7 AEO Experts Shaping AI Search in 2026, and What B2B Teams Should Learn From Them

I first published this article on 25 June 2026, when my main conclusion was that AEO was moving from theory towards measurement.

I still think measurement is central. But the way I think about the problem has become more precise.

The original article leaned heavily on ideas such as visibility, citations and share of answer. Those are useful signals, but they do not tell you whether an AI system is describing a company accurately, whether an observed citation helped produce a particular claim, or whether a change to a page caused a later answer to move.

The old Kojable Blog developed that thinking through subsequent pieces including:

• Build An AI Answer Alignment Baseline
• The AI Representation Gap: What It Is, Why It Happens, and How to Fix It
• How To Verify Whether An AEO Or AI Answer Alignment Change Worked
• Comparable Retesting in AI Search

The progression matters.

The first version of this article was mainly asking, “How do we become visible in AI answers?”

The later work asked harder questions:

What exactly is the AI saying? Is it accurate? What evidence is associated with the answer? Which gap actually deserves action? And when we make a change, how do we retest without pretending that correlation proves causation?

Kojable's current position reflects that evolution. We describe the broader problem as AI answer alignment: reducing the gap between company reality, public evidence and the story AI systems tell buyers. Visibility and citations sit inside that problem rather than defining the whole thing.

So this update is not simply a newer list of names. It is a reassessment of what the seven perspectives actually teach us.

Answer Engine Optimisation, or AEO, is the practice of improving how information is discovered, understood and represented in answer-oriented search experiences. Generative Engine Optimisation, or GEO, is commonly used for similar work focused on generative AI systems.

There is no universally settled boundary between the two terms.

Google describes AEO and GEO as third-party terms used for work focused on visibility in AI search experiences. From Google's own Search perspective, optimisation for its generative AI features remains part of SEO, and its current guidance emphasises established SEO fundamentals rather than a separate collection of AI-search hacks. (Google Search Central)

That is a useful reminder because the AEO conversation can easily collapse several different problems into one acronym.

A team might be trying to improve:

• whether its information can be found;
• whether a page can be understood and quoted cleanly;
• whether the company is represented accurately;
• whether credible public evidence supports its claims;
• whether it appears in comparisons;
• whether it is cited;
• whether it is recommended;
• whether different AI systems behave differently;
• whether an agent can evaluate or select the company;
• or whether any intervention produced observable movement.

Those are related questions. They are not identical.

That is why comparing practitioners is more useful than searching for one universal AEO playbook.

Jason Barnard matters because the answer-engine idea did not begin with ChatGPT.

His published record documents a BrightonSEO presentation in 2018 titled A Universal Strategy for Answer Engine Optimization, built around the shift from merely appearing in search results towards becoming part of the answer itself. His archive also traces AEO work to the period before that presentation. (Jason Barnard)

That history is useful because it stops us treating every AI-search problem as unprecedented.

Featured snippets, knowledge panels, voice answers and other direct-answer interfaces had already changed what “visibility” could mean. Generative systems made the problem much larger because they can combine information into new explanations, comparisons and recommendations.

What Jason's lens gets right

The object being optimised is no longer necessarily a ranked page.

Sometimes the relevant outcome is whether a system can correctly identify an entity, understand what it does and use its information when producing an answer.

That makes clear entity information, factual consistency and useful answer-level content important areas to inspect.

Where I would set the boundary

I would not treat phrases such as “algorithmic confidence” or “machine trust” as directly measurable internal states unless the system exposes evidence for them.

What we can observe is simpler:

• what answer appeared;
• which claims recurred;
• what sources were shown;
• how the entity was described;
• and whether those observations changed under comparable tests.

That distinction between an explanatory framework and an observable result is important throughout AEO.

Rand Fishkin's contribution is less about a specific GEO tactic and more about the economics of modern discovery.

His work with SparkToro has documented the growth of zero-click search for years. Using Similarweb clickstream data, SparkToro estimated that 68.01% of US Google searches in the first four months of 2026 ended without a click. That number describes a specific Google-search population and methodology. It should not be treated as a generic zero-click rate for ChatGPT, Gemini or every AI interface. (SparkToro)

The strategic point is more durable than any single percentage.

A user can encounter information, evaluate a company or change their view of a category without producing a visit to the source website.

That means traffic is no longer a sufficient proxy for influence.

What Rand's lens gets right

Marketing teams need to distinguish:

• visibility;
• source exposure;
• website traffic;
• brand consideration;
• and conversion.

