Answer Engine Optimization: The Playbook for Getting Cited by AI

Answer Engine Optimization (AEO) is the discipline of making your brand the source an AI system can trust when it answers a buyer question. If someone asks ChatGPT, Perplexity, Claude, Gemini, or Google AI Overviews who to hire, what tool to use, or which local pro to call, AEO is why your name appears — or why a competitor’s does.

This playbook is the operating manual Spurlock Studios uses on visibility engagements. It covers definition, why rank-only SEO fails the new surface, the five-layer method (entities, machine-readable facts, citeable passages, earned corroboration, measurement), crawler policy, common failures, and a 90-day implementation sequence. Spoke posts under this pillar go deeper on each tactic. Start here for the system.

The short answer

• Search engines return ranked lists. Answer engines return synthesized answers that may cite zero, one, or several sources.
• You win when a model can retrieve you, compress you into a sentence, and defend that sentence against contradictory pages.
• Five layers: entity architecture, on-domain facts machines can extract, passages that stand alone, off-site agreement, and a prompt panel you actually log.
• Training opt-out and citation eligibility are separate robots.txt decisions. Treat them as separate.
• Run 90 days with weekly measurement. One publish sprint without a baseline is a hope, not a program.

What is Answer Engine Optimization?

AEO is not a rebrand of SEO. Search engines return ranked lists of pages. Answer engines return synthesized answers. Your job shifts from “rank page 1 for keyword X” to “be the entity and the passage the model can defend when it writes a sentence.”

Three surfaces matter for most B2B and local brands in 2026:

1. Retrieval-augmented chat — ChatGPT with search, Perplexity, Bing Copilot, Claude with web tools. These systems fetch live pages, compress them, and cite.
2. AI Overviews and similar SERP answers — Google injects a generated block above organic results when it decides the overview is additive. Citations there drive brand memory even when the click never happens.
3. Model memory and training residue — Older facts about your company that persist even when the site has changed. Wrong NAP, dead product names, and competitor comparisons that never update.

AEO work attacks all three. You publish machine-readable truth, make that truth easy to retrieve, earn corroboration off-site, and measure whether models actually use you.

AI visibility is the scoreboard for this work: inclusion in the answer, not position in a list.

AEO vs SEO vs GEO

Operators hear three acronyms and assume they compete. They do not.

Discipline	Primary unit of winning	What you optimize
SEO	Document rank for a query	Crawlability, links, relevance, UX
AEO	Citation / inclusion in an answer	Entities, passages, schema, corroboration
GEO (Generative Engine Optimization)	Same family as AEO; often used for generative SERPs and chat	Citeability inside generated text

In practice Spurlock Studios treats GEO as a subset of AEO focused on generative engines, and keeps SEO as the foundation that still feeds crawl and authority. You still need indexable pages. You also need pages that survive summarization.

For the acronym split without stealing this playbook’s method, see GEO explained. For the scoreboard change (rank/CTR vs citation/accuracy), see AEO vs SEO.

Why AEO is a revenue problem, not a vanity metric

Buyers already ask AI before they ask Google, or they ask Google and get an Overview that never clicks through. That does not kill websites. It changes the job of the website from “win the click” to “win the citation and still convert the people who dig deeper.”

Pew Research Center’s July 2025 analysis of March 2025 browsing — 900 U.S. adults — found Google users clicked a traditional result on 8% of visits when an AI summary appeared, versus 15% when it did not. Clicks on links inside the summary itself happened on 1% of those visits. About 18% of the Google searches in the study produced an AI summary; 88% of those summaries cited three or more sources (Pew Research Center).

Read that the operator way: most Overview traffic will not show up as a session. Being unnamed in the answer is a consideration-set loss, not a “we’ll catch them in organic” delay.

Google’s own site-owner guidance says AI Overviews and AI Mode may use query fan-out — multiple related searches across subtopics — and that eligibility for a supporting link requires the page to be indexed and snippet-eligible. There are no extra technical requirements beyond ordinary Search eligibility (Google Search Central: AI features). Google also claims clicks from pages with AI Overviews tend to be higher quality (more time on site). Pew and Google are measuring different things. Do not pick one and ignore the other.

