The AI Search Playbook: How to Stay Visible When Search Is No Longer Just Google

1. How AI Search Works: Google AI Overviews, ChatGPT, and Perplexity

Understanding why your content is or is not being cited starts with understanding that these three systems work completely differently.

Treating them as interchangeable is one of the most common and most expensive mistakes in GEO right now.

– Annoyed Subu

Google AI Overviews: the query fan-out machine.

When a user types a query that triggers an AI Overview, Google does not simply retrieve the top result and summarise it. It runs what is called a query fan-out it decomposes the original query into multiple related sub-queries and retrieves sources across all of them simultaneously. The pages that appear most consistently across those sub-query results are the ones that end up cited in the Overview.

This changes the targeting strategy entirely. A page that ranks brilliantly for one specific long-tail query may be less likely to get cited than a page that covers a topic comprehensively enough to appear across multiple related sub-queries. Depth and breadth within a topic matter for AI Overview citation in a way they did not for traditional ranking.

AI Overviews appear on approximately 13 percent of all queries as of early 2026, with BrightEdge measuring them on 48 percent of searches by February 2026 up 58 percent year on year. They predominantly trigger for informational and complex queries. Purely transactional queries like “buy laptop” or “order pizza” rarely trigger them. Google is not going to cannibalise its own ad revenue by answering commercial intent queries with an AI summary.

What Google selects as AIO sources has also shifted. Multiple 2025 studies showed a correlation between top-10 organic rankings and AIO citation, but this relationship has weakened since Google upgraded AI Overviews to Gemini 3 in January 2026. Domain-level authority still matters significantly, but the fan-out process means that pages outside the top 10 for a primary query can be cited if they rank well for sub-queries within the same topic.

The practical implication: topical authority across a content cluster is more valuable for AIO citation than a single high-ranking article.

Perplexity: the three-layer quality gate.

Perplexity retrieves candidates from a hybrid index combining Bing’s web index with its own internal cache. Every page in that candidate pool then passes through three stages of evaluation. The first two layers handle semantic relevance and initial filtering. The third layer is the one that matters for GEO strategy: an L3 XGBoost machine learning reranker that scores each source against approximately 59 ranking signals including semantic depth, domain trust, content freshness, and engagement patterns.

A quality threshold determines which sources clear the gate entirely. Sources that fail the threshold are dropped from consideration regardless of how well they performed in earlier layers. Perplexity also maintains manually curated authority domain lists that give algorithmic preference to sources referenced by or associated with high-authority platforms including GitHub, Reddit, LinkedIn, and Amazon. Being cited in those environments is a direct trust signal that influences Perplexity visibility.

The practical result: Perplexity rewards content that opens with a direct, specific, factual answer before elaborating. It rewards original data and expert-attributed claims. It penalises outdated content through a time-decay curve that reduces visibility for pages that have not been recently updated. It is not interested in your introduction, your brand story, or the four paragraphs of context you wrote before getting to the actual answer.

ChatGPT: the consensus aggregator.

ChatGPT’s web browsing capability, when enabled, approaches source selection through a different lens. It looks for consensus what does the collective authoritative internet say about this topic? Rather than primarily evaluating individual page quality, it weighs what multiple authoritative sources agree on.

This means the third-party ecosystem matters as much as your own website. Reddit discussions about your product category. G2 and Capterra reviews. Industry roundup articles that include or exclude your brand. High-authority listicles. If you claim your product is the best but authoritative third-party sources suggest otherwise, ChatGPT will surface the consensus view, not your marketing copy.

The internet always wins the argument against your own website. For ChatGPT to be able to surface your content at all, your robots.txt must allow the OAI-SearchBot crawler. Block it and you do not exist in ChatGPT’s web-browsing results. No exception, no workaround. Check your robots.txt right now.

– Firm Subu

3. Optimising Content for AI Overviews

Every word here is about the practical rewrite, not the theory. If you have read this far and want someone to summarise research at you, there are approximately nine thousand blog posts that will do that. This section is about what to actually change in your existing content.

Answer first. Context second. Always.

The single most impactful structural change you can make to existing content is moving the answer to the first one or two sentences after each heading. Not the second paragraph. Not after the context setting introduction. The first sentence.