They can move independently.

A company may receive less search traffic while still being represented prominently in the interface. It can also rank well and receive traffic while being absent or poorly described in an AI-generated comparison.

Where the lens stops

Zero-click behaviour tells us that measurement needs to expand.

It does not tell us what caused an AI system to mention one company rather than another, or how to change that result.

That is where the other practitioner lenses become useful.

Amanda Natividad makes the zero-click problem more operational.

Her current GEO/AEO framework breaks the problem into four areas:

1. Retrievability: can the system access the information?
2. Extractability: can the information be understood, summarised and cited?
3. Credibility: is there a defensible reason to rely on the source?
4. Public evidence: does supporting evidence exist outside the company's own website? (Amanda Natividad)

I like this framework because it moves the conversation away from vague instructions to “make content AI friendly”.

It gives a team four different diagnostic questions.

Retrievability

A page that is inaccessible to the relevant retrieval system cannot contribute through that retrieval route.

For a website team, this brings familiar issues back into scope: crawlability, indexation, internal linking, rendering and stable URLs.

Extractability

The important question is whether a source communicates its claims clearly enough to be used.

Definitions, explicit factual statements, descriptive headings, tables and well-structured comparisons can make information easier for both people and machine-mediated systems to interpret.

That does not mean a table or FAQ schema automatically earns a citation. Structure is something to test, not a guaranteed ranking factor.

Credibility

This is where I would make the language more cautious than much of the AEO industry does.

We can evaluate evidence quality, source provenance, named expertise, primary data and corroboration. We should be careful about claiming that a model internally “trusts” one page because it has those characteristics.

Public evidence

This is perhaps the most important expansion beyond owned-site SEO.

A company's own website is only part of the information environment. Reviews, interviews, independent articles, partner pages, public discussions, research and other third-party material can supply additional evidence about the entity.

The practical question is not “How do I manufacture mentions everywhere?”

It is:

What important claim about this company needs credible public evidence, and where would that evidence realistically belong?

AI-search excitement has created an incentive to declare every established practice obsolete.

Lily Ray is useful because her work pushes in the opposite direction.

Her current biography records more than 15 years in search, including building and leading a 35-person SEO team at Amsive. Her current work explicitly spans conventional SEO and AI-search platforms. (Algorythmic)

The larger point is not the team size.

It is methodological discipline.

Technical health still matters. Indexation still matters. Content quality still matters. Clear site architecture still matters. The authority and reliability of information still matter.

Google's own guidance reinforces that point for AI Overviews and AI Mode: its established SEO best practices remain foundational to appearing in generative Search experiences. (Google Search Central)

AEO adds to the audit rather than replacing it

For a B2B company, an AI-search audit may need to examine:

• technical search health;
• indexation;
• content quality;
• entity consistency;
• important buyer questions;
• representation across AI systems;
• competitor framing;
• cited sources;
• missing evidence;
• and changes across repeated tests.

That is more than traditional SEO.

But abandoning the parts of SEO that make information accessible and understandable would solve the wrong problem.

The better model is extension.

Rohit Singh's value in this comparison is the category question itself.

SEO, AEO, GEO, AIO, LLMO and now agent-oriented labels are routinely presented as though the industry has already agreed on neat boundaries between them.

It has not.

Rohit has argued that GEO and SEO may diverge further as agents retrieve and consume information through different environments, while also acknowledging that they remain closely related today. (Rohit Singh)

Google, meanwhile, currently treats AEO and GEO as third-party terminology and considers optimisation for its own generative Search experiences part of SEO. (Google Search Central)

Both observations can be true because they answer different questions.

Google is describing its own Search ecosystem.

Practitioners are trying to create useful language for a wider environment that includes standalone assistants, retrieval systems, recommendation interfaces and agents.

The practical lesson is methodological

Do not let the acronym become the strategy.

Before deciding whether a piece of work is SEO, AEO or GEO, define:

• which system you are observing;
• which user question matters;
• what outcome you are measuring;
• what evidence the system exposes;
• and what action the team can realistically take.

A shared vocabulary is useful.

A false sense of standardisation is not.

Ethan Finkel, founding product manager at Gauge, adds one of the most useful technical distinctions in this group.

In an Ascend interview, he described two broad answer paths: a model can rely on information already represented in its internal knowledge, or the product can retrieve information from the web at answer time and use that material to construct the response. (Ascend)

This distinction matters because marketers often discuss “LLM visibility” as though every answer is produced in the same way.

