SEO After AI: What the Snake Oil Salesmen Won't Tell You

Introduction and Framing

Andrew Palmer introduced Pete Everitt as a long-standing colleague from the UK SEO community, connected through the events organised by Lee Matthew Jackson. Palmer noted that Everitt productises SEO through his company, SEO Hive, and that the session title was chosen deliberately and provocatively.

Everitt opened by acknowledging the title's clickbait appearance before immediately asserting that the session was intended as an honest, grounded conversation. He positioned the talk as a counterweight to the volume of noise circulating on social media about AI and SEO, noting that contradictory claims often appear in the same feed, sometimes from the same person.

He committed to covering three specific things about AI SEO (also referred to as AEO or GEO depending on the layer of the stack being discussed) that are under-discussed, counterintuitive, or simply not what the loudest voices online want to say.

The Two Dominant Narratives (and Why Both Are Wrong)

Everitt identified two opposing camps that dominate online discourse about AI and SEO: Camp One: "AI is killing SEO." This position typically arrives with a sales pitch. The person making the claim is usually selling a course, a retainer, or a subscription product that promises to show you how to "rank in AI." The implication is that SEO as previously understood is dead and their new system is the replacement.

Camp Two: "Nothing has changed." This is the reassurance narrative. The person making this claim typically wants clients to stay calm, keep their existing SEO retainers running, and trust that things will work out. It downplays disruption.

Everitt's position is that the truth is messier than either camp admits. Neither is wholly right. Both are inadequate responses to a genuinely complex and evolving situation. The purpose of his session was to offer something more honest and more practically useful than either narrative.

Why Bad Advice Proliferates: The Categorisation Error

Before addressing the three core points, Everitt identified what he sees as the root cause of poor AI SEO advice: a categorisation error.

When AI search became a significant phenomenon, the responsibility for understanding and optimising for it was handed to SEO practitioners and web developers. This made surface-level sense. AI search involves being found online. It involves content, rankings, and brand visibility. So the people who had been handling those things for websites got handed the AI question.

The problem is that AI is not search. It operates on fundamentally different principles. It categorises information differently, it models relationships differently, and it generates outputs through a mechanism that has little in common with a search index. Most of the bad tactical advice in the space, Everitt argues, flows directly from this mismatch. People trained in SEO logic are applying SEO solutions to a non-SEO problem.

What AI Actually Is: The Character Predictor Model

Everitt offered a plain-language explanation of what AI language models are actually doing, citing Rand Fishkin (previously of Moz) as someone who has articulated this clearly.

When a user queries an AI, the system is not going away and searching a database. It is not retrieving the "best" answer. It is predicting what the next most logical character, word, or sentence would be, based on everything that came before it in the prompt and based on everything in its training data.

Everitt applied this specifically to the visual domain as well, noting that the same logic applies to pixels (images) or frames (video), but for the purposes of text-based AI the essential mechanism is word and sentence prediction.

The implications of this are significant: AI is not finding the best answer. It is generating what the best answer looks like.

This is a subtle but meaningful shift. The optimisation problem is not about ranking. It is about whether you exist coherently within the model's training data.

If the model does not have a clear picture of what you or your business does, you will not appear. Not because your ranking is too low, but because the model has no reliable entity for you at all.

This is a different problem from traditional SEO, and it requires a different kind of solution.

The Fragmented AI Landscape vs. Google's Dominance

A further complicating factor Everitt identified is that the AI landscape does not have a single dominant player the way search does. In traditional SEO, Google holds over 90% of search share. Bing and other engines are minor enough that they largely conform to whatever Google does. Practitioners can essentially optimise for Google and expect secondary benefits across the board.

AI is different. OpenAI, Google (Gemini), Microsoft (Copilot), Anthropic (Claude), Perplexity, and others all operate with their own models, their own training data, and their own ranking logic. None of them has to conform to a single standard. This creates genuine uncertainty and grayness in how to approach optimisation.

