on visibility rather than providing a complete account of citations, prompts, clicks and subsequent customer behaviour, but it represents an important step towards understanding how websites participate in Google’s generative search experiences. [6]
Google Analytics also introduced a dedicated AI Assistant channel within its default channel grouping in May 2026. Recognised visits from assistants such as ChatGPT, Gemini and Claude can now be assigned the ai-assistant medium and analysed alongside the events, key events and revenue recorded during those sessions. This does not solve the entire attribution problem because an AI platform cannot pass referral information when no link is clicked, and someone may return through another channel before converting, but it makes the directly measurable part of AI traffic much easier to isolate. [7]
These tools allow marketers to build a more meaningful measurement framework. Search Console can indicate whether eligible pages are being surfaced inside Google’s generative experiences. Bing Webmaster Tools can reveal citation activity across Microsoft’s supported AI surfaces. GA4 can show which identifiable AI assistants send users to the website and whether those sessions contribute to enquiries, conversions or revenue. CRM data, sales notes and customer interviews can then capture the influenced journeys that analytics cannot observe reliably.
That final layer is especially important. At Fly High Media, some of our highest-quality AI-assisted leads did not arrive as a clean, attributable referral from an AI platform. They came through branded search, direct traffic or another recognisable source after the prospect had already used AI to research agencies, understand the problem and form an initial shortlist. The most revealing information often emerged during the sales conversation when the prospect explained how they had found us and what they already knew about our work.
This is why I would not judge AI search purely through traffic volume. A smaller number of highly informed prospects may be more valuable than a much larger number of casual visitors. Businesses should examine lead quality, conversion rate, sales velocity, average contract value, influenced pipeline and the questions prospects ask when they arrive. If people have already used AI to compare providers and understand their options, the role of the website may shift from introducing the category to validating the decision.
Measurement should also include the quality of the brand’s representation. Is the company described accurately? Are outdated services still being mentioned? Does the system associate the organisation with the subjects it wants to own? Are competitors consistently cited where the company is absent? Which pages appear repeatedly, and which important areas of expertise are missing? These questions turn AI monitoring from a vanity exercise into a strategic review of how the market understands the brand.
Where SEO goes next
The most common prediction about AI search is that it will reduce the value of websites because people can receive answers without visiting them. There is some logic behind that concern. When an interface resolves a simple question directly, fewer users may need to click through to the source. Publishers built around large volumes of easily summarised information face a genuine challenge if the answer can be reproduced without the reader needing the original context.
However, that does not make authoritative websites less important. It changes the role they play. AI systems still need current information, first-hand evidence, accurate product details, specialist explanations and original sources from which to construct their responses. Google’s guidance explicitly connects generative search visibility to pages in its Search index and continues to emphasise crawlability, technical structure and valuable content. ChatGPT Search provides links to web sources, while Microsoft now gives publishers direct visibility into the pages cited across its AI experiences. [1][3][4]
The vulnerable website is not necessarily the one that loses a few informational clicks. It is the one that contributes nothing an answer system needs to retrieve. When a site consists mainly of generic summaries, duplicated descriptions and claims that cannot be independently supported, it becomes difficult to explain why it should be selected over thousands of alternatives. When it contains proprietary data, experienced analysis, definitive product information, useful tools and a recognised body of expertise, it remains valuable even as the interface changes.
This is why I believe SEO, digital PR, content strategy and brand building are converging. Technical SEO creates access. Content expresses knowledge. PR and reputation provide independent validation. Brand creates familiarity and trust before the person arrives. Analytics connects those activities to commercial outcomes. Treating each discipline as an isolated service misses the way modern search systems and customers combine information.
Search Everywhere Optimisation is my way of describing that broader responsibility. It is not an argument that SEO is dead or that every business needs a fashionable new acronym. It reflects the reality that discovery and evaluation now happen across an ecosystem. The same organisation may be encountered through Google, ChatGPT, Copilot, Reddit, LinkedIn, YouTube, customer reviews and traditional media during one buying journey. Success depends on whether those different encounters create a consistent and credible understanding of the brand.
The title Found, Trusted, Chosen describes the commercial journey most businesses need customers to complete. They must first become visible, then credible, and finally preferable to the alternatives. AI adds another decision-maker to parts of that journey. Information must be found by retrieval systems, trusted as a useful source and chosen for inclusion in the answer. The principles are similar even though the interface is changing.
That is why I do not believe the future belongs to the companies that produce the largest quantity of content. It belongs to those that create the strongest body of evidence. They will document their experience, publish original knowledge, communicate it clearly and earn recognition beyond their own websites. They will continue practising good SEO, but they will judge success by more than rankings. They will ask whether their expertise is shaping the answers customers receive and whether that influence leads to real commercial opportunity.
SEO has not become less important. It has become part of a bigger challenge. The first era of SEO was primarily about helping people find information. The next era will also require us to help machines understand why that information deserves confidence. Businesses that achieve both will not merely occupy a position in a search result. They will become part of how their market understands the subject.
The future of SEO is not about ranking. It is about becoming the evidence.
