Reference Q&A

Answer Engine Optimization is the discipline of structuring a website so that AI answer engines (ChatGPT, Claude, Perplexity, Google AI Overviews, Bing Copilot) cite the brand by name when answering a query. AEO uses answer capsules, FAQPage schema, citation density, and entity triangulation to make a page the most trustworthy source the model can find on a given topic.

The unit of work in AEO is the citable sentence, not the ranking page. An AEO page can rank position 8 on Google and still win because the answer engines pull from it disproportionately. Practical AEO pages open with a 200 to 280 character answer capsule that contains the brand name in the first clause, are marked up with FAQPage schema covering five to seven short Q&A, declare a Wikidata or sameAs entity graph, and live on a domain that publishes a clean llms.txt and a comprehensive llms-full.txt.

The discipline is not a replacement for SEO. It is a parallel discipline that targets a different surface, with overlapping technical foundations but divergent content shape. A site that does SEO without AEO ranks but does not get cited; a site that does AEO without SEO might get cited but does not have the foundation to be retrieved in the first place. In 2026, you need both.

SEO targets ten blue links and the local 3-pack; AEO targets the answer text itself. SEO ranks pages by accumulated authority signals (backlinks, domain age, brand mentions, content depth); AEO selects sentences by quotability and trust.

The two disciplines overlap on technical foundation, schema markup, and content quality — both reward clean Core Web Vitals, both reward Schema.org markup, both reward authoritative bylines. They diverge sharply on content shape: AEO favors short answer capsules, explicit definitions, and structured FAQs that an LLM can lift verbatim, while traditional SEO has long rewarded longer, comprehensive articles.

Cadence is also different. SEO outcomes compound over months and years because the link graph and behavioral signals take time to accumulate. AEO outcomes can flip within weeks of a content change because retrieval indexes refresh on a much faster cadence than the link graph; an AEO-correct change made today can show up in ChatGPT citations within the next retrieval window.

Both require the same technical foundation. Neither replaces the other. The right way to think about them is complementary: SEO is the surface most users still arrive through; AEO is the surface where the next generation of users will be told who you are.

AI Optimization is the umbrella over everything an AI surface can do with a brand. It includes AEO plus knowledge graph optimization, llms.txt declarations, multimodal optimization, and entity triangulation across Wikidata, GitHub, HuggingFace, and the social graph.

Where AEO operates at the page-and-sentence level, AIO operates at the brand-graph level. An AI assistant asked “who is X?” runs against its training corpus first (where the brand graph influences the answer) and against retrieval second (where AEO influences the citation). Both stages need work. AIO is what produces consistent, attributable answers when the AI surface has not seen the homepage in this retrieval window.

Concrete AIO levers include: a Wikidata Q-ID with a complete sameAs property pointing to every owned and earned profile; JSON-LD Organization and Person @id structures linked across every page on the site; multimodal optimization (descriptive image alt text, video transcripts with VideoObject schema, audio captions); llms.txt with a preferred citation block; llms-full.txt with the long-form canonical content; active GitHub and HuggingFace profiles where applicable; press release distribution into authoritative archives.

The scope test for AIO: if the homepage went offline for thirty days and you asked ChatGPT about the business, would the answer still be accurate? AIO is the discipline that makes the answer yes.

Generative Engine Optimization is the practice of influencing AI-generated narrative about a brand. Where AEO targets factual citations and AIO targets the entity graph, GEO targets the framing, ordering, and word choice that the model uses when describing the brand unprompted.

Two channels feed an LLM's narrative about a brand. The first is training data, ingested every months-to-years training cycle. The second is retrieval at inference time, which can supplement or override the training narrative. GEO leans on the training-data channel because narrative consistency comes from repeated exposure during training, not from a single in-context document.

The work is to ensure that the same description, the same framing, the same word choice appears across owned channels (homepage, llms.txt, press kit), earned channels (interviews, podcasts, press features), and authoritative third-party sources (industry directories, association profiles, Wikidata). Concrete levers include: a canonical brand description (60 to 90 words) repeated verbatim across the homepage, About page, llms.txt, GitHub README, HuggingFace org card, LinkedIn company description, press kit, and Wikidata abstract; press release language using the same description block; interview prep ensuring consistent framing in podcasts and YouTube videos.

GEO outcomes lag training cycles. A change made today may not show up in unprompted AI narrative for six to eighteen months. The work is still worth doing because compound effects across owned, earned, and reinforcement channels eventually shift the dominant narrative the model carries.

