The AEO and GEO Playbook

The AEO and GEO Playbook in a nutshell

• Search is now generative. Buyers move between Google, ChatGPT, Perplexity, Gemini, Copilot, and Claude in a single research session. The page of ten blue links is no longer where the decision gets made.
• AEO and GEO are the ground floor, not an add-on to SEO. Brands that treat them as a side project lose visibility slowly for months, then all at once.
• Nobody has this fully solved. Models and their citation behavior shift every few weeks. If someone guarantees you a fixed method, they are selling confidence they have not earned.
• The brands that win run a loop. They publish, measure how often they get cited, learn, and refresh on a weekly cadence.
• Citation has replaced ranking. Being the source an AI answer quotes is worth more than sitting third on a page nobody scrolls.

I wrote this as the working manual we use inside SOLID with our clients. It is opinionated, it does not favor any one vendor, and we keep updating it as the engines change. Treat it as a reference you act on, not a script to follow line by line.

Chapter 1: Why search just changed (and why this playbook exists)

1.1 From ten blue links to one synthesized answer

For twenty years, search gave you choices. You typed a question, scanned ten results, clicked one, judged it, and maybe went back for another. The brands that won were the ones that could rank, write a headline worth clicking, and earn that click.

That model is giving way. A large and growing share of queries now end with an answer instead of a list. Google summarizes the top of the page in an AI Overview. ChatGPT, Perplexity, Gemini, and Claude answer the question directly and cite a few sources underneath. The ten-link page still exists, but for the queries that decide your revenue, it is not where your buyer’s attention lands anymore.

That costs you something real. The work brands poured years into, exact-match keyword pages, click-through rate, meta descriptions, counts for less than it used to. What matters now is whether your brand shows up inside the answer itself, named, quoted, or linked. A page that ranks well and never gets cited is, more and more, a page nobody reads.

1.2 AEO vs. GEO vs. SEO

The acronyms have outrun the clarity, so here is how we use them at SOLID.

• SEO (Search Engine Optimization) is the long-running discipline of being found in traditional search results. Crawling, indexing, ranking, the click. It still matters, and most AI engines lean on that same search infrastructure underneath.
• AEO (Answer Engine Optimization) is the practice of structuring your content so an AI engine can lift it cleanly into an answer. Google AI Overviews, ChatGPT search, Perplexity, Gemini. It rewards clarity, structure, and a sharp definition of who you are.
• GEO (Generative Engine Optimization) is the wider game: being cited, referenced, and remembered by generative systems even when they are not running a live search. GEO includes AEO and adds off-site presence, entity authority, training-data reputation, and how your brand co-occurs across the web.

In practice, AEO is mostly content and technical work. GEO sits closer to digital PR fused with brand strategy. Serious programs need both running at once.

1.3 The new visibility stack: Google, ChatGPT, Perplexity, Gemini, Copilot, Claude

Your buyer no longer has one front door. Depending on who they are, where they are, and what they are trying to do, a single research journey now crosses several surfaces.

• Google still owns the largest share of search by volume, but AI Overviews now sit above the organic results on a growing share of queries.
• ChatGPT has become a primary research surface for professionals, with hundreds of millions of weekly active users globally according to public usage data.
• Perplexity has won over the power users who want sourced, cited answers for serious research.
• Gemini is embedded across Google Workspace and Android, so it shows up in the middle of work, not only in moments of search.
• Microsoft Copilot runs inside Windows, Edge, and Microsoft 365, often right in the workflow of a B2B buyer.
• Claude is widely used by knowledge workers, especially for longer analysis where careful citations matter.

Chasing one of these is a trap. The work is to build a content footprint and an entity definition that hold up across all of them, then watch each surface on its own, because they cite very differently.

Chapter 2: A field that is still being written

Before the tactics, a word on how new this is and how fast it moves.

2.1 Why nobody has a fixed playbook

If someone tells you they have AEO and GEO completely figured out, be skeptical. The honest position in 2026 is that the field is still being written. The models get retrained, the way they pick citations shifts, new surfaces appear, and a tactic that worked in January is only half-true by May.

This is not a marketing problem you can wait out. It is what happens when you work inside systems that are being rebuilt while you use them. Anyone selling a fixed, guaranteed method is either selling a tool or has not been doing the work long enough to see how quickly it dates.

What holds up is the operating loop. Ship work grounded in what performs today, watch citation behavior continuously, and refresh both content and tactics on a fast cadence. The loop outlasts any single tactic, because the tactics keep rotating out from under you.

2.2 What we actually know vs. what we are testing

We know a few things with high confidence:

• Clear, extractable content with strong H2/H3 hierarchy gets quoted more often than dense walls of text.
• Entities defined consistently across your site, schema, and external mentions are easier for models to attach claims to.
• Citation by AI engines tracks closely with traditional search authority. The brands ChatGPT cites are usually the brands already ranking well in Google.
• Freshness counts in answer engines, often more than in classic SEO. Stale pages get demoted in the synthesis layer.

