The rise of AI-generated answers has created a new layer of uncertainty around content strategy.
Businesses that once asked how to rank in search now ask how to appear in AI answers, how to become a cited source, whether they need "GEO" or "AEO," whether their pages should be written differently, and whether old SEO practices still matter.
The Genius Talk conversations assigned to this question do not support a magic formula.
They support a more durable approach.
Melanie Gorman emphasizes a hub-and-spoke content structure built around important services and the real problems that lead customers to them. Matthew Edgar treats AI-search visibility as an experimental technical discipline and warns against turning observations into permanent rules. Brandon Leibowitz stresses relevance, competition, useful content and the broader web signals that have long mattered in search. Anna Covert distinguishes traditional SEO from newer AEO and GEO language while emphasizing clear points of difference. Jack Turk brings the editorial discipline: AI can accelerate work, but the person using it still needs the judgment to recognize useful, persuasive and strategically sound output.
Current platform documentation adds an important boundary around those expert observations.
Google's own guidance says that its established SEO practices remain relevant to AI Overviews and AI Mode. It does not require a special AI file, special schema or a separate style of writing for those features. OpenAI, meanwhile, documents a distinct search crawler, OAI-SearchBot, that site owners can allow or disallow independently from GPTBot, which is used for model-training-related crawling.
That leaves a practical content strategy built around four jobs:
- Publish material that is specific enough to be useful.
- Organize it so people and systems can understand the relationship between topics.
- Support important claims with evidence and clear attribution.
- Make the content technically accessible to the systems you want to discover it.
Those jobs matter whether the eventual interface is a familiar search result, an AI-generated answer, a cited source link or a person navigating the site directly.
Search results and AI answers are different interfaces
Traditional search and AI-generated answers can both begin with a person expressing a need, but the experience is different.
A familiar search result gives the user a set of links and snippets. The user does more of the synthesis.
An AI answer may summarize, compare or organize information before the user clicks. It can also break a complex question into related searches or retrieve information from multiple sources.
That changes how visibility is experienced.
A page may contribute useful information without receiving the same kind of prominent blue-link exposure marketers learned to track. A user may encounter a brand inside a comparison or cited answer rather than by opening a result immediately. The same query can produce different source selections across time, users or systems.
This makes AI-answer visibility harder to reduce to one position number.
Edgar's interview captures that uncertainty well. His team has experimented with technical variables such as JavaScript rendering, pre-rendering, HTML structure, internal linking, bot activity and prompt mentions. He is careful to frame the findings as evolving observations rather than fixed ranking factors.
That discipline should carry into any AI-search strategy.
An experiment can tell you what happened in a particular test. It does not automatically reveal a platform rule.
Start with useful specificity
Gorman's strongest contribution is a content strategy built around specificity, without requiring a theory about hidden model-ranking behavior.
Her hub-and-spoke structure starts with the services or core areas a business needs to be known for. Each important service gets a clear landing page. Supporting articles answer the real questions, situations and problems that cause someone to need that service. Internal links connect the supporting material back to the relevant core page.
This structure works because it solves a human information problem before it solves a search problem.
A visitor can understand what the business does.
They can move from a question into the relevant service.
They can see deeper evidence of expertise around that subject.
The site stops looking like a random collection of blog posts.
Specificity also helps the business avoid publishing generic material that could belong to almost any competitor.
A broad article titled "How to Improve Your Marketing" asks the reader to do a lot of interpretation.
A specific article answering "How should a regional accounting firm structure service pages for three distinct buyer groups?" carries more context. It defines an entity, a situation and a problem. It gives the writer a narrower field in which to demonstrate real experience.
This is where Gorman's emphasis on niche, audience and problem becomes useful. A smaller business often cannot win by becoming the most comprehensive general source on a huge topic. It can become unusually useful for a particular kind of person facing a particular kind of decision.
Build a topical architecture, not a pile of pages
Content volume is easy to count. Architecture is harder to see.