Concrete failure modes we see in audits:

• A mid-market SaaS ranks for its category keywords but ChatGPT recommends three competitors because those competitors have clearer About pages, Wikidata IDs, and comparison posts with tables.
• A local HVAC company owns the map pack and still loses “best HVAC near me for heat pumps” style prompts because directories and city pages never state services, certifications, and service area in plain sentences.
• A founder corrects a wrong founding year on their site; Perplexity still cites an old press release. The model is not stubborn — the corroborating sources are.

If your pipeline includes inbound or high-consideration purchase, AI answers are part of the consideration set whether you measure them or not. Ignoring them is not neutrality. It is conceding the narrative.

• Priority buyer prompts exist as a written list, not a vibe
• Someone has run those prompts in at least two products this month
• You know whether you are named, cited, misstated, or absent
• Revenue-relevant prompts are tagged separately from vanity brand queries

How answer engines decide what to cite

You do not need the model weights. You need a working mental model:

1. Query understanding — Is this a definition, comparison, recommendation, local, or how-to?
2. Retrieval — Which URLs or indexed snippets look relevant and fresh?
3. Compression — Which passages compress into a confident sentence without contradictions?
4. Attribution — Which domains are safe to show as citations (or safe to name without a link)?
5. Safety / policy — Does the answer risk recommending something harmful or outdated?

Your site wins when it is easy to retrieve, easy to compress, and hard to contradict. That is why entity consistency and off-site agreement matter as much as word count.

Surface	What the vendor says it does	What you optimize
Google AI Overviews / AI Mode	Query fan-out; supporting links from snippet-eligible indexed pages; uses Search systems plus Knowledge Graph (Google AI Overviews PDF)	Indexable, people-first pages; extractable passages; consistent Organization facts
Perplexity	Live web search, then a summary with numbered citations you can open (Perplexity Help Center)	Fresh, factual pages; tables; dates; pages that survive a first-screen extract
ChatGPT search	OAI-SearchBot surfaces sites in search answers; opting out removes you from those answers (OpenAI crawler docs)	Allow the search bot; keep facts consistent; earn other domains that agree
Claude web tools	Claude-SearchBot indexes for search quality; Claude-User fetches on a user question (Anthropic Help Center)	Same split: do not confuse training opt-out with search opt-out

Google is explicit that AI Overviews often do not trigger. Absence of an Overview on a keyword is not a citation failure. Absence of your name when an Overview or chat answer does fire on a buyer prompt is.

For Overview-specific tactics, use how to get cited in Google AI Overviews. This pillar stays on the system that feeds every surface.

What “getting cited by ChatGPT” actually requires

People ask how to get cited by ChatGPT as if there were a submission form. There is not. Practical requirements:

• Pages crawlable by the bots and tools that feed browsing and search modes
• Clear, non-contradictory brand facts
• Content that answers the class of question buyers ask (not only keyword variants)
• Enough external mention that retrieval does not only find you as a thin homepage
• Ongoing freshness for claims that change

Paid ads do not buy citations in the chat product. Authority and clarity still do. OpenAI even documents a separate OAI-AdsBot that only visits landing pages submitted as ads and is not used to train foundation models (OpenAI crawler docs). That is ad safety, not a citation shortcut.

Training bots and search bots are not the same decision

This is the most expensive robots.txt mistake in the visibility lane.

Vendor	Training / model-use control	Search / citation crawl	User-triggered fetch
OpenAI	GPTBot — disallow to signal “do not use for foundation-model training”	OAI-SearchBot — disallow and you are not shown in ChatGPT search answers	ChatGPT-User — user-initiated; robots.txt may not apply
Anthropic	ClaudeBot — future training datasets	Claude-SearchBot — search indexing / answer quality	Claude-User — fetch on a user question; Anthropic says it honors robots.txt
Google Search	Googlebot + ordinary Search controls (nosnippet, noindex)	Same crawl feeds AI features in Search	Preview controls documented on the AI-features page
Google Gemini apps / Vertex grounding	Google-Extended product token — does not change Search inclusion or ranking (Google common crawlers)	Separate from Googlebot Search	N/A as a Search ranking lever

OpenAI says the settings are independent: you can allow OAI-SearchBot and disallow GPTBot. robots.txt changes can take about 24 hours to apply for search (OpenAI crawler docs).