And there is a lot we are still testing as I write this:

• The exact weight of schema markup versus raw HTML structure.
• The trade-offs of blocking versus allowing specific AI crawlers.
• How much of your “share of voice” inside a model comes from training data versus live retrieval.
• How durable a citation ranking is over months and quarters.

That list grows as the surfaces change, and we are running more experiments than we can keep current on a page like this. So treat any number anyone hands you on these open questions as a snapshot, not a law.

2.3 How the models change month to month

Two separate things move. The models themselves change as new versions ship, retraining cycles complete, and routing logic gets updated. Around them, the surfaces change too: how AI Overviews are presented, how Perplexity picks its sources, what ChatGPT does by default with its web tool, how Copilot grounds an answer.

The practical effect catches people out. A citation-rate measurement from three months ago is closer to history than to current state, and decisions made on stale data tend to over-correct.

2.4 How SOLID stays current

Inside client work we run three loops at different cadences.

Every week we run a prompt panel: 25 to 75 questions a real buyer would ask, fired across ChatGPT, Perplexity, Gemini, and Google AI Overviews, with a log of who gets cited. Every month we work a refresh queue of the top 10 to 20 pages, checking each for accuracy, freshness, and fit with how the engines are quoting right now. Every quarter we audit the brand’s entity: its definitions, schema, founder bios, and the way other sites describe it, watching for drift.

One loop on its own does not move much. Run together, they compound.

2.5 How to read the rest of this playbook

Read this as a working manual. Where the claims are strong, they are strong because we have tested them across client engagements. Where something is still uncertain, I have flagged it, and you should treat it as an open experiment rather than settled advice.

Chapter 3: How answer engines actually work

You do not need to be an ML engineer to do this well. You do need a working picture of what happens between the moment your buyer types a question and the moment the engine answers.

3.1 Retrieval, ranking, and generation

Almost every AI answer is the product of three steps, even though the user only sees one.

1. Retrieval. The engine pulls a set of candidate sources. For Perplexity and ChatGPT search, that is a live web search. For a pure generative answer, it can come from training data or a retrieval-augmented generation (RAG) pipeline.
2. Ranking. The candidates get scored and a subset is chosen. Traditional SEO signals carry real weight here, because the retrieval and ranking layers often reuse classical search infrastructure.
3. Generation. A language model writes the answer from the chosen sources, usually with citations attached.

Most of your AEO and GEO work goes into making content easy to retrieve, worth ranking, and clear enough to be useful when the model writes its answer. The generation step is the one teams tend to overlook. A page can rank perfectly and still be too dense or too vague for the model to quote without effort, so it gets passed over for one that reads cleaner.

3.2 How ChatGPT, Perplexity, Gemini, and Google AI Overviews source citations

Each surface behaves differently, and those differences should shape where you put effort.

• ChatGPT search tends to cite a small handful of authoritative sources per answer, leaning on recognized brands and well-structured pages.
• Perplexity usually cites more sources per answer, mixing mainstream and niche outlets, and rewards content that is direct and well organized.
• Google AI Overviews lean on the existing search index, so sites that already rank well are the ones that get cited.
• Gemini behaves like AI Overviews on commerce queries and more like ChatGPT on long research queries.
• Claude with web access cites fewer sources but weighs them carefully, and often prefers a primary source over an aggregator.

These behaviors change over time, so it is worth running the queries yourself and checking how each engine cites for your category rather than relying on a summary like this one.

3.3 Query fanout: why one user question becomes ten sub-queries

An engine rarely treats a question as a single lookup. It fans the question out into sub-queries that approach the topic from different angles, then pulls the results back together.

Ask “what is the best CRM for a 50-person SaaS company” and the engine may quietly run sub-queries on CRM pricing tiers, feature comparisons, integration ecosystems, support reviews, and migration cost. Your answer is assembled from all of them.

So optimizing one page for the head term and hoping to win is a losing bet. You need a cluster of content that covers the sub-queries the engine will explore on its own. This is why pillar-and-cluster architecture matters more in AEO than it ever did in classic SEO.

3.4 The role of training data vs. live retrieval

Your brand can land in an AI answer two ways.

1. Training data. What the model absorbed about you during its training run. Durable, but slow to update. Brands with a strong, consistent web presence built up over years tend to be well represented.
2. Live retrieval. The real-time searches a model runs to ground its answer. Fast to update, but only triggered when the engine decides it needs to look something up.

GEO is mostly aimed at the training data. AEO is mostly aimed at live retrieval. You want to be present in both, because you do not control which one a given answer leans on.