A site can publish frequently and still make it difficult for a reader to understand which pages matter, how they relate and where to go next.
The hub-and-spoke model offers one answer.
A core page establishes the central subject or service. Supporting pages go deeper into specific questions. Internal links create paths between them.
The same principle can be applied beyond service businesses.
A software company might organize content around major use cases.
A professional association might build hubs around recurring member problems.
A publisher might connect expert interviews, synthesis articles and practical guides around a defined theme.
An ecommerce business might connect category pages, comparison content, buying guides and product education.
The value of the hub-and-spoke model comes from making the relationship between pages visible.
Google's current guidance still recommends making content easy to find through internal links. That is a conventional SEO principle, and it remains relevant to Google's generative search features.
Internal links also help readers.
They reveal hierarchy, give context and offer a next step without forcing the visitor back to navigation or search.
Internal linking should express meaning
Weak internal linking is mechanical.
A site publishes a new article, then adds a handful of links because an SEO checklist requires them.
Stronger internal linking reflects the relationship between ideas.
If an article explains a problem that leads naturally to one service, the service page should be linked where that decision becomes relevant.
If a supporting article introduces a concept explained more fully elsewhere, the deeper source should be linked at the point of need.
If several articles belong to one larger subject, a hub page can help readers see the full set without relying on site search.
This makes the content library easier to navigate and easier to maintain.
It also creates a useful editorial test. If the team cannot explain where a new article belongs, why it exists and which pages it should support, the article may be too disconnected from the site's actual strategy.
A content calendar should not become a substitute for information architecture.
Service and entity specificity reduce ambiguity
AI-search discussions often use the word "entity." The term can sound more technical than the practical problem requires.
A business page should make clear who the organization is, what it does, where relevant, who it serves, and how its products, services, people and evidence relate.
That clarity helps humans first.
A vague service page filled with broad benefit language can leave a prospect unsure whether the business actually handles their situation. A specific page can identify the service, intended customer, process boundaries, geography when relevant, proof, common questions and next step.
Gorman's "why you, why this service, why now" test is useful because it pushes the page beyond category-level language.
"Experienced marketing support for growing businesses" says little.
A page that explains which businesses the firm works with, what problem triggers the engagement, how the service differs, what evidence supports the approach and when the service is a poor fit gives the reader something to evaluate.
That kind of specificity is also safer than inventing artificial "AI keywords."
The content becomes distinct because the business itself is distinct.
AEO and GEO are useful labels only if they improve the work
Anna Covert distinguishes SEO, answer-engine optimization and generative-engine optimization as different lenses on discoverability.
The terms can help teams notice that discovery is changing.
They become less useful when they are treated as separate systems with guaranteed tricks.
Google's current documentation is unusually direct on this point. Its 2026 guidance says that SEO remains relevant to its generative AI search features because those experiences rely on core Search ranking and quality systems. Google also says site owners do not need special AI text files, special schema, artificial "chunking" or a separate style of writing to qualify for its AI features.
That does not make every third-party AEO or GEO practice useless. It means claims about such practices need evidence.
A reasonable working definition is:
- SEO focuses on making useful web content discoverable and competitive in search.
- AEO can be used as a planning label for making answers clear enough to satisfy direct questions.
- GEO can be used as a planning label for visibility inside generative search or answer experiences.
The labels should not outrun the underlying evidence.
If a GEO tactic conflicts with official platform guidance or exists only because someone observed one short-lived pattern, treat it as a hypothesis.
Evidence is part of content quality
Gorman emphasizes credentials, experience, citations and proof.
That matters because expert content has two separate jobs.
It has to say something useful.
It has to give the reader a reason to trust the useful thing.
Evidence can take several forms depending on the claim:
- a first-party platform document for a current platform rule;
- a primary study for a research claim;
- a clearly attributed guest observation for professional interpretation;
- a case study for a reported business result;
- a screenshot, test record or technical log for an experiment;
- a direct customer quote for a customer experience;
- a clearly labeled example when the point is illustrative rather than empirical.