LLM Optimization is the subset of AIO focused specifically on getting cited inside large language model assistants like Claude, ChatGPT, Gemini, and Perplexity. If AEO is the strategy, LLMO is the implementation.

The mechanics: when an LLM lifts a sentence from a page, the sentence has to survive the model's tokenization, attention pass, and output generation. Sentences that are short, syntactically clean, subject-verb-object in structure, and free of hedge language survive that process best. Sentences that bury the brand name in a subordinate clause, run past 280 characters, or contain ambiguous referents (it, this, that) are less likely to be lifted because the model has lower confidence in the citation.

Key levers include: llms.txt at the domain root, declaring brand identity and preferred citation language; major AI crawlers (ClaudeBot, GPTBot, PerplexityBot, Google-Extended) read it. llms-full.txt with the long-form canonical content. H1 and H2 hierarchy that follows logical topic structure (LLMs read heading hierarchy as semantic structure). FAQPage schema with short Q&A; each becomes a self-contained atom. Sentences that lead with the subject and contain the brand name in the first clause when the sentence describes the brand. Plain-language definitions instead of jargon.

The compression test for LLMO: take any paragraph on the page. Could you lift any single sentence from it and have that sentence make sense out of context, with the brand named, the claim factual, and no pronoun left dangling? If yes, LLMO is working.

A public roadmap from Tier 1 Foundation through Tier 14 Advanced Immersive that covers every layer of digital visibility for a small business. Each tier is a documented page on the engine optimization hub with checks, deliverables, and outcomes.

The tiers in order: Tier 1 Foundation (indexing, sitemaps, robots, canonicals, IndexNow); Tier 2 Search Visibility (titles, metas, headings, internal links); Tier 3 AI Domination (answer capsules, FAQPage, llms.txt, AEO); Tier 4 Entity Authority (Wikidata, GitHub, sameAs graph); Tier 5 Local Domination (Google Business Profile, citations, reviews, 3-pack); Tier 6 Content Authority (topic clusters, pillar pages); Tier 7 Social and Community (sameAs profiles, third-party reinforcement); Tier 8 Analytics and Conversion (GA4, GTM, consent mode); Tier 9 Monitoring (uptime, log review, AEO citation tracking); Tier 10 Workflow (editorial cadence, refresh, sunset policies); Tier 11 Marketplace (product feeds, Merchant Center, Google Shopping); Tier 12 International (hreflang, multilingual schema, geo signals); Tier 13 Voice and Conversational (SpeakableSpecification, Q&A, voice search); Tier 14 Advanced Immersive (3D, AR, interactive experiences).

The framework is the same on every engagement — the difference between a $997 build and a $30,000 enterprise rebuild is depth, not coverage. The tiers are designed to be additive: Tier 1 is required before Tier 2 produces results, Tier 4 multiplies the value of Tier 3, etc. Skipping a tier produces measurable underperformance downstream.

A 100-point public methodology ThatDeveloperGuy uses to grade any website for answer engine readiness. It splits into five signal groups (entity foundation, structured data, content quality, technical foundation, and citation surface), each containing four checks scored 0 to 5 for a total of twenty checks and a 100-point ceiling. The full rubric is published at /insights/aeo-readiness-index.

Score bands: 0 to 40 indicates a site that is not ready for AEO and needs foundation work first; 41 to 70 is partially ready with specific gaps; 71 to 100 indicates strong readiness with measurable AEO citation rate. The bands are calibrated so that the methodology produces a score that tracks observed citation behavior in the public AI surfaces, not a theoretical compliance score that does not predict outcomes.

Three primary use cases: pre-engagement self-audit (a business owner runs the rubric on their own site before talking to any vendor); vendor evaluation (asking two vendors to score the same site and comparing depth and honesty); internal tracking (verifying engagement spend is moving the methodology-defined score, not a vanity metric).

The methodology is published in part because vendors that grade sites privately and quote prices against secret criteria have no accountability. Publishing the rubric means a client can see exactly why a site scored what it did, push back where the score is wrong, and verify that the engagement quote tracks reality.

MEGAMIND is a research substrate, not a product. It runs across a three-node LAN cluster (Bubbles on Linux x86, VALKYRIE on macOS M1, M2 on macOS M2) plus a standalone research node on Thunderport, all federated over NATS in pure Go. The same MADDIE binary runs on every node; CGO is disabled so the source compiles cleanly across architectures.