3.5 Why “ranking” is being replaced by “citation rate”

Position one on Google still has value. But for a growing slice of queries, your buyer never sees the results page at all. They see the answer. The question stops being “where do I rank” and becomes “how often am I quoted.”

Citation rate, the share of relevant prompts where your brand turns up as a cited source, is the closest thing AEO and GEO have to a north star metric.

Chapter 4: The new metrics that matter

4.1 From rankings to share of model voice

Share of model voice is the AEO equivalent of share of search. For a defined prompt panel, it is the percentage of answers your brand shows up in, weighted by how much each prompt matters to the business. It is the cleanest top-line number we have found.

4.2 Citation rate, inclusion rate, and answer presence

We track three related numbers per client, and they are not the same thing.

• Citation rate. The share of prompts where your domain shows up as a cited source link.
• Inclusion rate. The share of prompts where your brand name appears in the answer text, with or without a link.
• Answer presence. Whether your brand shows up at all, cited or mentioned, on a given prompt.

Citation rate is the strictest, because the engine has to grant you authority to link you. Inclusion rate captures awareness even when there is no link. Answer presence is the loosest read of the three.

4.3 Branded prompt volume

The new version of branded search is the branded prompt: someone asking ChatGPT about your brand by name, or asking Perplexity to compare you to a competitor. We estimate branded prompt volume by sampling prompts and watching how often people name the brand directly versus arriving at it indirectly.

It is harder to measure than branded search, because the engines do not publish search-volume data. Proxy signals fill the gap: navigational traffic from AI referrers, mentions in prompts people share publicly, and direct traffic that spikes after a piece of content goes live.

4.4 Attribution in a zero-click world

The hardest conversation I have with a CMO in 2026 is about attribution. AI engines often send no click at all, even when they cite you. Your brand gets the credit and your analytics get nothing.

Here is the framing that helps. AI visibility behaves more like a billboard or earned PR than like performance marketing. It builds consideration and trust, often invisibly, and the brands taking it seriously have already made peace with the fact that not every revenue dollar will trace back cleanly.

A few things make it less murky in practice. Track referrer traffic from the AI surfaces, which is climbing fast. Watch branded direct traffic in the weeks after a content launch. Ask new customers during onboarding where they first heard of you. And run the occasional controlled holdout on a content investment to see what moves.

4.5 The KPIs to put on the executive dashboard

If AEO and GEO performance has to fit on a single slide, use five charts.

1. Share of model voice across your top 50 prompts, trended weekly.
2. Citation rate per engine, so you can see where you are winning and where you are slipping.
3. Branded prompt volume as a proxy for awareness.
4. AI-referral traffic from ChatGPT, Perplexity, Gemini, and Copilot.
5. Pipeline or revenue from AI-influenced channels, measured as honestly as the data allows.

Perfection is not the goal. A small, durable set of numbers the executive team will actually watch over several quarters beats a precise dashboard nobody opens.

Chapter 5: The AEO and GEO content framework

5.1 The extractable-chunk principle

Write so an engine can quote you without doing extra work. An extractable chunk is a 40-to-80-word passage that answers one specific question and needs no context from the paragraphs around it.

One caveat worth being honest about: Google has said publicly that writing in discrete chunks is not necessary, and that its systems can understand normal, well-structured prose. Many SEOs structure content into extractable chunks anyway, because in practice it tends to get quoted more cleanly across the answer engines and it costs you nothing in readability. Treat it as a useful habit rather than a hard rule.

A non-extractable paragraph reads like a flowing argument that only makes sense in sequence. An extractable one reads like a clear claim followed by a short explanation. Your page needs both, but only the second kind earns citations.

There is a simple test for it. Take any paragraph, paste it into a blank document, and read it cold. If it stands on its own, an engine can lift it. If it leans on the sentence before it, rewrite it until it does not.

5.2 Question-first structure

Your H2s and H3s should echo the questions real people ask. “What is X.” “How does X work.” “When should you use X.” “X vs. Y.” Engines are trained to pair questions with answers, and your heading hierarchy is the clearest signal you can hand them.

Clever headings work against you here. “The new frontier of digital intelligence” might read well in a print magazine. To an engine trying to map your heading to a user’s intent, it is noise.

5.3 The “in a nutshell” summary pattern

Every long page benefits from a short, dense summary near the top. We label it “In a nutshell” or “[topic] in a nutshell” instead of “TL;DR,” because it scans as natural prose and the engines treat it as real content rather than a meta tag.

Keep it to four to six bullets, each one a complete idea, most important claim first. This is the section most likely to be quoted word for word, so write it last, once the body is finished and it can honestly reflect the page.

5.4 Entity clarity

An entity, in the AI sense, is a thing the model can identify and pin claims to. Your brand is an entity. So are your products, your founder, and each major service you offer.