The evidence should sit close to the claim it supports.
This is especially important in AI-search content because the field attracts confident statements about systems that change rapidly and disclose only part of how they work.
A page becomes more credible when it separates what is known from what is observed.
Separate platform facts from practitioner observations
Matthew Edgar provides a useful model for this distinction.
His team has run experiments around JavaScript, pre-rendering, HTML structure, internal links, crawling and AI-search behavior. Those tests can generate useful hypotheses. They do not become permanent rules simply because they produced a result.
The same applies to Brandon Leibowitz's comments about links, mentions and AI systems. His views come from long experience in SEO and current professional interpretation, but the assigned source material explicitly warns against turning those observations into fixed AI ranking rules.
A citation-quality article should therefore use clear language:
"Edgar reports that his team observed..."
"Leibowitz argues..."
"Google currently documents..."
"OpenAI currently says..."
"The platform does not publicly establish..."
This level of attribution is not timid. It tells the reader which kind of evidence they are looking at.
That distinction is especially valuable when the page may later be cited by another writer or summarized by an AI system.
Human judgment is the filter AI cannot supply for you
Jack Turk's contribution is editorial rather than technical.
He argues that AI can make a capable marketer faster, but the person still needs enough understanding to judge the output.
That problem becomes more serious at scale.
A model can produce a plausible article quickly. It can also produce generic language, flatten distinctions, mix sources, overstate evidence or repeat an error with confidence.
The publishing advantage therefore comes from using AI inside a stronger editorial process.
A subject-matter expert can identify what is missing.
A researcher can verify current facts.
An editor can remove false certainty.
A marketer can decide which question is commercially relevant.
A technical owner can confirm that the page is accessible and indexable.
AI may reduce the labor required for parts of that process. It does not remove the need for judgment at the points where meaning, truth and strategy are decided.
Generic mass production creates a quality problem before it creates an SEO problem
Businesses sometimes respond to generative AI by increasing output dramatically.
The logic is understandable. If drafts are cheaper, publish more.
The weakness is that production capacity and information value are not the same thing.
Google's current guidance warns that using generative AI to produce many pages without adding user value can violate its scaled-content-abuse policies. More broadly, a site full of lightly differentiated pages can make it harder for people to understand which page matters.
The better question is what the business can add that a generic model output cannot supply on its own.
That may include:
- direct operating experience;
- original examples;
- customer language;
- proprietary data the business is entitled to publish;
- expert disagreement;
- documented tests;
- current platform verification;
- a clear point of view grounded in evidence;
- useful synthesis across sources that are rarely considered together.
The advantage comes from information quality, not from pretending AI was never involved.
Technical access still matters
Excellent content can remain invisible when systems cannot reliably access it.
This is where Edgar's technical emphasis becomes important.
Google's current documentation says pages need to meet the usual technical requirements for Search and be indexed and eligible to appear with a snippet to be eligible as supporting links in its AI features. It also recommends allowing crawling, keeping important content available in textual form and making pages easy to find through internal links.
Google can render JavaScript, and its 2026 documentation explicitly notes that JavaScript itself is not inherently a barrier to Google Search. Yet JavaScript-heavy sites can still create additional complexity, and other crawlers may behave differently.
That distinction matters.
Do not turn "Google can render JavaScript" into "every AI crawler handles every client-rendered site exactly the same way."
When important content depends on a particular rendering path, test the actual systems that matter to the business.
Technical accessibility is an empirical question.
Search access and model-training access are not the same setting
OpenAI's crawler documentation makes one current distinction particularly clear.
OAI-SearchBot is used for ChatGPT search features. GPTBot is associated with crawling that may be used to train OpenAI's generative AI foundation models. OpenAI says site owners can configure these independently in robots.txt.
That means a publisher can allow OpenAI's search crawler while disallowing GPTBot.
The distinction matters because "block AI" is too vague to be operational.
A technical policy should identify which crawler, which product behavior and which business objective it is addressing.