Everitt acknowledged this fragmentation as the reason why the framework he went on to describe is intentionally broad: it needs to be broad enough to accommodate the quirks across different platforms, while still being specific enough to give practitioners a clear direction.

The EEAT to AEO to GEO Framework

Everitt presented a three-layer framework that he described as the most reliable guide available for navigating AI search. These layers build on each other sequentially.

Layer One: EEAT (Experience, Expertise, Authoritativeness, Trust)

EEAT is Google's long-standing framework for evaluating content quality. Everitt argued that rather than becoming less relevant in the age of AI, EEAT has actually become more important. This is because the internet's established conventions for signalling topic authority have informed how AI models were trained. The things that made content credible and authoritative to Google's evaluation systems are, in large part, the same signals that AI models use to determine whether a source is worth drawing on.

EEAT is the foundation layer. It builds the credibility that the subsequent layers depend on.

Layer Two: AEO (Answer Engine Optimisation)

AEO emerged chronologically after EEAT as a discipline and is concerned specifically with the AI Overviews and featured snippet-style outputs that now appear at the top of Google search results. This layer is about structured content and semantic markup.

At the practical level, AEO involves:

• Using standard semantic HTML and structured data so that AI can parse what is on the page.
• Ensuring that answers to specific questions appear within approximately 100 words of a heading or title that is clearly related to the likely prompt.
• Applying copywriting techniques that surface answers quickly and in a format machines can extract cleanly.

Everitt described AEO as a "halfway house." There is still an SEO element to it (the structured content and how it surfaces in search), but there are additional technical and copywriting requirements beyond traditional optimisation.

Layer Three: GEO (Generative Engine Optimisation)

GEO is the most advanced layer and is about being present in the actual language model outputs. This is the "generating" or "predicting" layer: being what the model reaches for when constructing an answer in a given topic area.

Everitt summarised the three layers as a progression:

• EEAT builds your credibility.
• AEO proves to machines that you have the structure and content to answer specific questions.
• GEO gets you into the LLMs themselves, so that you feature in what they generate.

- GEO gets you into the LLMs themselves, so that you feature in what they generate.

The Three Things Snake Oil Salespeople Won't Tell You

Thing One: There Is No Proven Playbook The first and most direct claim Everitt made is that there is no proven, step-by-step method for guaranteed AI SEO success. Anyone claiming otherwise and asking for your money in exchange for that claim should not get your money.

His reasoning:

• AI is still developing and evolving rapidly. What worked six months ago may not work now. What worked three months ago may not work now. What worked three weeks ago may not work now.
• AI citation data is harder to track reliably than organic search rankings. Traditional SEO can track keyword positions with reasonable accuracy. AI citations require tracking exact prompts, which can be paragraphs long. The patterns of appearance in AI outputs are far more inconsistent than anything seen in traditional search.

- Traditional SEO has 20-plus years of data behind it. Practitioners understand the black hat and white hat divides, they understand what builds lasting authority, and they can predict outcomes with reasonable confidence. AI has no equivalent foundation yet. - AI is "a surface tactic," he said. It "constantly keeps on changing" and there are no solid foundations established yet from which anyone can say definitively "this is how it works."

He was emphatic on this point: do not waste money on courses or SaaS products claiming to guarantee AI search visibility.

Thing Two: Optimising for AI Search Now Is Not Enough The second point is that current text-based AI is only the beginning. Optimising narrowly for what AI looks like today means aiming at the wrong target.

Everitt noted that the current dominant paradigm is text-in, text-out: type a prompt, get a text response. This is the MVP, the minimum viable product of AI. The technology is already expanding significantly:

• Midjourney and native phone/tablet image editing tools have brought AI image generation to everyday users.
• Voice AI agents are becoming commercially viable. Everitt referenced a consultancy client in the United States building voice AI agents for restaurants and automotive businesses. These agents can take bookings, handle cancellations, process modifications, manage special requests, and send location-aware reminders. This is AI taking actions, not just answering questions.
• AI agents more broadly are becoming more capable of compounding decisions over time.