Google’s robots.txt file is a crawl-traffic control, not a hide-from-Search switch. A disallowed URL can still be indexed without a snippet if other pages link to it (Google robots.txt intro).

• OAI-SearchBot allowed if ChatGPT search citations are a goal
• Claude-SearchBot and Claude-User allowed if Claude citations are a goal
• GPTBot / ClaudeBot / Google-Extended decided on purpose, not copied from a “block all AI” gist
• CDN / WAF is not silently 403ing those user-agents
• Important answers are in HTML text, not only in client-rendered widgets

A “block every AI bot” policy is a product decision. It is not an AEO strategy.

The five-layer AEO method

Spurlock Studios runs visibility work as five stacked layers. Skip a layer and the stack wobbles. The rest of this playbook is those layers, then the 90-day sequence that ships them.

Layer	Job	Failure if skipped
1. Entity architecture	The brand is a named thing, not a slogan	Model substitutes a better-defined competitor
2. Machine-readable facts	Domain publishes extractable truth	Retrieval finds marketing fog
3. Citeable content	Passages survive 40–80 word compression	You rank; you never get quoted
4. Off-site corroboration	Other domains repeat the same facts	One perfect site loses to a chorus of directories
5. Measurement loops	You see citations, errors, and gaps	You optimize vibes

I have been SEO-certified since 2021 and now spend that same discipline on AEO / GEO surfaces. The crawl work did not get less important. The acceptance test changed.

Layer 1 — Entity architecture

Models name things. If your brand is not a clear thing — Organization, Person, Product, Place — the model hedges or substitutes a better-defined competitor. The spoke that owns this layer is entity architecture.

schema.org Organization is the vocabulary. Google’s Organization structured data docs say markup on the home page can help disambiguate your organization; some properties stay behind the scenes, others can influence knowledge-panel and attribution UI. There are no required properties — add what is relevant and true (Google Organization structured data). Google expanded Organization support in November 2023 (name, address, identifiers, logo) and said it can feed knowledge panels and attribution (Google Search Central Blog).

sameAs is the identity hinge: “URL of a reference Web page that unambiguously indicates the item’s identity” — Wikipedia, Wikidata, official site (schema.org/sameAs). Point it at profiles that are actually you. A wrong LinkedIn or a similarly named company’s Crunchbase row is worse than a short sameAs list.

Minimum entity stack for a brand:

Entity	Minimum proof	Common miss
Organization	Legal name, trade name, canonical URL, logo	Hero slogan instead of a name
sameAs profiles	LinkedIn, GBP, Crunchbase, YouTube — real URLs	Stale or competitor lookalikes
Person (when the pitch is a founder)	Name, role, sameAs	Ghost-written “team” with no identifiers
Product / Service	Specific names, not “Solutions”	Three pillars that mean nothing
Place / area served	City, region, or “remote / US” stated in a sentence	Implied by a map embed only

Wikidata is a structured knowledge base anyone can edit. Use it when the item already exists or notability is honest. Do not spam a promotional stub. If a Q-ID exists, put the official website on the item and put the item in sameAs. That closes a loop. If no item exists, skip Wikipedia theater and spend the week on directories that already rank in retrieval.

• One legal name and one trade name, used the same way on site, GBP, LinkedIn
• Organization JSON-LD on the home or About page, validated
• sameAs URLs resolve and describe this company
• Retired product names listed as retired, not deleted into silence
• Similarly named companies called out in one disambiguation sentence if confusion is real

Layer 2 — Machine-readable facts on your domain

Humans skim. Models extract. Give them extractable facts.

llms.txt is a briefing, not a sitemap

llms.txt is Jeremy Howard’s September 2024 proposal: a Markdown file at /llms.txt (or a more specific subpath) that tells language models what the site is and which URLs settle which questions (Answer.AI proposal). The only required section is an H1 with the project or site name. Optional: a blockquote summary, preamble, then H2 file lists of [name](url) links.

It is a convention, not an IETF or W3C standard. Shipping a 400-URL dump of your sitemap is how you waste the file. Shipping a briefing — who you are, who you serve, who you do not serve, and five to fifteen canonical URLs — is the AEO use.