For every entity that matters to the business, write a one-sentence definition and use it consistently across your site, your schema, your About page, and your external profiles. Inconsistency blurs the picture. “SOLID is a paid media agency” on one page and “SOLID helps brands grow with AI-driven marketing” on another splits the entity into two weaker versions of itself.

5.5 FAQ blocks, comparison tables, and review summaries

These three formats show up in AI citations more than their share, because they match the shape of a generated answer almost exactly.

• FAQ blocks answer common sub-queries in a structured form. Add FAQ schema where it fits.
• Comparison tables make differences explicit, which engines reward on shopping and software queries.
• Review summaries synthesize outside opinion, which is the same job the engine is trying to do.

There is nothing magic about them. They get quoted because they are easy to quote.

5.6 Freshness signals and dated claims

Engines weight freshness more heavily than they did even in 2024, and the signals are concrete.

• Visible “Published” and “Updated” dates on the page.
• Year-specific phrasing where it is relevant (“as of 2026”).
• Schema with accurate datePublished and dateModified values.
• Genuine updates rather than cosmetic timestamp changes, since some engines now compare the current version against the previous one.

A page that still says “2024” in its body in the middle of 2026 is being quietly demoted in answer engines, whether or not it is holding its Google ranking.

5.7 Non-commodity content: the new bar Google is setting

Google’s recent guidance, reinforced through the Helpful Content System, the updated Search Quality Rater Guidelines, the March 2024 spam policy on scaled content abuse, and the AI features and your website optimization guide, has made one thing clear. Commodity content no longer earns visibility. Commodity content is anything a generic AI model, a junior writer with a keyword brief, or a five-minute web summary could produce. Definitions, history filler, restated competitor outlines, and AI-style “ultimate guides” that compile public knowledge without judgment all sit in that bucket.

The same shift is underway across the answer engines. A model trained on the open web has already seen every commodity take on every common topic. What it quotes is the source that adds something the rest of the internet does not have.

So for any brand publishing in 2026, the bar is no longer “is this well written and on topic.” The bar is “could only this brand, this founder, this team, this client base have produced this.” If the answer is no, the page gets ignored by Google and the AI engines alike, however clean the schema or however tight the internal linking.

What non-commodity content actually looks like, in rough priority order:

• First-hand experience. What your brand, founder, or team learned doing the thing. Specific incidents, projects, clients, or workflows nobody else can write about.
• Original data or research. Internal numbers, survey results, anonymized customer data, audit findings, or benchmarks your team actually ran.
• Behind-the-scenes operational detail. How a decision got made, what the trade-offs were, what changed after rollout, what your team measures and why.
• Specific case studies and failure stories. Named situations with real inputs and outcomes, including what did not work and why.
• Counterintuitive or contrarian expert takes. Where the brand publicly disagrees with the conventional advice in its category, with the reasoning and evidence to back it.
• Decision criteria from real engagements. The actual checklist, scoring rubric, or framework your team uses internally, not a generic list pulled from other articles.
• Expert caveats and edge cases. Known gotchas, regulatory nuance, customer-segment exceptions, the scenarios where the standard advice breaks down.
• Synthesis the reader cannot get elsewhere. Comparing primary sources, reconciling contradictory data, or pulling a fragmented topic into one clear view.

Patterns to retire from your editorial calendar:

• Generic listicles built from common knowledge (“7 tips for first-time buyers” with no insider angle).
• Restating what other sites already say, rewritten in your tone of voice.
• AI-style topic summaries with no brand voice, no experience, and no point of view.
• Broad definitions and history filler at the top of every page.
• “Ultimate guide” structures that compile public knowledge without adding judgment, hierarchy, or opinion.

A practical test: if you deleted every paragraph that could appear on a competitor’s site without anyone noticing, would the page still be worth publishing? If the answer is no, it is commodity, and it should not ship.

Chapter 6: Technical foundations

You cannot optimize what an engine cannot crawl, parse, or render. The technical layer rarely gets attention, but it sets the ceiling on everything else you do.

6.1 Schema markup that AI engines actually use

Schema is having a second life in the AI era. The types that consistently earn their keep:

• Article for editorial content and playbooks.
• FAQPage for question-and-answer sections.
• Product for ecommerce listings.
• Organization for your brand, with a clean sameAs array pointing to verified profiles.
• HowTo for step-by-step instructions, used sparingly, since Google scaled back its rich-results treatment but the engines still parse it.
• BreadcrumbList for hierarchy.

Validate every block in Google’s structured data tester. Broken schema is worse than none, because it teaches the engine to distrust the markup it does find.

6.2 Crawlability for GPTBot, ClaudeBot, PerplexityBot, Google-Extended

The AI crawlers announce themselves with distinct user-agent strings. The ones worth knowing:

• GPTBot (OpenAI)
• ClaudeBot (Anthropic)
• PerplexityBot (Perplexity)
• Google-Extended (Google’s training crawler)
• CCBot (Common Crawl)
• Bingbot (Microsoft, including Copilot)

Most sites should allow all of them by default. Blocking can make sense for a publisher sitting on paywalled content or a unique data asset, but for a growth-stage brand, blocking is close to opting out of the next decade of discovery.