The implication for optimisation strategy is that if you build only for text-based AI search as it exists today, you will miss the much larger picture of what AI will look like in months and years ahead. The foundations being built now need to be durable enough to remain relevant as these more complex, multimodal, agentic forms of AI become mainstream.

Thing Three: The Big Players Still Control the Rules The third point is that no tactic or framework will protect a practitioner from platform-level rule changes made unilaterally by the major companies. Google, OpenAI, Microsoft, Anthropic, and Perplexity control the game. They change the rules when it suits them. They always have.

Everitt offered a concrete, recent example to illustrate this.

In September (of the prior year at time of the session), Google removed the num=100 URL parameter from search. For those unfamiliar: appending num=100 to a Google search URL would programmatically return the top 100 results for any query, rather than the default 10.

AI systems had been using this parameter extensively to scrape a broad set of results when training and when refreshing their knowledge of a topic.

Google removed the parameter with minimal advance notice. Effectively overnight, AI systems that relied on it were limited to scraping only the top 10 results for any given query. The consequence was that AI models lost all context for content sitting below position 10. If you were ranking 11th or lower for a term, AI could no longer discover you through this mechanism. Unless the model already had you in its training data, you effectively ceased to exist for that topic.

Everitt was clear that Gemini did not face this limitation (since Google controls both the search index and Gemini, it had access to broader data). But for every other AI system, this one platform decision immediately changed the landscape.

The lesson: the rules can and will change, without warning or lengthy transition periods. No "master the algorithm" tactic survives a deliberate platform-level change. The only durable strategy is to build real authority and real credibility over time, because that is what tends to persist across rule changes.

The Good News: Inherent Authority

Having laid out three sobering truths, Everitt offered genuine encouragement. If a practitioner has been doing the basics properly and doing them consistently over a long period of time, they likely already have a degree of inherent authority that will serve them well in AI contexts.

The point is that AI models, when they discover a source that has the right foundational signals in place, can rapidly place it well ahead of competitors in a given niche. Long-term investment in credibility and structure pays off in ways that are not entirely predictable but are genuinely advantageous.

What You Actually Need to Do: The Practical Foundations

Everitt outlined three core foundational actions.

Build Your Schema Graph Properly A schema graph is conceptually similar to a sitemap, but constructed by applying schema markup to the URLs across a website. Schema markup declares the entities present on each page and, critically, how those entities relate to one another.

Everitt explained that AI models work in entities and relationships, not keywords. He used a simple illustrative example: "Pete Everitt, Sheffield, Digital Agency." Each of those is an entity. When a model queries "digital agency," it can also surface a relationship to Pete Everitt and Sheffield because the schema has declared those relationships explicitly. This is the logic of a relational data model, not a keyword index.

He also noted that linking to external sources, including Wikipedia, which has traditionally been discouraged in SEO contexts, is actively beneficial in schema and AI contexts. Those external references help models verify and contextualise entities.

Key points about schema graph construction:

• It cannot be applied only to key pages (homepage, main service pages, contact). That is not enough context.
• It must be applied consistently across the entire site.
• It must be maintained over time. It is not a set-and-forget exercise.
• The volume of data these models work with is enormous. A sparse or inconsistent schema graph will not register.

Author Everything Everitt argued that authorship is one of the highest-value signals available. Named individuals with real credentials, consistently attributed across content, and treated as entities in their own right within the schema graph, carry enormous authority with AI models.

This aligns with what EEAT already demanded around the Experience and Expertise dimensions. If practitioners have already been attributing content to named authors with credentials, they have a head start.

The emphasis is on consistency and on making the author an entity: not just a byline, but a structured entity with schema, credentials, social profiles, and a feed of attributed content.

Understand That AI-Generated Content Underperforms in AI Search This point is counterintuitive and likely uncomfortable for practitioners who have leaned into AI content generation. Everitt stated plainly that AI-generated content underperforms in AI search outputs relative to human-authored content with genuine expertise and real examples.

He offered two explanations:

1. AI-generated content is often not properly attributed to a real author, which weakens its entity signals.
2. More fundamentally, AI models appear to detect their own generated content and discount it.