Deep spec and examples: llms.txt for brands. How to write one that models actually use: llms.txt done properly.

JSON-LD that matches the visible page

Prefer JSON-LD for Organization, WebSite, Article, and LocalBusiness / ProfessionalService as relevant. Google’s AI-features guidance repeats a rule that already applied to Search: structured data must match the visible text (Google AI features).

Validate with the Schema Markup Validator. Accuracy beats volume. Five types on every page, half invalid, is schema soup.

FAQPage is still a real vocabulary for pages that actually contain questions and answers. Do not ship fake FAQ markup to chase a SERP accordion. Write the Q&A because a model can lift a 40-word answer. Markup is secondary.

Details: schema markup for answer engines.

Canonical fact pages

Page	Question it must answer in the first screen	Ages?
About	Who are you, since when, where, for whom	Medium
Services / offer	What you sell, what you refuse, how you engage	High
Pricing or packages	Numbers or a clear “contact” posture	High
Locations / service area	Where you work, in a sentence	Medium
Team	Named people when they are part of the pitch	Low
Changelog or “updates”	What changed and when	High

Dates on claims that age (pricing, product names, certifications) are not decoration. Stale facts become hallucination fuel.

• /llms.txt returns 200, starts with # Brand Name, links to answer pages
• Organization JSON-LD matches the About page word for word on name, URL, founding year
• Pricing posture is identical on the site, in schema, and in sales decks
• One owner can approve a fact change in a day

Layer 3 — Citeable content architecture

Answer engines prefer passages they can quote without rewriting half the paragraph. Structure content so a 40–80 word block still makes sense alone.

Google’s inclusion guidance for AI features is boring on purpose: allow crawling, make content findable with internal links, put important content in textual form, match structured data to visible text, keep Business Profile / Merchant Center current (Google AI features). The same page points at ordinary helpful, people-first content. There is no secret AEO schema type.

Patterns that win citations:

• Direct answer in the first two paragraphs
• Definition boxes and comparison tables
• Numbered methods and checklists
• FAQ sections with real questions
• Original data, screenshots of method, or named case outcomes — not adjective stacks

Build this as clusters, not random posts. Pillar + spokes that cover a question family outperform isolated “thought leadership.” This page is the cluster hub. The spokes are the question family. Write so the first breath is the answer — answer-first pages — and group them as content clusters for AI visibility.

Format	When it compresses well	When it fails
Definition	“What is X?” in two paragraphs, then depth	1,200 words of throat-clearing first
Comparison	Table with explicit criteria and a “who should pick which”	Feature laundry lists with no loser
Playbook	Numbered steps, prerequisites, failure modes	Motivational essays
Checklist	Auditable items a practitioner can run today	Vague “be authentic” items
Local service page	City + service + proof + NAP	Doorway spam
Small original research	Even a narrow anonymized benchmark	Recycled tips with no source

Avoid the opposite formats when citation is the goal: pure opinion with no extractable claims, infinite-scroll listicles without sources, and “ultimate guides” that bury the answer.

When you brief writers, specify format, not vibes. “Write a comparison with a six-row table and a one-sentence recommendation per row” is a brief. “Make it authoritative” is not.

Layer 4 — Corroboration off-site

One perfect site is not enough when retrieval samples the open web. Models look for agreement across sources. Perplexity’s own help text says it searches the live web and attaches numbered citations so a reader can verify (Perplexity Help Center). If the live web disagrees with you, the answer will too.

Sources that move the needle for most brands:

Source class	Why retrieval likes it	Do not
Niche publications with editors	Independent sentence, same facts	Pay-for-play dumps with wrong NAP
Industry directories / associations	Repeated name, URL, category	Ten spam listings with three legal names
Podcast transcripts / event pages	Spoken name in a dated artifact	“As seen on” graphics with no URL
Partner and customer case pages	Third-party domain, first-party story	Quotes you wrote for them that they never published
Wikidata / Wikipedia	Structured identity	Promotional stubs that get deleted

Tactics live in the spoke cluster. The playbook rule is simpler: the same three sentences — who you are, who you serve, what you sell — should appear on your site and on the pages retrieval already trusts. If those sentences conflict, fix the conflict before you pitch another article.