6.3 The robots.txt decision

If you allow every AI crawler, do it on purpose and write down why. If you block some, document the business reason for that too. The worst setup is a silent default that nobody on the team chose or understands.

A sensible default for a growth-stage brand allows GPTBot, ClaudeBot, PerplexityBot, Google-Extended, CCBot, and Bingbot, with the same crawl-delay and disallow rules you already apply to Googlebot.

6.4 Sitemaps, canonicals, and rendering for LLM crawlers

Sitemaps are still the cleanest way to declare what you want crawled. Keep your XML sitemaps current, include lastmod dates, and split them logically once you pass 50,000 URLs.

Canonicals matter more than they used to. When an engine finds duplicate or near-duplicate pages it picks one and demotes the rest, and with messy canonicals it often keeps the wrong one.

Rendering is the quiet killer. If your most important content renders client-side and the crawler does not run JavaScript, you are invisible to it. Server-side rendering, static generation, or hybrid rendering with prerendering keeps you safe. We see this break most often in React and Vue apps that grew faster than their SEO discipline.

6.5 Core Web Vitals in an AI-mediated world

Core Web Vitals (LCP, INP, CLS) still count. Engines read slow or unstable pages as lower quality in both classic ranking and retrieval scoring. The target has not changed: fast loads, stable layouts, responsive interactions.

Chapter 7: Entity and authority building

This chapter looks the most like brand work, because it is brand work.

7.1 What an entity means to an LLM

When a model reads about your brand, it is building or updating a probabilistic picture of what you are, what you do, who you serve, and how you relate to the other entities in your category. The more consistent your inputs, the sharper that picture gets.

Inconsistency dulls it. When your site, your LinkedIn, your Crunchbase profile, your press, and your podcast appearances each describe you a little differently, the model averages across them and loses precision on all of it.

7.2 Wikipedia, Wikidata, and the knowledge graph

A Wikipedia page is one of the most powerful entity signals you can earn, because nearly every major model has trained on Wikipedia and keeps doing so.

You cannot self-publish one. What you can do is become eligible by accumulating independent, secondary coverage in named outlets, then have an experienced editor, not your marketing team, draft a neutral, well-cited entry.

Wikidata is the structured layer beneath Wikipedia and it is far more accessible. A clean Wikidata entry helps engines connect your brand to the entities around it.

7.3 Building unambiguous brand definitions

Settle on a one-sentence definition of your brand. Use it on your homepage hero or close to it, in your Organization schema, in your social bios, in your founder bios, and in your boilerplate.

Then repeat it, and resist the urge to reinvent it for variety. Variety serves storytelling. Consistency serves the entity, and the entity is what the engines are reading.

7.4 Consistent NAP, founder bios, and structured “About” data

NAP consistency (name, address, phone) is older than AEO and still matters. Extend it to founder names, founding date, headquarters city, and a single “what we do” descriptor.

Then cross-check the top 30 places your brand appears: directories, social profiles, app stores, partner pages, conference bios, podcast show notes. Drift creeps in everywhere, so audit it twice a year.

7.5 The role of third-party validation

Press, podcast appearances, awards, partnerships, and customer case studies all tell the engine that someone other than you vouches for you. Models weight that third-party validation more heavily than anything you say about yourself.

This is the point where GEO and PR become the same discipline. The brands winning citation share are, almost always, the brands earning real third-party mentions in named publications.

Chapter 8: Off-site GEO: getting cited where models learn

If your entire AEO strategy lives on your own site, you are leaving most of the leverage on the table.

8.1 The sources LLMs trust most

Models do not weigh all sources equally. The patterns hold up across engines:

• Major mainstream publications (Reuters, AP, BBC, FT, NYT, WSJ).
• Trade publications respected inside their industry (TechCrunch in tech, Modern Retail in commerce, and so on).
• Reference sites such as Wikipedia, Britannica, and government domains.
• Substantive long-form blogs from recognized practitioners.
• Reddit and YouTube transcripts, weighted heavily for certain kinds of queries.

Get cited inside these and the visibility compounds across every surface at once.

8.2 Digital PR for AI visibility

Digital PR is the most underrated GEO lever there is. The old version was scored on domain authority, links, and referral traffic. The new version is also scored on what gets remembered by the models.

The best digital PR for GEO is not a press release. It is original data, sharp expert commentary, or a genuine opinion tied to your brand, placed in outlets the engines already trust.

8.3 Reddit, Quora, YouTube, and forum presence

Models read these heavily, and Reddit threads in particular surface in AI citations far more often than most marketers expect.