AI extraction works by identifying the most concise, accurate, self contained passage that answers the question implied by the surrounding heading. If that passage is in sentence one, the AI can cite it cleanly. If it is in sentence four after three sentences of setup, the AI either skips the page or extracts a garbled version that misrepresents the source.

Go through your highest traffic informational pages right now. For every H2 and H3 subheading, ask: does the very first sentence after this heading directly answer the question in the heading? If the answer is no, rewrite it so that it does.

– Brutally Honest Subu

This is a one afternoon project on most sites and it is one of the highest leverage AIO optimisation actions available.

Fact density over adjective density.

The claim “many users saw significant improvements after implementing this strategy” is useless to an AI system. It is vague, unverifiable, and provides no anchor for factual extraction. The AI will not cite it.

The claim “in Semrush’s 2025 study, 63 percent of businesses reported improved organic visibility after AI Overviews rolled out” is specific, attributable, and factually verifiable. The AI can extract it with confidence because it can verify the source, the number, and the context.

Every piece of informational content needs specific, attributed, verifiable data points woven through it. Not marketing assertions. Not vague summaries. Actual numbers with actual sources. Each data point acts as an anchor for AI extraction.

The more anchors a page has, the more extractable it becomes across multiple related queries. This is why original research, proprietary data, and cited external studies are disproportionately cited in AI Overviews relative to equally well written pages that contain no specific verifiable claims.

Question based H2 and H3 headings.

Google’s fan out process decomposes queries into sub queries and retrieves sources for each. The headings on your page are one of the primary signals the AI uses to evaluate whether your page is relevant to a specific sub query.

A heading that reads “Benefits of Structured Data” is less specific than “What Does Structured Data Do for AI Search Visibility?” The second heading explicitly maps to the type of question a user might ask a conversational search engine. It also tells the AI precisely what the following passage answers.

Rewrite generic topic headings as explicit questions whenever the content is answering a question rather than describing a topic. This is not keyword stuffing. It is structural alignment between your content’s purpose and the query format that AI systems process.

Three to five FAQ questions per important page, formatted cleanly.

FAQ sections are not a vintage SEO technique due for retirement. They are one of the most effective AI citation formats available in 2026. A well structured FAQ with specific, concise answers to common questions on the topic is essentially a pre formatted extraction target for AI systems. Each FAQ item is a self contained question answer unit that the AI can directly cite.

Write FAQ questions in the natural language your audience actually uses. Not “What are the benefits of [service]?” but “Is [service] worth it for a small business?” Answer each FAQ in two to four sentences maximum. The goal is a complete, specific, non hedged answer that stands alone without requiring surrounding context.

– Helpful Subu

5. Conversational Content and Question Based Formats

Here is the query that changed everything for a client of mine last year. They ran a D2C supplement brand. Solid Google rankings on a handful of product-adjacent informational terms. They came to me because their organic traffic had plateaued despite consistent content output and solid on-page work.

I pulled their Search Console data and filtered for queries driving impressions from AI Overview appearances. Zero. Their content was generating zero AI Overview impressions. I looked at how their content was structured. Every article opened with a three-paragraph brand story. The actual answer to the question in the title appeared somewhere around paragraph six.

Headings were topic labels: “Magnesium and Sleep,” “Benefits of Ashwagandha,” “Understanding Cortisol.” Not questions. Labels. Meanwhile, the queries their audience was sending to ChatGPT and Google’s AI Mode were things like: “what magnesium type is best for sleep and anxiety at the same time” and “can I take ashwagandha if I have high cortisol in the morning.” Full sentences. Conversational intent. Specific circumstances baked into the question itself.

Their content had the answers. It just was not structured in a way that let any AI system find them, extract them, or confidently cite them. The rewrite took two weeks. Headings became questions. Answers led every section. FAQ sections with natural-language questions got added to each article. Within eight weeks, AI Overview impressions went from zero to appearing on 23 of their target queries. The content had not changed. The structure had.

– Annoyed Subu

Why search queries are now conversational and why that rewrites your formatting strategy.

Voice search, AI assistants, and the normalisation of talking to search engines like ChatGPT have fundamentally changed the average query structure. Users no longer type “running shoes knee pain.” They ask “what are the best running shoes for someone with bad knees who runs on trails?”