The same discipline should be applied to any platform control.
Do not copy a robots.txt rule from a social post without understanding what that rule changes.
Review the first-party documentation, decide what access the business wants to permit, implement the rule, then verify the result.
Robots controls can affect visibility
Search and AI features depend on access.
If a crawler is blocked, the system may not be able to retrieve the content in the same way.
Google documents robots.txt and preview controls for its Search features. OpenAI documents OAI-SearchBot as the control for inclusion in ChatGPT search crawling.
These are current platform facts and should be treated as date-sensitive.
The content team needs a reliable handoff to infrastructure owners even when it does not manage the infrastructure itself.
Someone should know:
- which crawlers the site allows;
- which important pages are blocked;
- whether a CDN or firewall is blocking legitimate bot traffic;
- whether canonical, noindex or snippet controls are intentional;
- whether the page renders meaningful content without requiring a fragile client-side path;
- whether internal links to the page are crawlable;
- whether structured data, where used, matches the visible page.
A publishing checklist that ignores these questions can fail before content quality is evaluated.
Structured data still has a role, but there is no special AI schema shortcut
Structured data can help search systems understand specific page information and can make pages eligible for supported rich results.
That does not mean adding more schema causes an AI system to cite the page.
Google explicitly says there is no special schema.org markup required for its generative AI search features. It also says structured data should match the visible content.
The practical standard is therefore accuracy.
Use supported structured data when it helps describe content that exists on the page.
Keep it aligned with the visible information.
Do not create markup for claims, reviews, authorship or entities the page does not actually support.
A technical annotation is only as trustworthy as the content it describes.
Content should be answerable without becoming robotic
The growth of answer engines has encouraged a style of writing where every page begins with a blunt two-sentence answer followed by dozens of tiny headings.
Sometimes that format is useful.
It should not become a universal template.
Google's current guidance specifically rejects the idea that content must be artificially "chunked" for its generative systems. It also says there is no ideal page length.
The human question comes first.
A simple factual query may deserve a direct answer near the top.
A complex strategic question may need definitions, disagreement, examples and qualifications before the conclusion is responsible.
A comparison may work best in a table.
A technical process may need numbered steps.
A nuanced expert article may need long-form synthesis.
The page should fit the task.
This is where V06, the Genius Talk synthesis on writing for skimmers, becomes a useful companion. Structure helps readers move through depth. It should not flatten every subject into the same layout.
Distinctive expertise matters more as commodity text becomes easier to produce
Covert emphasizes defensible points of difference.
That becomes more important when generic explanatory text is abundant.
If ten businesses can publish competent introductory articles on the same topic in an afternoon, the advantage shifts toward material with a stronger reason to exist.
What can this source say that a generic source cannot?
What experience does it have?
What evidence can it show?
What tradeoff can it explain?
What does it know about this audience, geography, product, process or market that changes the answer?
This is the editorial version of entity specificity.
The page should give both the reader and any system processing the page a clear reason to associate the source with the subject.
That reason should come from substance.
Backlinks and mentions need careful language in AI-search discussions
Leibowitz places substantial weight on relevant, authoritative backlinks in traditional SEO and sees guest appearances as potentially useful when they produce a linked web page.
That is a familiar SEO strategy.
The AI-search extension requires more caution.
It may be reasonable to test whether broader web presence, authoritative references and relevant mentions correlate with AI visibility. Edgar and Leibowitz both discuss forms of this idea.
What cannot responsibly be said from these interviews is that a particular number of backlinks or mentions will cause a specific AI system to cite a brand.
Google's current generative-search guidance also warns against pursuing inauthentic mentions as a tactic.
The durable recommendation is reputational rather than mechanical.
Create work worth citing.
Earn relevant references where there is a real editorial or business reason for them.
Publish enough clear first-party information that other sources can describe the business accurately.
Avoid manufacturing signals that have little value outside the optimization tactic itself.