That is not an invitation to spam. Models can tell the difference between a contribution a community welcomed and a promotional drop nobody asked for. Pick two or three communities next to your category and contribute something real over months, not days.

8.4 Industry roundups, “best of” lists, and comparison sites

If your category has “best X” or “top X” lists in major outlets, getting into them is high leverage. Those pages are cited heavily on shortlist and comparison queries.

The way in is rarely a cold pitch. It is a relationship with the writer, original data they can quote, and a track record their editor can verify.

8.5 Podcast appearances and transcript indexability

Podcasts have turned into a real GEO surface, because transcripts are increasingly indexed. A founder appearance on a respected show can now do more for entity authority than a guest post would have five years ago.

Two things make the difference. Make sure the host publishes a full transcript. And early in the conversation, name your brand and your category clearly at least once, because the models extracting from a transcript need an anchor phrase to attach the rest to.

Chapter 9: Content architecture: pillars, clusters, and playbooks

9.1 The pillar and cluster model rebuilt for AEO

Pillar-and-cluster was already best practice in classic SEO. In AEO it is the price of entry, because query fanout forces you to cover the whole surface of a topic rather than just the head term.

A pillar page covers the topic in full. Cluster pages handle the specific sub-topics, sub-queries, and use cases, and they link tightly back to the pillar and across to each other.

This playbook is a pillar. The AI engine optimization guide is a cluster page that feeds up into it.

9.2 Programmatic pages without the spam penalty

Programmatic SEO can still work, but the threshold has jumped. Thin programmatic pages now get demoted in answer engines even faster than they did in classic search.

The test is simple: does each programmatic page carry at least one piece of unique, useful information the others do not? If not, consolidate them until they do.

9.3 Internal linking as a retrieval signal

Internal links tell engines how your pages rank against each other in importance. They also help with retrieval, because a well-linked page is easier to surface inside a multi-step query.

We hold to two rules. Every pillar page should be linked from at least five places. Every cluster page should link up to its pillar and across to at least two siblings. What you want is a graph, not a tidy tree.

9.4 Comparison and “vs.” pages that win citations

On shortlist queries, comparison pages punch well above their weight. A well-built “X vs. Y” page is one of the most frequently cited formats in commerce and software.

Build it with care. Neutral framing, real data, honest pros and cons, current pricing. The “we are obviously better” version gets demoted. The genuinely useful, well-structured one gets quoted.

9.5 The case for fewer, deeper pages

If I could give a brand starting fresh today a single instruction, it would be this: write fewer pages, and make each one the best resource on its specific topic. Engines reward depth and punish thinness harder every quarter. Twenty exceptional pages will beat two hundred mediocre ones, and it is not close.

Chapter 10: Measurement: how to prove AEO and GEO works

10.1 Tools worth paying for

The AEO measurement market is young and consolidating. The players we use or watch include Profound, Otterly, AthenaHQ, Peec AI, and SE Ranking, among others, and each has its strengths and gaps.

Do not over-invest in tooling before you have a measurement habit, though. A prompt panel in a spreadsheet that you actually run every week will teach you more than an expensive platform you open once a quarter.

10.2 Building a prompt panel: 50 prompts that matter

A prompt panel is a fixed set of questions you run against the engines on a schedule. Built well, it is the single most useful artifact your AEO team will own.

Start with 50 prompts across four buckets:

1. Head category queries. “Best email marketing agency for ecommerce.”
2. Comparison queries. “Klaviyo vs. Omnisend for DTC brands.”
3. Branded queries. “Is SOLID a good agency for cross-border growth.”
4. Long-tail use cases. “How do I scale Meta Ads after the iOS 14 changes.”

Run them weekly, log the answers, tag who gets cited. The patterns show up inside a month.

10.3 Tracking citation rate weekly without burning hours

Several tools now automate citation-rate tracking across the major engines, including the ones above. They run your panel on a schedule, log who is cited, and surface the trends for you. Once the panel stabilizes, paying for one is usually the right call.

If you are still doing it by hand, automate the prompt execution wherever you can. The log itself can be a simple sheet: prompt, engine, your brand cited yes or no, competitors cited, notes.

Two hours a week from one analyst is enough to keep a 50-prompt panel running across four engines without tooling. Less than that and the discipline slips. More than that and you should be paying for a tool instead.

10.4 Connecting AEO visibility to revenue

This is the hardest part, and where most teams give up. The honest method is to triangulate.

• Track AI-referral traffic by engine. It is growing.
• Watch branded direct traffic and branded search after a content launch.
• Ask new customers in onboarding where they first heard of you.
• Run six-week holdout tests on specific content investments.

No single signal proves anything on its own. Put together, they form a picture you can act on.

Chapter 11: Industry playbooks

The principles hold across the board. How you apply them shifts by category.