Citation-gap work is competitive: list the URLs that appear when you are absent, then decide whether you can earn a similar page or out-publish a clearer owned page. Do not “out-PR” a hallucination with more adjectives.

Layer 5 — Measurement that survives non-determinism

If you cannot see whether AI cites you, you are optimizing vibes. AI answers are non-deterministic. The same prompt can cite different sources on different days. Design measurement accordingly.

Google reports AI Overviews and AI Mode inside Search Console’s overall Web performance — not as a magic “AEO” tab (Google AI features). That is useful for traffic. It is not a citation log. You still need a prompt panel.

Core KPIs

KPI	How to capture	Cadence
Citation rate	% of prompt-panel runs that link you	Weekly
Mention rate	% that name you with or without a link	Weekly
Share of voice vs named competitors	Same panel, competitor set fixed	Weekly
Position in answer	Named first / mid / only in an “also” list	Weekly
Fact accuracy	Wrong claims about you (yes/no + severity)	Biweekly
Referral traffic from AI hosts	Analytics referrers + UTM where available	Monthly
Overview presence	Manual / tool checks on priority SERPs	Weekly

Mention without citation still moves a shortlist. Citation without a click still moves a shortlist. Pew’s 1% in-summary click rate is why you cannot run this program on sessions alone.

Building a prompt panel

Start with 25–40 prompts, not 400. Buckets:

Bucket	Example shape	Why it exists
Category definition	“What is X?”	Tests whether you own the term
Vendor recommendation	“Best X for Y”	Tests whether you are in the shortlist
Comparison	“A vs B”	Tests tables and honesty
Local	“X near [city]”	Tests NAP and service-area sentences
Brand	“Who is [Company]?” / “Is [Company] legit?”	Tests the fact packet
Objection	“How much does X cost?”	Tests pricing posture

Run them in ChatGPT, Perplexity, and one Google AI Overview sample per week. Log: date, model/product, cited URLs, whether you appear, whether facts are correct.

Date	Product	Prompt ID	Named?	Cited URL	Fact error	Competitor named first
YYYY-MM-DD	ChatGPT search	rec-04
YYYY-MM-DD	Perplexity	rec-04
YYYY-MM-DD	AI Overview	rec-04

Semrush and similar suites help with SERP/Overview monitoring and competitive URL discovery. They do not replace the chat prompt panel. Surfer-style content scoring helps page structure for the human/SERP layer. It is not a citation score.

Full audit mechanics: the AEO audit checklist.

• Panel size is 25–40, frozen for a quarter
• Each prompt has an owner and a “revenue / brand / research” tag
• Logs live in a sheet or base, not in screenshots
• A wrong founding year is an incident, not a shrug

Common AEO failures (and the fix)

Failure: Treating llms.txt as a sitemap dump

Symptom: File exists; answers still ignore you.
Fix: Rewrite as a briefing with entities, services, and deep links to answer pages. See the llms.txt spoke.

Failure: Schema soup

Symptom: Every page has five types, half invalid.
Fix: Ship accurate Organization + page-type schema. Validate. Remove vanity markup.

Failure: Blog volume without question coverage

Symptom: 80 posts, zero comparison or definition pages for the category.
Fix: Map the question cluster, write the missing answer pages, prune or redirect fluff.

Failure: One site, zero corroboration

Symptom: Site is clear; AI still cites directories and competitors.
Fix: Digital PR, partner pages, listings — then re-measure citation gaps.

Failure: Blocking the search bot you wanted to impress

Symptom: “We blocked AI crawlers for safety”; ChatGPT search never cites you.
Fix: Split training vs search. Allow OAI-SearchBot / Claude-SearchBot if citation is the goal. Confirm the WAF.

Failure: Correcting the site but not the sources of the lie

Symptom: Hallucinated founding year / HQ / product persists.
Fix: Find the corroborating wrong sources, update or outcompete them, strengthen canonical facts.

Failure: Measuring only organic rank

Symptom: Rankings up, AI share of voice flat.
Fix: Add the prompt panel. Treat Overview and chat as first-class surfaces.

Failure: Ignoring local pack vs AI local answers

Symptom: Strong Maps presence, weak chat recommendations.
Fix: Service-area pages with plain-language proof, reviews that mention services, consistent NAP.