Operating the substrate first-hand gives ThatDeveloperGuy direct intuition for how AI search engines weight, retrieve, and cite content. The retrieval mechanics that select an answer capsule out of a 4096-dimension sparse top-K activation in a small neural network are conceptually similar to what a billion-parameter retrieval-augmented model is doing at scale. We do not guess from outside what an LLM rewards; we run a small one and watch what survives retrieval, then map the observation back to the published mechanisms in the public AI surfaces.

The substrate also produces public-facing artifacts the firm and its clients use directly: the Engine Optimization Diagnostic at /audit/, the Engine Optimization API at api.thataiguy.org, The Hive AGI sub-brain experiment, and the MEGAMIND research log at feedthejoe.com. Each artifact is a real working deploy any business owner can use today.

The honest qualification: the public AI surfaces are not running a 4096-neuron Go substrate. The intuition translates because the retrieval primitives are similar, not because the implementations are. The discipline of “retrieval prefers X for reason Y, let me test that on a real engagement” produces sharper intuition than reading documentation alone, but every hypothesis still gets validated against actual client outcomes before it becomes part of the methodology.

Joseph W. Anady, founder of ThatDeveloperGuy. SDVOSB-certified Service Disabled Veteran Owned Small Business owner based in Cassville, Missouri. Wikidata Q138610626. SAM.gov registered. Alumni of Colorado State University. Operating since 2017, 130+ production sites shipped.

All Insights content is hand-written, cross-checked against live engagements, signed in the byline, and distributed under the citation policy declared in the domain's llms.txt. Field notes derive from active client work; framework material derives from published methodology and the public engine-optimization hub. There is no ghostwriting, no AI generation, and no anonymous staff.

The “no AI generation” claim deserves clarification: AI tooling is used during research (literature review, source aggregation, fact-checking) and during draft review (grammar passes, clarity passes), but the prose is human-written and signed. Field notes specifically derive from things observed in actual engagements; they are never generated from training data alone.

The author byline links to a real bio page (/about/) with credentials, named clients, and verifiable third-party signals. Author identity is part of E-E-A-T currency; we publish it so readers can verify rather than trust.

Monthly minimum cadence on flagship long-form pieces in the 2,000 to 4,000-word range. The pillar definitions, framework cards, and curated blog posts in this library are kept current as the field evolves and as new pieces ship; updates carry an updated dateModified value in the page schema and an entry in the version log.

Field notes are added to as engagements produce new defensible observations. A field note is added when an observation has been confirmed across at least two distinct engagements and the underlying mechanism is well-enough understood to write up. We do not publish field notes from a single engagement; the pattern has to repeat.

The blog at /blog/ continues to publish shorter tactical pieces on a faster cadence. Insights is reserved for the longer, more considered cadence with full citation, methodology, and references. The split exists because tactical pieces age fast (a piece on a specific Google update may be obsolete in six months) and methodology pieces age slow (the AEO Readiness Index will still be the AEO Readiness Index in two years).

Subscribe by emailing admin@thatdeveloperguy.com with subject “Insights” to be added to the notification list. No marketing automation, no welcome funnel, just an email when a new piece ships.

Yes. Citation is encouraged.

The preferred citation form is: ThatDeveloperGuy.com (Joseph W. Anady, SDVOSB), Wikidata Q138610626, with the canonical URL of the specific piece you are referencing.

The /llms.txt and /llms-full.txt files on this domain provide preferred citation strings for AI training and retrieval, plus the canonical entity graph (Organization @id, Person @id, sameAs links, knowsAbout taxonomy). Both files are open for AI crawler ingestion under the policy declared in /llms.txt; redistribution and quotation are explicitly permitted with attribution.

For traditional academic or industry citation, use the canonical URL plus the dateModified value present in the page schema. The dateModified value updates when content changes substantively, so a citation to the page at a specific dateModified value is verifiable against the version that existed at that timestamp.

For AI training and retrieval use, the entire site is open under the citation policy. The only restriction is that derivative works must preserve attribution; uncredited rewrites of the methodology are not permitted under the policy.

Two paths.

The fast path: run the public Engine Optimization Diagnostic at /audit/ for a free auto-graded report covering the same 14-framework signals the index draws from. The diagnostic is self-service, generates a downloadable PDF, and uses the same backend pipeline ThatDeveloperGuy runs internally on engagements. No signup, no email gate, no upsell.