11.1 Ecommerce and DTC brands

Product pages and category pages are your two highest-leverage surfaces, and AI engines cite product pages heavily on shopping queries. Mark them up with structured data, write clear descriptions, surface real reviews, and keep pricing accurate.

Comparison content (“X vs. Y for Z use case”) pulls more than its share of citations on shortlist queries, and so do honest “best X for Y” pages.

For DTC the integration with the rest of the stack is what matters most. AI visibility feeds retargeting pools, email capture, and warm acquisition, so we run it as part of the broader ecommerce growth program rather than a side project.

11.2 SaaS and B2B software

B2B SaaS buyers lean on AI engines to build their shortlist. The pattern that wins is a strong category presence, a clean comparison hub, and a steady cadence of expert content from people on your team.

Founder-led content is unusually effective here, because models attach claims to named experts faster than to a faceless brand. Get your founder and product leaders onto podcasts, into op-eds, and writing substantive long-form posts on LinkedIn.

11.3 Local and service businesses

Local intent still anchors to Google Business Profile, but the answer engines are catching up fast. Your Business Profile, your service-area pages, and your reviews now feed both the local pack and the AI answer.

For multi-location brands, consistency across location pages is the single biggest factor. Inconsistent NAP, hours, and service descriptions fracture the entity and weaken every location at once.

11.4 Marketplaces and aggregators

Marketplaces carry a built-in tension. The user-generated content that makes them valuable is also unstructured, inconsistent, and hard for an engine to parse. The ones that win build a structured layer on top: curated guides, comparison hubs, and editorial content that gives the engine something clean to cite.

11.5 Regulated industries

For finance, health, and legal brands, EEAT signals (Experience, Expertise, Authoritativeness, Trustworthiness) matter more than anywhere else. Author bios with real credentials, citations to primary sources, and visible review by qualified experts are requirements, not nice-to-haves.

Engines stay cautious on YMYL queries (Your Money or Your Life). They cite established, credentialed sources and demote anonymous or shallow content faster than in any other category.

Chapter 12: The AEO and GEO rollout plan

If you are starting from scratch, this is the sequence we run with clients.

12.1 Phase 1: audit and baseline

• Build the 50-prompt panel and run it once across four engines.
• Audit the top 20 pages for structure, freshness, and entity clarity.
• Audit schema markup, robots.txt, and crawler access.
• Inventory your entity definitions across the top 10 external surfaces (LinkedIn, Crunchbase, Wikipedia if present, and so on).
• Stand up the baseline dashboard.

12.2 Phase 2: foundational fixes

• Fix schema markup on the top 20 pages.
• Standardize entity definitions across site, schema, and external profiles.
• Repair canonicals, sitemaps, and crawler access.
• Add or rewrite “in a nutshell” summaries on top-traffic pages.
• Make a deliberate decision on AI crawler access in robots.txt.

12.3 Phase 3: content rebuilds on top 20 pages

• Rewrite or restructure the top 20 pages for extractable chunks, question-first structure, and FAQ blocks.
• Build or rebuild your most important pillar pages.
• Refresh dated language and timestamps where they need it.
• Tighten internal linking across the top 50 URLs.

12.4 Phase 4: off-site GEO push and measurement loop

• Launch the first round of digital PR aimed at AI-visible outlets.
• Pick two or three communities (Reddit, niche forums, Substack) to contribute to over the next quarter.
• Book three to five founder-led podcast appearances.
• Lock in the weekly measurement habit and start reviewing the dashboard with the executive team.

12.5 What to ship every week

A rough weekly cadence we use:

• Monday: prompt panel run, results logged.
• Tuesday: content production or a refresh on one page.
• Wednesday: off-site work (a PR pitch, podcast outreach, a community contribution).
• Thursday: a technical or schema fix.
• Friday: dashboard review and a short note to leadership.

The compounding only happens if the rhythm holds. The weeks you skip are the weeks your competitors do not.

Chapter 13: Common mistakes and how to avoid them

13.1 Over-optimizing for one engine

We watch teams fixate on ChatGPT citations and ignore Perplexity, or chase Google AI Overviews while quietly losing ground in Copilot. The engines diverge. Optimize for the principles underneath them, not for whichever surface you happened to check last.

13.2 Stuffing schema or faking FAQs

Schema works when it describes the page accurately. It backfires when a team pads FAQ schema with questions that are not on the page, or invents reviews that never happened. Engines catch this faster than they used to, and the demotion is sharp.

13.3 Ignoring crawler access entirely

The most common mistake we find is a site that has never thought about AI crawlers at all: default robots.txt, no schema, no entity consistency. These brands are not blocked. They are simply invisible, and fixing the technical layer is often the single highest-ROI move available to them.

13.4 Mistaking traffic loss for failure

Plenty of brands are losing organic clicks while gaining AI-driven influence at the same time: clicks down, brand mentions up, branded direct traffic up. Measure clicks alone and you will misread the situation and over-correct in exactly the wrong direction.