Failure	First diagnostic	First ship
Invisible in chat	robots.txt + WAF for search bots	Allow search bots; resubmit key URLs
Named wrong	Prompt-panel fact log	Fact packet + source cleanup
Never named	Competitor citation URL list	One cluster + two corroborating pages
Overview-absent on a keyword	Confirm an Overview even fires	Do not treat a no-Overview SERP as a loss

90-day implementation roadmap

This is the sequence we actually run. Do not invert it. PR before a fact packet amplifies the lie. Content before a baseline means you cannot tell if citations moved.

Days 1–14: Audit and baseline

• Run the AEO audit checklist
• Build the prompt panel and capture baseline citations
• Inventory entity consistency across the top 10 profiles
• Crawl for conflicting facts (founding year, HQ, product names)
• Identify the top 10 competitive citation URLs
• Dump robots.txt and WAF rules for the bots in the table above

Deliverable: baseline report with gaps prioritized by revenue-relevant prompts.

Audit output	Pass looks like
Prompt panel v1	25–40 prompts, three products, one week of logs
Entity sheet	Name / URL / founding / HQ / products, with conflicts highlighted
robots.txt decision	Written, not inherited from a template
Competitor citation list	Real URLs from real runs, not a guessed “they have more DR”

Days 15–35: On-site truth layer

• Ship or rewrite llms.txt
• Fix Organization / LocalBusiness JSON-LD and sameAs
• Rebuild About, Services, and primary offer pages for extractability
• Add FAQ blocks only where questions are real
• Align NAP and service-area language
• Put dates on anything that can rot

Deliverable: machine-readable brand packet live on the domain.

Days 36–60: Citeable content sprint

• Choose one pillar topic (this playbook’s pattern) and 6–12 spoke questions
• Write definition, comparison, and how-to pages with answer-first structure
• Add tables, steps, and original proof where you have it
• Internal-link the cluster; update the sitemap
• Point llms.txt at the new cluster, not only at Home

Deliverable: one complete question cluster live.

Days 61–90: Corroboration and loops

• Pitch or place 3–8 digital PR / niche mentions with correct facts
• Close citation gaps against the competitor URL list
• Re-run the prompt panel; document deltas
• Open a monthly hallucination / fact-drift review
• Decide: continue content, deepen local, or expand entities (products, people)

Deliverable: measured lift on citation rate for priority prompts, or a clear next experiment.

Local or multi-location brands should parallelize GBP hygiene and city pages in days 15–60 rather than waiting for the content sprint to finish.

Window	If you only have one engineer-week	If you only have one writer-week
1–14	robots.txt, schema, crawl conflicts	Prompt panel + competitor URL harvest
15–35	llms.txt + JSON-LD + About	Offer page rewrite, first screen
36–60	Internal links + sitemap	One comparison + one definition
61–90	Profile cleanup on the top 10 listings	Two pitch emails with the fact packet attached

A 90-day program with no week-12 panel is an article series. Keep the log.

Operating cadence after the first quarter

AEO is not a one-time project. Minimum ongoing rhythm:

Cadence	Work	Done when
Weekly	10–20 prompt-panel runs; log citations	Sheet has dates, not vibes
Monthly	Fact audit on About / pricing / product; refresh stale claims	lastModified or changelog moved
Quarterly	Cluster refresh against new buyer questions; PR burst	One new spoke or one retired spoke
Anytime a launch or rebrand ships	Truth layer first, content second, PR third	llms.txt and schema updated the same day

Spurlock Studios visibility retainers are built around that cadence plus the audit offer for teams that want a sharp baseline before they commit to build.

Tooling notes (honest)

Tools help; none of them are the strategy.

Tool	What it is good for	What it is not
Semrush	Competitive URL discovery, keyword → question mapping, Overview/SERP monitoring where available	A ChatGPT citation score
Surfer (or similar)	On-page structure and topical coverage for the human/SERP layer	Proof you will be cited
Manual prompt panels	Ground truth for chat citations	Scalable without a human or a script you trust
Schema Markup Validator	Catch broken JSON-LD	A ranking lever
Search Console	Traffic and indexation, including AI-feature clicks rolled into Web	Mention/citation rate
Crawl tools	Orphan pages, conflicting titles, blocked JS-only answers	Entity truth

We disclose Semrush and Surfer when they appear in client workflows because they influence recommendations. They do not generate citations by themselves.