These are not the same query. The second one carries context, specificity, and a scenario that a good answer needs to address directly. The first one is a keyword. The second one is a question looking for a real answer. AI systems are specifically built to process the second format. They parse the question, identify the intent and the constraints within it, and look for sources that address the full question rather than just the surface keyword.

Content that is written as a direct answer to a specific, conversational question is structurally easier for an AI to cite than content written around a keyword.

The structural rewrite: what changes and what does not.

The goal is not to make your content sound like a chatbot wrote it. Subu’s voice, short punchy paragraphs, dry wit, directness all of that stays. What changes is the architecture underneath the voice.

• Question-based headings for every section that answers something. A heading that reads “Keyword Research in 2026” is a topic label. A heading that reads “Is Traditional Keyword Research Dead in 2026?” is a question that maps directly to what users are asking AI systems. The content under both headings can be identical. But the second heading tells the AI immediately: this section answers this specific question. Not every heading needs to be a question. Sections that introduce, connect, or provide narrative context can stay as statements. But every section that answers something the user came to the page to find out should be a question that matches the natural-language version of that thing.
• Answer in the first sentence. Every time. This has been mentioned in the AI Overviews section and it bears repeating here because it is the most consistently missed structural fix on informational content across the web. The first sentence after a question-format heading is the extraction target. The AI looks at the heading, evaluates the first sentence or two of the following content, and decides whether this passage cleanly answers the question. If it does, it is citation-eligible. If the first sentence is context-setting, narrative, or transitional, the AI either moves on or extracts something garbled from further down the section.
• Chunk content into 120 to 180 word sections. Long unbroken sections of prose are hard for AI systems to extract from cleanly. When the relevant answer is embedded in 400 words of flowing text, the AI has to decide where the answer starts and ends. That decision introduces the risk of a garbled or incomplete extraction. Breaking content into sections of 120 to 180 words, each answering one specific sub-question, gives the AI clean, bounded extraction targets. Each section is self-contained enough to be cited independently. The sections string together for human readers into a coherent argument. For AI systems they are a library of citable passages organised by sub-question.

The format that works: heading as a question, first sentence as a direct answer, following sentences as the explanation, evidence, and nuance. This is not dumbing down the content. It is inverting the pyramid that most writers were taught to build right-side up. The conclusion goes first. The supporting argument follows.

– Helpful Subu

FAQ sections: the most reliable AI citation format available.

A properly structured FAQ section is essentially a pre-formatted extraction table for AI systems. Each FAQ item is a self-contained question-answer pair. The AI does not need to parse surrounding context. It does not need to evaluate where the answer starts. It reads the question, reads the answer, and either cites it or moves on.

Write FAQ questions in the exact natural language your audience uses when asking AI systems. Not “What are the SEO benefits of structured data?” but “Does structured data actually help you rank in 2026?” Keep answers to two to four sentences. Direct, complete, non-hedged. If the answer requires more than four sentences to be accurate, it belongs in a full section of its own, not in a FAQ.

Aim for five to eight FAQ questions per page on informational content. Place the FAQ section near the bottom of the page after the main content has established the full context, but before the CTA and sign-off. This placement means users who read straight through see the FAQ as a useful summary, while AI systems parsing the page find a clean, structured extraction target regardless of where they start processing.

7. E-E-A-T in the Age of AI Search

Google added the first “E” for Experience to E-A-T in December 2022. It was not an aesthetic update. It was a structural signal that Google was specifically looking for content produced by people who have done the thing they are writing about, not just people who have researched what others say about it.

In 2026, with AI tools capable of producing credible-sounding content on any topic in seconds, the gap between genuine E-E-A-T and fake E-E-A-T has become the primary quality battleground in search.

Google’s August 2025 Spam Update targeted it directly. And the arms race between AI-generated content and AI-powered detection has made genuine human expertise more algorithmically valuable than at any point in search history.

– Firm Subu

Experience: the signal AI content cannot fake.

Experience is demonstrated through specific, concrete, first-person detail that could only come from someone who has actually done the thing.

The difference: “applying for a business loan requires preparation and documentation” is something anyone can write.