13.5 Hiring the wrong agency or tooling

The agency market is full of teams that relabeled their old SEO service as “AEO” without changing the method. The tooling market is full of dashboards that count things without telling you what to do about them.

Ask any potential partner three questions. How do you measure citation rate today? What does your weekly operating loop look like? What have you changed in your method in the last six months? The honest ones answer with specifics. The rest deflect.

If you want to see how we run this in practice, our SEO, GEO, and AEO agency engagements are built around exactly the loop this playbook describes.

Chapter 14: The next 24 months

I will not pretend to know exactly what happens next. But a few directions already have enough momentum to plan around.

14.1 Agentic search and the rise of buying agents

AI agents that browse, evaluate, and buy on a person’s behalf are arriving faster than most brands expect. When the buyer is an agent, the priorities shift again. Clean product data, machine-readable pricing, structured guarantees, and verifiable reviews stop being nice-to-haves and become the requirements.

14.2 Agentic commerce: the new shopping surfaces

Shopping is moving inside the AI conversation. The developments worth watching right now:

• Shopify Agentic Storefronts launched in the Winter ‘26 Edition and began activating by default for Shopify stores in late March 2026. Merchants can sell directly inside ChatGPT, Perplexity, and Microsoft Copilot conversations, with more surfaces rolling out.
• The Universal Commerce Protocol, built by Shopify with Google, is an open standard for bringing commerce to AI agents. Native shopping is rolling out across Google AI Mode and the Gemini app.
• ChatGPT checkout, Perplexity Shopping, and Microsoft Copilot’s commerce features are each maturing in parallel. Every one of them is a new shelf your brand either sits on or does not.

The strategic point is the same whichever platform leads. Once a chat surface can complete a purchase end to end, the funnel collapses, and the brands that are easy to discover, recommend, verify, and transact with take a disproportionate share. That means clean product data, accurate inventory, structured returns and shipping information, and a presence on whichever commerce protocol your platform supports.

For ecommerce brands this is no longer a 2027 problem. The shelves exist now.

14.3 Multimodal answers

Voice answers, video answers, and visual citations are all growing. Optimizing for text alone is no longer enough. Video transcripts, podcast transcripts, and image alt text are becoming first-class content rather than afterthoughts.

14.4 The likely consolidation of measurement tools

The crowded AEO tooling market will thin out. Expect two or three category winners inside 24 months, and expect the survivors to integrate with the traditional SEO platforms rather than try to replace them.

14.5 What to future-proof now

Three things will hold their value through whatever comes next.

1. A strong, consistent entity across every surface your brand appears on.
2. A library of genuinely useful, extractable content that does not depend on any single platform’s algorithm.
3. An operating loop with measurement, learning, and fast iteration. The loop is what survives the rule changes.

Build for those three and the rest is tactics, and tactics rotate.

Want help running this loop for your brand? We will look at your current visibility, build the prompt panel, and tell you candidly where the leverage is.

Get your free growth plan →

Frequently asked questions

What is the difference between AEO and GEO?

AEO (Answer Engine Optimization) is mostly on-site work: structuring content so an AI engine can extract it as an answer. GEO (Generative Engine Optimization) is broader. It includes AEO and adds off-site presence, entity authority, and being remembered by generative models even when they are not running a live search.

Is SEO dead?

No. SEO is the foundation underneath AEO and GEO. Most AI engines reuse classical search infrastructure to retrieve and rank candidate sources, so strong SEO is a prerequisite rather than a competing priority.

How long does it take to see AEO results?

Foundational fixes (schema, structure, freshness) can show citation-rate improvement within four to eight weeks. Entity building and off-site GEO compound over six to twelve months. Anyone promising overnight results is overpromising.

Should I block AI crawlers from my site?

For most growth-stage brands, no. Blocking opts you out of both training data and live retrieval. The exceptions are publishers with paywalled or proprietary content, where the trade-off is deliberate and documented.

What is the single most important thing to do first?

Build a prompt panel, run it once, and look at the results. Most teams find they are less visible than they assumed, or that a competitor they were not watching is dominating. The panel forces an honest baseline, and everything else follows from it.

How is this different from the AI engine optimization guide?

The guide is the on-ramp, a short primer on why AI search matters and how to start adapting. This playbook is the operator’s manual: the deeper framework, the metrics, the rollout phases, and the ongoing loop we run with clients. Read the guide first if the topic is new to you, then come back here to operate.

Can I just use a tool to handle this?

No tool currently handles AEO and GEO end to end. Tools help with measurement and with surfacing patterns. The strategy, the content, the entity work, and the off-site GEO still take human judgment and craft. The brands winning right now treat tooling as one layer inside a broader operating system rather than a substitute for it.