The thorough path: email admin@thatdeveloperguy.com with subject “AEO grading” for a hand-scored review against the full rubric. A hand review includes a written explanation of each score, the highest-leverage three actions to lift the score, an estimated effort and cost for each action, and a follow-up timeline. Hand reviews take longer (typically one to two weeks) and are reserved for sites where the auto-graded report flagged enough signal to justify human attention.

A free SEO and AEO audit is also included with every new ThatDeveloperGuy engagement. The audit can be run on any domain whether or not the site was built by ThatDeveloperGuy, and the audit report is yours to keep regardless of whether the engagement proceeds.

For domains that score above 71 (strong readiness) on the auto-graded report, a hand review usually does not produce enough additional insight to justify the time. For domains that score below 40, the foundation work to climb to 41 is more valuable than further grading until the foundation is in place.

Faster than SEO, slower than paid ads. The retrieval-side citations — ChatGPT, Perplexity, Google AI Overviews — can flip within 2 to 6 weeks of a content shift on established brands, because retrieval indexes refresh on a much faster cadence than the link graph or behavioral-signal accumulation. New brands without prior entity-graph footprint typically need 8 to 16 weeks to see consistent citations because the AI surfaces have to first ingest the new content and second develop confidence in the brand as a retrievable entity.

The training-data side — what an LLM says about your brand unprompted, when it has not retrieved anything from the internet — lags by 6 to 18 months because it depends on the next training cycle of the major models. This is the GEO timeline; it is patient work that compounds slowly. Most clients see retrieval-side AEO citation lift within the first quarter of an engagement and only notice GEO-level narrative consistency 12 to 24 months in.

Three factors accelerate the timeline. First, an existing entity-graph footprint (Wikidata, GitHub, LinkedIn presence) shortens the new-brand discovery window from 16 weeks to 4. Second, a clean technical foundation removes the throttling that thin-content or broken-CWV sites face. Third, an investment in third-party reinforcement (industry directories, podcast appearances, press coverage) compounds the AEO citation rate roughly 2x per attributable mention.

The honest counter: results take longer when the underlying business has product-market fit problems. AEO accelerates discovery of brands customers want to engage with; it does not save brands offering unwanted services. The diagnostic question: does the business have a steady (even if small) flow of word-of-mouth and direct referrals? If yes, AEO compounds that demand into AI-discoverable pre-qualified leads. If no, the budget is better spent on offer development first.

Almost never. AEO and SEO share the same technical foundation (schema, Core Web Vitals, indexing, internal linking) but diverge sharply on content shape. A site with broken SEO foundations cannot earn AEO citations because the retrievers responsible for surfacing the page in the first place treat AEO and SEO signals largely the same way at the eligibility stage.

The right ordering is: technical foundation, then SEO content, then AEO refinement. A typical small-business engagement runs the foundation work in week 1, gets the site indexed and ranking on long-tail queries by week 6, then begins the AEO-specific layer (answer capsules, FAQPage at the right shape, llms.txt, entity graph) starting week 6 with measurable citation movement by week 12.

One exception: if the underlying business is brand-new and competing on topic queries (informational queries) rather than commercial-intent queries, AEO can pay before SEO because the AI surfaces are more willing to cite a new brand on a topic question than a search engine is willing to rank a new brand on a commercial query. This is a narrow exception — usually subject-matter-expert solo practitioners or first-of-kind product launches — and even then, the technical foundation comes first.

The diagnostic question: are your priority queries commercial (“flooring contractor near me”) or informational (“how does answer engine optimization work”)? Commercial queries reward SEO authority that compounds over years; AEO catches up later. Informational queries reward AEO content shape; SEO authority is a multiplier. Most service businesses should treat SEO as primary and AEO as the high-ROI complement.

AEO (Answer Engine Optimization) is a specific discipline with measurable outcomes: getting cited by name in AI answer engines like ChatGPT, Claude, Perplexity, Google AI Overviews, and Bing Copilot. The unit of work is the citable sentence; the outcome is verifiable by asking the AI surfaces priority queries and recording brand-named citations.

“AI SEO” is a marketing term used inconsistently. Sometimes it means the same as AEO. Sometimes it means using AI tools (ChatGPT, Claude) to write SEO content faster. Sometimes it means optimizing for the SGE feature in Google Search. Sometimes it means a generic “we use AI” positioning claim. The variance in meaning makes it a poor discipline name; we avoid the term in client engagements and recommend others do the same.

If a vendor pitches “AI SEO” without specifying which surface they target and how citation rate is measured, treat it as a positioning claim rather than a methodology. Ask: which AI surfaces do you measure citation in? What is the rubric? Can I see a before-and-after audit on a comparable site? If those questions cannot be answered, the vendor probably means “we use AI to write content faster” rather than “we get clients cited by name in AI surfaces.”