“When I applied for a working capital loan for my second client, the bank required 24 months of GST filings, not 12 as stated on the website, and we nearly missed the approval window because we found out three days before submission” is something only someone who has been through that process can write.

AI systems can detect the difference through the specificity of the detail, the presence of outcome data, the candour about what went wrong, and the absence of the hedged, cover-all-bases language that characterises synthesised content.

Content that demonstrates genuine first-hand experience with specific processes, timelines, outcomes, and failures is structurally harder to replicate with AI and structurally more likely to be cited by AI as a trustworthy source.

Build real experience signals into content: first-person case studies with specific data, documented processes with genuine intermediate steps including the ones that did not work, original before-and-after analyses with methodology described, and expert commentary that reflects an actual professional opinion rather than a synthesised consensus.

Expertise: author credentials as verifiable entity signals.

An article attributed to a real person with a verifiable professional profile is more likely to be cited by AI systems than an article with no author, or one attributed to a generic author name with no external presence.

Build author entity pages on your site for every regular contributor. Include professional credentials, experience in the specific subject area, links to published work elsewhere, and a profile photo.

Link every article’s author byline to this author page. Include Person schema on the author page with the sameAs property linking to the author’s LinkedIn profile and any other verified external profiles.

This creates a verifiable entity chain: Article → Person entity → External professional presence.

An AI system that can trace this chain has a basis for evaluating the author’s expertise before citing the content. An AI system that finds no author information has no basis for that evaluation and defaults to domain-level authority signals instead.

Authoritativeness: the third-party validation layer.

Authority is not claimed on your own website. It is assigned by others.

The signals that build domain authority for AI citation purposes: external publications citing your content, experts in your field referencing your work, your domain appearing in industry-standard resource lists, your authors being quoted in news coverage, and your brand appearing as a consensus recommendation in community discussions.

This is why the off-page work described in the ChatGPT section maps directly to E-E-A-T. Building genuine authoritative presence in the places AI systems use as consensus signals is not just a ChatGPT strategy. It is the authoritativeness component of E-E-A-T applied to the AI search environment.

– Helpful Subu

Trustworthiness: the signals Google explicitly checks.

Google’s Quality Rater Guidelines flag specific trust deficiencies that directly impact AI Overview citation eligibility: no clear author attribution, no accessible contact information, no transparent editorial standards, no disclosed business relationships, and no mechanism for correcting factual errors.

Make sure every page has a clearly identified author. Your About page should identify who runs the site and what their credentials are.

Your contact information should be accessible. If you publish affiliate content or sponsored content, disclose it.

These are not just ethical standards. They are trust signals that AI systems use to evaluate whether a domain is a legitimate, accountable source.

9. How AI Is Changing Keyword Research

Traditional keyword research is not dead. It has just become the floor rather than the ceiling.

The marketers who built entire strategies around monthly search volume rankings are discovering that a keyword with 10,000 monthly searches and a Google AI Overview answering it completely is now worth a fraction of what it was two years ago.

The click never materialises. The content ranks. The traffic does not arrive.

The strategy looks correct on paper and fails in revenue.

Here is what has actually changed and what it requires you to do differently.

The click-value problem with high-volume informational keywords.

Search volume is a measure of how many people ask a question.

It has never been a measure of how many people will click through to your page for the answer.

Those two numbers used to be correlated closely enough that volume served as a reasonable proxy for traffic potential.

In 2026, AI Overviews have severed that correlation for a specific and growing category of queries.

Any query that is primarily informational, has a clear and concise answer, and does not require current, source-specific, or deeply opinionated information is a candidate for zero-click AI Overview treatment.

For those queries, the click-through rate collapses regardless of ranking position.

First place gets clicks that second place used to get. The whole category shrinks.

The keyword research implication: before investing in content for any informational target, check whether that query currently triggers an AI Overview. Open an incognito browser, search the query, and observe whether an AI Overview appears above the organic results. If it does, evaluate the Overview’s quality. If it answers the question thoroughly, your traffic potential for that keyword is dramatically lower than the search volume suggests.

– Firm Subu

Intent specificity over volume: the long-tail shift.

The queries that still reliably drive clicks are the ones that require source-specific, opinionated, contextual, or current information that AI cannot fully satisfy from its training data or retrieved summaries.