Worked example: a category recommendation prompt

Imagine a buyer asks Perplexity: “Best fractional AI automation partner for a 40-person e-commerce brand.”

A weak brand presence looks like this in retrieval:

• Homepage hero: “We reinvent growth with AI”
• Services page: three vague pillars, no ICP, no proof
• No comparison or “who we serve” page
• Directory listings with an old company description
• robots.txt copied from a “block GPTBot” blog that also blocked OAI-SearchBot

A citeable presence looks like this:

• Opening paragraph on the offer page names ICP, engagement model, and exclusions
• Case section with measurable outcomes you actually have (hours saved, error rate, cycle time) — no invented logos
• llms.txt points at the offer page, About, and a methodology page
• Two niche articles and a partner case study repeat the same ICP sentence
• Organization schema sameAs ties LinkedIn and Crunchbase
• Search bots allowed; training bots decided in writing

The model does not “prefer” you emotionally. It finds a compressible, corroborated story. Build that story on purpose.

I will not attach fake lift percentages to this example. The acceptance test is the week-12 panel versus the week-1 panel on the same prompts.

Governance: who owns brand truth

AEO fails when marketing ships copy that contradicts legal, product, or sales. Assign an owner for the canonical fact packet:

• Legal name, trade name, and “also known as”
• Founding year and HQ
• Product and package names (and retired names)
• Pricing posture (published numbers vs “contact us”)
• Certifications and partnership badges
• Service area and industries served / not served

That owner approves llms.txt, Organization schema, and About. PR and sales enablement reuse the same sentences. Drift is how hallucinations start.

Activity	Owner	Consulted
Fact packet	Marketing ops or founder	Legal, product
Schema / llms.txt	Web eng + marketing	SEO lead
Cluster content	Content lead	Sales (real questions)
Digital PR	PR / founder	Marketing ops (facts)
Prompt panel	SEO / growth	Demand gen
Hallucination incidents	Marketing ops	Support, legal

Keep it small. AEO dies when “everyone owns it” and nobody runs the weekly panel.

Budget framing for a first quarter

Rough allocation that works for many mid-market teams — a planning split, not a benchmark study:

• 30% truth layer (pages, schema, llms.txt, profile cleanup)
• 40% citeable content cluster
• 20% corroboration / digital PR
• 10% measurement and iteration

Underfunding measurement is how you publish a cluster and never know if citations moved. Underfunding the truth layer is how you amplify wrong facts with PR.

Buyer questions that should trigger this playbook

If your team hears any of these, the playbook applies:

• “ChatGPT recommended a competitor — why not us?”
• “Perplexity’s description of our product is wrong.”
• “We rank well but AI Overviews never include us.”
• “We’re relaunching / renaming a product — will AI keep using the old name?”
• “We expanded into a new city — Maps looks fine, chat doesn’t.”
• “Legal wants every AI bot blocked. Sales wants to be recommended.”

Those are not vanity concerns. They are narrative-control problems with pipeline consequences. The last one is a policy decision: write the robots.txt matrix, then pick. Do not let a default gist decide it.

How Spurlock Studios runs visibility work

The visibility lane is built for founders, marketers, and operators who need AI systems to describe them accurately and cite them when buyers ask. Typical engagement path:

1. Audit — baseline prompt panel, entity and schema review, citation gaps, prioritized roadmap (/visibility)
2. Build — truth layer + cluster content + corroboration plan
3. Operate — measurement loops and iterative content/PR

I have shipped hundreds of production sites and 500+ automations. The visibility work is the same operator habit: define the acceptance test, instrument it, then change one layer at a time. We do not sell a “get cited by Friday” switch.

If you want the full system applied to your domain, start with a visibility audit. If you only need one tactic, use the spoke posts linked throughout this playbook and come back when the stack needs to connect.