Human-authored content, grounded in genuine experience with real examples, consistently outperforms in LLM-generated outputs.

Practical Agency Advice

Everitt then pivoted to concrete, actionable guidance for digital agencies.

Apply the Foundations to Your Own Agency First Before selling AI SEO services to clients, apply all of the above to your own agency website.

The reasons are twofold:

• It gives you real, lived experience of how these foundations work, which makes you a more credible and informed advisor.
• It reveals the true scope of the work. Everitt was candid: "It's a grind." It is not a quick project. Understanding that from your own experience before you sell it to a client prevents you from under-scoping or under-pricing the work.

He noted that most active agency websites already have a reasonable number of assets: core service pages, a reasonably active blog, case studies, a contact page. Adding author pages and beginning to apply schema to that existing structure is a manageable starting point.

Sell It as a Service This is a genuine commercial opportunity. Clients will start asking for it if they have not already. Everitt recommended:

• Start with an audit to establish where a client currently stands.
• Present the opportunity: what could their visibility and authority look like if they invest in this properly?
• Get the skills, tools, and processes in place now, before client demand peaks, rather than scrambling to catch up later.
• Agencies that do not build this capability will leave money on the table.

Change How You Build Sites This is the most forward-looking of the three recommendations and, Everitt acknowledged, the most disruptive. Clients who commission a new website regard it as a business investment and expect it to be ready and fit for purpose when delivered. Increasingly, "fit for purpose" must include the foundational elements of AI readiness.

That means:

• Schema markup capability should be built in, not retrofitted.
• Author page templates should be a standard part of any build, not an optional extra.
• The architecture of the site should be designed from the start to support the kind of consistent, entity-rich schema graph that AI models need.

The goal is that clients receive a site that is already ready for these foundations, rather than needing to come back months later for an expensive retrofit.

Understanding that from your own experience before you sell it to a client prevents you from under-scoping or under-pricing the work.

Closing Summary

Everitt summarised his three points before wrapping up: SEO is still fickle. It will keep changing. Nobody knows exactly how it will change. Do not spend money on courses or products claiming otherwise.

Text-based AI is still just the beginning. The future will include voice, agents, multimodal interaction, and more. Build for where this is going, not just where it is now.

The big players control the rules and they will change them. You have to play their game.

The only durable position is one built on genuine authority and solid foundations, not tactical tricks.

His closing framing: "Stop chasing tactics, start building AI-proof foundations."

He promoted SEO Hive's Scout Reports product and noted that an AI audit component was being added. At the time of the session, these were not yet publicly available.

Q&A

The session ran slightly over time, leaving only a brief Q&A window.

Question from Stephanie Hudson: How do you qualify as an author, or how do you properly qualify an author for these purposes?

Everitt's answer was detailed. The ideal author setup consists of:

• A byline on every post or page attributed to that author, including a short bio.
• A link from that byline to a dedicated author page.
• The author page contains proper schema itemising the author's name, position, qualifications, and social media profiles. If the author appears on Wikipedia or similar reference sources, those should be referenced.
• The author page also has a feed of all posts and pages attributed to that author.
• This feed keeps the author page's timestamp current, signals to both AI models and Google that the author is still active, and creates a coherent entity with a documented, growing body of work.

He noted that once an agency has built this template properly once or twice, it can be rolled into their starter kit or theme, making it straightforward to include in every future build.

Question from Andrew (relayed from attendee): How do you learn how to do this?

Everitt recommended his own course, "Demystifying SEO," which includes an AI module. He noted with some self-deprecating humour that the course was already several weeks old, which in the current pace of AI development might render parts of it dated. He also pointed 1.

to SEO Hive's blog content and offered to have direct conversations with practitioners who wanted to explore the topic further.

About the speaker Pete Everitt Co-founder, SEO Hive

Pete Everitt is co-founder of SEO Hive, an agency and consultancy focused on search. He is known for cutting through hype around AI search with a technically grounded view of how these systems actually work.