“What is bounce rate” AI Overview completely satisfies this. Zero click value.

“Our bounce rate went from 45 percent to 78 percent after a site redesign, what should we check first” This is a scenario-specific diagnostic question.

The AI can provide a generic framework. A practitioner with experience in post-redesign traffic analysis can provide a specific, prioritised diagnostic process.

That specific answer is still worth a click.

The keyword research shift is toward extreme specificity and scenario-based targeting.

Not “best CRM software” but “best CRM software for a five-person SaaS sales team managing enterprise accounts with under 10,000 annual contacts.”

Not “how to write meta descriptions” but “how to write meta descriptions for e-commerce category pages when products overlap across multiple collections.”

The more specific and scenario-embedded the query, the more likely it is that AI cannot fully satisfy it, and the more likely a user is to click through to a source that engages with their specific situation.

Semantic clustering over individual keyword targeting.

AI search systems do not rank individual pages for individual keywords in isolation. They evaluate topical authority across content clusters.

A domain that has ten articles covering different angles of the same subject area, each linking to the others, each reinforcing the same topical entity signals, has higher AI citation probability for queries on that subject than a domain with one excellent article on the same subject surrounded by unrelated content.

The keyword research process now needs to build topic clusters rather than keyword lists. Start with a core subject.

Map every sub-topic, related question, adjacent concept, and scenario-specific variation. Build content that covers the cluster comprehensively.

The individual keyword volumes matter less than whether the cluster as a whole establishes genuine topical authority that AI systems recognise as expert coverage of the subject.

Mining conversational AI for keyword discovery.

The best source of natural-language keyword data in 2026 is the AI systems themselves.

Type the beginning of a query into ChatGPT, Perplexity, or Claude and observe the related questions they surface. Ask Perplexity “what do people want to know about [topic]?” and it will return a structured list of the questions its users actually ask on that subject.

– Helpful Subu

This data is more current and more conversational than any traditional keyword tool’s autocomplete.

It reflects how users phrase questions when they are not constrained by the keyword-optimised search bar muscle memory that traditional Google search created.

Build your keyword strategy around how people actually talk to AI systems, and your content is structurally better positioned for AI citation from day one.

11. The Future of SEO: What Still Works and What Does Not

Let’s clear this up because there is an enormous amount of confused hot-taking on both ends of this conversation.

“SEO is dead” and “nothing has changed” are both wrong. Here is what is actually true.

What is absolutely still working and matters more than ever.

Technical health is the table stakes that never changes. If Googlebot cannot crawl your site, if AI crawlers are blocked in your robots.txt, if your pages take six seconds to load on mobile, if your JavaScript is preventing content from rendering nothing else in this playbook matters.

The foundational technical requirements of crawlability, renderability, and page speed are not legacy SEO concerns. They are prerequisites for everything.

Backlinks remain a significant signal, but the reason they matter has evolved.

A strong backlink profile was always a proxy for the judgment of the broader web about your content’s quality.

In the AI era, that proxy relationship has become more direct: the same external links that pass PageRank authority are also the citations that appear in the authoritative sources AI systems use as training data and live retrieval signals.

Building genuine editorial backlinks is simultaneously traditional SEO and LLM brand authority building.

Deep topical authority has become more valuable, not less. The content cluster model where one domain covers a subject comprehensively enough to be the authoritative resource on it is exactly what AI systems reward through the fan-out citation logic described in section two.

Narrow, single-article coverage of a topic is increasingly insufficient for sustained AI visibility.

Comprehensive cluster coverage that demonstrates genuine subject expertise is the durable strategy.

What has stopped working and will not come back.

Keyword stuffing and optimising for exact-match density never made content better for readers.

It made it visible to a crawling algorithm that evaluated keyword frequency as a quality signal.

Modern algorithms, and certainly AI systems, do not work that way. Keyword density optimisation is a 2015 strategy. Stop.

– Annoyed Subu

Vague, padded, context-heavy introductions that delay the actual answer by four paragraphs exist to make word counts feel justified.

AI systems have no patience for them. Neither do users anymore.

Every piece of content should answer its primary question in the first two sentences after the main heading.

The context and nuance can follow.