Playbook summary (one screen)

1. Define the brand as an entity with consistent facts.
2. Publish machine-readable truth (llms.txt, schema, canonical pages).
3. Write answer-first content in clusters tied to buyer questions.
4. Earn corroboration so retrieval finds agreement, not a lone homepage.
5. Measure citations with a prompt panel; fix hallucinations at the source.
6. Split training-bot and search-bot policy on purpose.
7. Run a 90-day roadmap, then a weekly/monthly operating cadence.

That is Answer Engine Optimization as practiced at Spurlock Studios — not a buzzword, a shippable system.

What to do when the week-12 panel is flat

A flat citation rate is information. Do not “write more blog posts” as the default next move. Read the log the way you would read a failing eval set.

What the panel shows	Likely broken layer	Next ship
Never named, never cited	Retrieval / crawl or entity	robots.txt + WAF + Organization + About first screen
Named, never linked	Citeability	One table-heavy comparison or definition page
Linked on brand prompts only	Cluster coverage	Recommendation and comparison spokes
Linked, facts wrong	Corroboration	Hunt the URL that still states the old year
Linked in Perplexity, absent in ChatGPT	Bot policy or freshness	Confirm OAI-SearchBot; recrawl key URLs
Overview never appears on the keyword	Trigger, not citation	Stop scoring that query as a miss

• Week-1 and week-12 runs used the same prompt IDs
• At least two products were logged, not one favorite chatbot
• You can point at a URL that changed between the two runs
• The next experiment touches one layer, not all five

If you cannot name the layer, you are not ready for another content sprint. Re-run the AEO audit checklist before you brief another writer.

Cluster map

This playbook is the hub. Use the spoke that matches the broken layer — not the one that is most fun to write.

Scoreboard and definition

• What AI visibility actually means
• AEO vs SEO
• GEO without the buzzwords
• Mentions vs citations
• Measuring AI search visibility

On-site truth

• Entity architecture
• llms.txt for brands
• llms.txt done properly
• Schema markup answer engines use
• AI crawlers and robots.txt
• When AI gets your brand wrong

Pages that get quoted

• Answer-first pages
• Content clusters for AI visibility
• Get cited in AI Overviews
• Citation gaps
• The AEO audit checklist

FAQ

What is Answer Engine Optimization?

Answer Engine Optimization is the practice of making your brand and pages easy for AI systems to retrieve, trust, and cite when they generate answers. It includes entity clarity, machine-readable facts, citeable content, off-site corroboration, and measurement of citations across ChatGPT, Perplexity, AI Overviews, and similar products. The unit of winning is inclusion in the answer, not a blue-link rank.

How is AEO different from SEO?

SEO optimizes for ranked documents in a results list. AEO optimizes for inclusion and accurate representation inside generated answers. You still need crawlable, authoritative pages (SEO), but you also need extractable facts and corroboration so a model can name you without inventing details. AEO vs SEO is the scoreboard spoke; this page is the method.

How do I get cited by ChatGPT?

There is no submission portal. Allow OAI-SearchBot if you want ChatGPT search answers to include you, publish clear crawlable pages that answer the questions people ask, mark up your organization accurately, earn mentions on other trustworthy sites, and keep facts consistent everywhere. Then measure with a prompt panel and close gaps where competitors are cited instead. OpenAI documents the search bot and the training bot as independent controls.

Does AEO replace SEO?

No. Weak technical SEO and thin pages still lose. Google’s AI-features documentation says eligibility for supporting links is ordinary Search eligibility — indexed, snippet-eligible, no extra technical bar. AEO extends SEO into generative surfaces. Treat them as a stack: crawl and authority first, then citeability and entity truth.

What is the fastest win for most brands?

Fix contradictory brand facts, ship a real llms.txt briefing, strengthen Organization schema with honest sameAs, rewrite the About and primary service pages so the first paragraphs answer “who / what / for whom,” and check that you did not block the search crawlers you care about. Then baseline citations so you know if it moved.

How long until we see citation changes?

On-site clarity can show up in browsing-mode and live-search answers within days to a few weeks — OpenAI says robots.txt updates for search can apply in about a day. Model memory and training residue can lag for months. Plan for a 90-day program with weekly measurement, not an overnight switch.

CTA

If you want this playbook applied to your domain — baseline citations, entity and schema gaps, content priorities, and a 90-day plan — book a visibility audit or review the visibility lane.