It is not safe to make that assumption.

Retrieval changes what can be observed

In a retrieval-enabled answer, you may be able to inspect:

• search activity;
• exposed candidate sources;
• final citations;
• or the URLs shown alongside the answer.

Those observations can create faster feedback loops than trying to infer anything about a model's training data.

But even here, we should distinguish the stages.

A source appearing in a retrieval candidate set is not the same thing as becoming a final citation. A citation is not proof that the source caused every claim in the answer.

Kojable's own research reinforces the cross-provider problem

In Kojable's Different Answers, Different Evidence research, we tested the same designed B2B buyer questions across Claude, Gemini, OpenAI and Perplexity. Under that fixed benchmark, the observed source environments could differ materially between provider stacks.

The important application is not a precise overlap percentage.

It is this:

Do not assume that being visible in one provider's evidence environment means you are visible in another.

The study is observational. It does not establish permanent provider preferences or prove why a particular URL was selected. That methodology and the detailed quantitative evidence belong on the canonical Kojable Research page, rather than being reproduced here.

For a marketing team, the practical decision is straightforward: measure the provider and surface you actually care about.

Joseph Ho pushes the discussion one step further.

His recent AEO/GEO commentary distinguishes conventional search, answer engines, generative systems and Assistive Agent Optimisation, framing the last category around agents that can make decisions or take actions for a user. (Joseph Ho)

The strongest version of that idea is not that “agents always pick one winner”. Current systems and tasks vary too much for that claim.

The more defensible point is that selection can become a different outcome from visibility.

Those are different outcomes and should be measured separately.

The AAO terminology also does not originate solely with Joseph. Jason Barnard published a February 2026 Search Engine Land framework explicitly describing Assistive Agent Optimization as optimisation for environments where an AI agent moves beyond recommending and towards acting. (Search Engine Land)

What changes for marketers

The agent-first question forces companies to make information more operationally useful.

Consider:

• Is the product clearly described?
• Is its intended audience explicit?
• Are important constraints visible?
• Is pricing or commercial information current where it should be public?
• Can an external system distinguish the company from a competitor?
• Is there defensible evidence for the claim being evaluated?
• Can the relevant information be accessed in a form suitable for the task?

These are useful questions even before AAO becomes a stable industry term.

This is where my own view has changed most since the original article.

In June 2026, I described the next phase of AEO principally as measurement.

I still agree with that, but I would now be more specific about what needs measuring.

A visibility score can tell you whether a company appears.

A citation metric can tell you whether a source was shown.

Neither tells you by itself whether the answer is accurate, whether the company is placed in the right category, whether the correct differentiation appears, or whether the public evidence supports the claims a buyer actually cares about.

That is why we now use AI answer alignment as the broader frame at Kojable.

Kojable is an AI answer alignment platform for B2B companies. The objective is to reduce the gap between:

1. what is currently true about the company;
2. what public evidence exists;
3. and what AI systems tell buyers when they research, compare and validate that company.

The original AEO article concentrated heavily on visibility.

The old Blog posts that followed progressively introduced the baseline, the representation gap and comparable retesting.

The current model connects them.

Visibility is a signal. AI representation is the observable output. Alignment is the problem. Monitor → Diagnose → Improve → Verify is the operating process.

That is a more useful model for a marketing team than chasing a new acronym every quarter.

AEO in 2026 is not one technique.

Jason Barnard reminds us that the transition towards answers started before generative AI.

Rand Fishkin shows why a click is no longer a sufficient unit of discovery.

Amanda Natividad gives teams a practical way to think about access, extraction and public evidence.

Lily Ray keeps technical and search discipline inside the conversation.

Rohit Singh reminds us that a fast-moving industry can mistake vocabulary for consensus.

Ethan Finkel makes the retrieval architecture visible.

Joseph Ho pushes the question from being present in an answer towards being considered when AI systems recommend or act.

The mistake is turning any one of these into the complete doctrine.

For B2B teams, the better approach is to define the buyer question, observe the answer, diagnose the evidence gap, take a justified action and retest.

That is where AEO stops being a trend and becomes an operating discipline.

The useful test of an AEO framework is not how persuasive it sounds. It is whether the team can move from an observed answer to a defensible diagnosis, take an appropriate action and verify what changed.

Explore Kojable's case studies to see how answer-alignment work moves from observation and evidence through to improvement and comparable retesting.