Mass-produced AI content generated at scale to cover every keyword variation with thin, templated responses actively triggers Google’s spam detection systems post-August 2025 update.

The store that publishes three hundred AI-generated product comparison articles, all following the same template, all adding no original insight, is not building topical authority.

It is building a spam profile that will eventually be penalised. Scale and quality are not interchangeable.

Generic guest posting for link building without editorial selectivity about placement quality is increasingly transparent to both algorithms and AI systems.

A thousand links from low-authority, topically irrelevant domains contribute less to AI citation probability than ten links from genuinely authoritative sources in your field.

The link quality bar has risen with the sophistication of the systems evaluating it.

The practical reframe.

Traditional SEO gets you into the index. Generative Engine Optimisation gets you into the answer. These are not competing strategies.

They are sequential dependencies.

A page that cannot rank cannot be cited. A page that ranks but is structured for keywords rather than extraction cannot be cited.

A page that is well-structured but has no E-E-A-T signals will not be trusted enough to be cited.

The full stack in 2026: technical health → crawlability for all bots → ranking eligibility → E-E-A-T signals → extraction-ready structure → schema markup → third-party consensus signals → LLM citation. Each layer depends on the one below it. Skip a layer and the ones above it underperform regardless of how well they are executed.

– Helpful Subu

TL;DR: The AI Search Survival Protocol

Print this. Screenshot it. Stick it somewhere you will look at it every time you are about to spend three hours writing a 2,000-word article that answers a question Google’s AI Overview already answers in four sentences.

– Brutally Honest Subu

Zero-click is structural, not temporary.

64.82 percent of searches end without a click. On mobile, 77.2 percent. This is not a bug. It is the product. Plan around it.

Answer first. Always.

The first sentence after every heading is the extraction target for every AI system reading your page. If that sentence is context-setting prose instead of a direct answer, you have missed the citation opportunity. Rewrite every important page with this principle applied to every section.

Fact density is citation currency.

Replace “many users saw improvements” with “63 percent of businesses reported improved organic visibility in Semrush’s 2025 study.” Specific, attributed, verifiable numbers are what AI systems anchor citations to. Vague assertions get ignored.

Structure for machines without writing like one.

Question-based headings. Chunked 120 to 180 word sections. Five to eight FAQ questions per page with FAQPage schema. Answer-first passages. None of this requires you to sound like a corporate knowledge base. Subu’s voice lives in the sentence-level writing. The structure underneath it serves the machines.

Schema is now a citation prerequisite, not an enhancement.

Organisation schema. Article schema with author entity. FAQPage schema. Product schema with availability in Schema.org vocabulary, numeric price, MerchantReturnPolicy, and ShippingDetails. Validate everything in the Rich Results Test. Fix every error before publishing at scale.

You cannot win ChatGPT on your own website.

ChatGPT reads consensus. Build your presence on Reddit, G2, Capterra, authoritative industry publications, and high-ranking listicles in your category. If the authoritative internet does not know you exist, ChatGPT does not know you exist.

Perplexity will time-decay you into irrelevance if you do not update your content.

Freshness is a documented ranking signal. Update your most important existing content quarterly at minimum. Change the dateModified in Article schema. Add new data. Remove outdated references. Give Perplexity’s freshness algorithm a reason to keep you in the results.

Check your robots.txt for AI crawlers right now.

OAI-SearchBot, PerplexityBot, ClaudeBot. If any of them are blocked, you do not exist on those platforms regardless of how good your content is.

Traditional SEO is the floor, not the ceiling.

Technical health, crawlability, and ranking eligibility are prerequisites for AI visibility. A page that cannot rank cannot be cited. Fix the foundation before expecting the superstructure to perform.

Measure citation rate, not just click-through rate.

Track your brand mention frequency across ChatGPT, Perplexity, and Claude on non-branded queries monthly. This is the leading indicator of AI search visibility that traffic metrics will never show you directly.

The era of ten blue links is done. The era of being the source the AI trusts enough to quote has started. These skills are learnable. The mechanics are specific. The stores, brands, and practitioners who understand them early will own disproportionate visibility in the answer economy while everyone else is still writing keyword-stuffed blog posts for a search engine that no longer exists in its old form.

Get into the answer. Everything else is noise.

Subu, SEO by Subu