Join my personal newsletter
For more practical insights into SEO, AI search and Search Everywhere Optimisation, subscribe to my personal newsletter.
I share what I am observing, testing and learning as search continues to evolve, with a focus on the changes that are producing meaningful commercial outcomes rather than short-lived tactics.
Methodology
This article combines four types of evidence. First, I reviewed current documentation published by Google, Microsoft and OpenAI concerning generative search, web retrieval, citations and measurement. Secondly, I used foundational academic research into Retrieval-Augmented Generation to explain the distinction between knowledge stored within a model and information retrieved from external sources. Thirdly, I drew on observations across Google AI Overviews, AI Mode, ChatGPT, Gemini, Perplexity and Microsoft Copilot. Finally, I incorporated more than a decade of practical experience in SEO and digital marketing through my work as the founder of Fly High Media, author of Found, Trusted, Chosen, and adviser to businesses across ecommerce, healthcare, finance, professional services and B2B markets.
The Confidence Model is my interpretation rather than an official methodology published by any AI provider. No individual quality discussed in the article guarantees a citation, ranking or inclusion in a generated answer. The model is intended to help businesses prioritise enduring characteristics of useful information without pretending that citation selection can be reduced to a simple formula.
The Fly High Media commercial figures cover a rolling 12-month period and include directly identifiable AI referrals, prospects who explicitly stated that an AI assistant contributed to their research and opportunities where AI was recorded as an influential discovery touchpoint. They should not be interpreted as perfect last-click attribution.
Glossary
Retrieval-Augmented Generation (RAG)*
A method that combines a language model with information retrieved from an external source before or during the generation of an answer. In plain English, the system can look for relevant evidence rather than relying only on what it learned during training.
Query fan-out*
A process in which an AI system creates several related searches to investigate different parts of a complex request. Google gives the example of turning one lawn-care question into additional searches about herbicides, chemical-free removal and prevention.
Large Language Model (LLM)*
A type of machine learning model trained to understand and generate language. ChatGPT, Gemini and Claude are examples of products powered by large language models, although the models, retrieval systems and interfaces behind them differ.
Entity*
A recognisable person, company, product, place or concept that a search or AI system can distinguish and connect with related information. For example, an organisation may be connected with its founders, products, locations, industry and published work.
Grounding*
The process of connecting an AI-generated response to external information or evidence. Grounding can improve freshness and make it easier for users to inspect the sources supporting an answer.
Corroboration*
The process of supporting or validating a claim through additional evidence or independent sources. Corroboration does not require every source to express the same opinion, but it helps readers and systems assess whether a claim is defensible.
Frequently asked questions
Does AI search replace traditional SEO?
No. Google states that its generative search experiences are rooted in its existing Search ranking and quality systems, and pages still need to be crawlable, indexed and eligible for display. Technical SEO, helpful content, links, user experience and authority therefore remain important. The difference is that AI search introduces additional ways for information to be selected, summarised and presented.
Does ranking first guarantee that a website will be cited?
No. A strong ranking can help a page become discoverable, but it does not guarantee that the page will be selected for a particular generated response. The AI system may retrieve several sources, examine different passages or use query fan-out to investigate related aspects of the request.
Do I need an llms.txt file to appear in Google AI Overviews?
Google says it does not use llms.txt files and that creating one will neither improve nor harm visibility in Google Search. Other platforms may develop their own standards, so this answer is specific to Google’s published guidance.
Should content be divided into smaller chunks for AI?
Not automatically. Google says there is no requirement to break content into tiny pieces for its AI systems and that publishers should create pages for their audience rather than follow an arbitrary length or chunking rule. Clear sections and descriptive headings can help readers, but they should support the argument rather than fragment it unnecessarily.
Is structured data required for AI search?
Google says there is no special schema markup required for its generative search features. Structured data remains useful when it accurately describes visible content and helps pages qualify for established rich-result experiences, but it should not be treated as a guaranteed AI citation tactic.
How can a business measure AI visibility?
Businesses can combine platform reporting, analytics and CRM information. Bing Webmaster Tools provides AI citation insights, Google Search Console is rolling out a generative AI impressions report, and GA4 now includes an AI Assistant channel for recognised referral traffic. Sales teams should also record when prospects mention using AI during their research because many influenced journeys will not appear as direct AI referrals.
What type of content is most likely to remain valuable?
Content grounded in genuine experience, original evidence, transparent methodology and specialist knowledge is difficult to replace with a generic summary. Useful examples include proprietary research, case studies, benchmarks, expert explanations, calculators, tools and first-hand reviews.
Sources and further reading
1. Google Search Central: Optimising your website for generative AI features on Google Search
2. Lewis et al. (2020): Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
3. OpenAI: Introducing ChatGPT Search
4. Microsoft Bing: Introducing AI Performance in Bing Webmaster Tools
5. Google Search Central: Creating helpful, reliable, people-first content
6. Google Search Console: Introducing Search Generative AI performance reports
7. Google Analytics: What’s new in Google Analytics – AI Assistant traffic measurement
8. Google Search Central: AI features and your website