Measurement should match what the interface can actually reveal
Traditional search measurement is imperfect, but marketers are used to impressions, clicks, positions and landing-page traffic.
AI-answer measurement is still evolving.
Edgar's emphasis on experimentation is appropriate because a single "AI rank" can be misleading. Generated answers can change with wording, context, location, model behavior and retrieval.
Google's current Search Console documentation has also changed over time as generative features became more measurable. As of September 2026, Google provides a Generative AI performance report for its Search experiences. That is useful first-party measurement for Google, but it does not turn the wider AI-answer landscape into one stable metric.
A practical measurement system can combine:
- impressions and clicks from conventional search;
- traffic and conversions from AI or referral sources where identifiable;
- Google Search Console's available generative-search reporting;
- query or prompt testing recorded with date, wording and context;
- branded-search changes;
- assisted conversions;
- citations or source appearances sampled over time;
- qualitative sales evidence about how prospects discovered the business.
Use the measurement system to detect trends while avoiding false precision.
If the measurement process cannot be repeated, it is difficult to use for strategy.
Controlled experiments beat folklore
Edgar's testing mindset is one of the strongest protections against AI-search mythology.
A useful experiment starts with a specific question.
Does making a core resource available in server-rendered HTML change how often a target crawler reaches and retrieves it?
Does improving the internal-link path to an orphaned page change discovery?
Does adding a clearly sourced first-party dataset change whether the page earns references?
Does consolidating several weak pages into one stronger resource improve search performance?
The experiment should change as few variables as practical, record the starting state, define the observation window and capture the result.
Even then, treat the outcome as evidence about that site and test.
Search systems change. Competitors change. User behavior changes. Indexes refresh.
The goal is to replace confident guessing with documented learning.
A publishing checklist for search and AI answers
A strong article can be reviewed across four layers.
1. Editorial specificity
- Is the intended reader or user problem clear?
- Does the page answer a question the business is qualified to answer?
- Does it contain specific information rather than category-level filler?
- Is there a clear reason this source should exist alongside competing pages?
- Are examples, tradeoffs and limitations concrete?
2. Evidence and attribution
- Are factual claims supported by appropriate sources?
- Are current platform statements checked against first-party documentation?
- Are guest opinions and professional observations clearly attributed?
- Are case-study outcomes labeled as reported when they have not been independently verified?
- Are uncertain findings presented as experiments rather than rules?
- Are source dates recent enough for the claim?
3. Topical architecture
- Does the page have a clear home in the site's subject structure?
- Does it link to the relevant core service, product or hub?
- Do related pages link back where useful?
- Are headings descriptive enough to help a skimmer understand the argument?
- Does the page avoid duplicating another page's job?
4. Technical access
- Is the page indexable where intended?
- Can the relevant crawlers access it?
- Is important information present in text that the systems can retrieve?
- Do canonical, robots and snippet controls match the publishing intent?
- Are internal links crawlable?
- Does structured data, if present, match the visible page?
- Has the page been tested in the relevant webmaster or inspection tools?
This checklist does not guarantee visibility.
It reduces avoidable reasons for invisibility.
What should a business actually do next?
The strongest synthesis across the interviews and current documentation is conservative in the useful sense.
Keep the foundations.
Make the site architecture clearer.
Go deeper on the specific subjects the business has a reason to own.
Publish evidence that can be checked.
Use internal links to show relationships.
Keep important content technically accessible.
Use AI to accelerate research and drafting where it helps, then apply human judgment before publication.
Run experiments where platform behavior is uncertain.
Measure what can be measured without pretending generated answers behave like a fixed search-results page.
The temptation in a fast-moving field is to chase a new tactic every week.
A better strategy is to build a source that remains useful even when the interface changes.
If the page gives a reader a specific answer, shows where the answer came from, fits into a coherent body of expertise and can be accessed by the systems the business cares about, it has a strong foundation for both search and AI-mediated discovery.
That foundation is less dramatic than a secret GEO trick.
It is also easier to defend when the platforms change again.