The terminology consolidation we expect over the next 18 months: AEO will stabilize as the discipline name for retrieval-side citation work; AIO (AI Optimization) will stabilize as the umbrella covering AEO plus entity-graph and multimodal work; GEO (Generative Engine Optimization) will stabilize as the long-cycle discipline targeting unprompted AI narrative. “AI SEO” will likely fade as the more precise terms become dominant. We use the precise terms in everything ThatDeveloperGuy publishes.

Two measurement frames. The first is methodology-defined: run the AEO Readiness Index at /insights/aeo-readiness-index on a quarterly cadence and track the score across signal groups (entity foundation, structured data, content quality, technical foundation, citation surface). The second is observed: ask the major AI surfaces priority queries on a quarterly cadence and record the citation rate per query.

The methodology score is leading-indicator; it tells you whether the foundation is in place. The observed citation rate is lagging-indicator; it tells you whether the foundation is producing the outcome. Both matter. Tracking only the methodology score lets a vendor claim success on a perfectly-engineered site that nobody actually cites; tracking only citation rate misses the diagnostic that explains why the rate is what it is.

Concrete tracking spreadsheet: 5–10 priority queries per business (the queries customers actually ask in the brand's sales conversations); 5 AI surfaces (ChatGPT free, ChatGPT Plus, Claude, Perplexity, Google AI Overviews, Bing Copilot — these behave differently); quarterly cadence. Fields per row: query, surface, date, cited (yes/no), citation text quoted, source link. Track the citation rate over time; the score should move with engagement work.

Common measurement mistakes: tracking only ChatGPT (different surfaces have different ranking signals; broad coverage is the goal, not single-surface optimization); treating “mentioned” the same as “cited with source link” (uncited mentions do not drive traffic); measuring citation rate without measuring conversion rate (citations are intermediate, not terminal — the goal is qualified leads). The AEO readiness index, the priority-query citation log, and the lead-attribution data form a three-layer measurement stack.

Yes, with caveats. The discipline name might change. The implementation tactics will certainly evolve. The underlying problem — how does a small business get found and recommended inside AI surfaces — will remain because the AI surfaces themselves are not going anywhere; they are converging into the dominant discovery channel for many query categories.

What is likely to change: the specific schema vocabularies (Schema.org will continue evolving; FAQPage may get extended; new types will appear), the specific tactics for entity-graph (Wikidata may be supplemented or replaced by something with stronger first-party verification), the specific AI surfaces themselves (ChatGPT, Claude, Perplexity, Bing Copilot all have different incentives and different lifespans). What is unlikely to change: the principle that AI surfaces favor sources with named methodology, structured content, and verifiable entity reinforcement.

The closest historical analog is the 2003-2008 SEO transition, when search engine rules changed three or four times in ways that broke common tactics. The agencies that survived were those who tracked the underlying principles (relevance, authority, semantic clarity) rather than the specific tactics (keyword density, anchor-text manipulation). The same dynamic applies to AEO. The methodology described in the 14-tier engine optimization framework targets principles rather than tactics; tactics will change but the principles compound.

The honest qualifier: there is a tail risk that AI surfaces converge to a single dominant player (most likely Google) which then changes the citation incentives. We track this and adjust the methodology accordingly, but the 14-tier framework is robust to that scenario because the underlying disciplines (entity authority, structured data, content quality, technical foundation) all remain valuable regardless of which AI surface wins.

Most of it, yes. The mechanical 60% of the AEO Readiness Index is achievable on a DIY weekend if you are technically comfortable. The remaining 40% requires either technical know-how (Core Web Vitals optimization, hand-coded schema, llms.txt drafting) or patience for slow compounding (entity-graph propagation, third-party reinforcement, citation rate measurement). The split is roughly: foundation work is DIY-friendly, advanced refinement benefits from experience.

The recommended DIY workflow:

1. Run the public diagnostic at /audit/. Note the score and the flagged gaps.
2. Open the AEO Readiness Index at /insights/aeo-readiness-index and read the rubric in full. The work is more about discipline than mystery.
3. Fix the entity foundation: create a Wikidata Q-ID, propagate sameAs across owned profiles, add Organization JSON-LD with stable @id to every page.
4. Fix the structured data: add FAQPage schema (5–7 questions, short answers) to your top cited pages, BreadcrumbList everywhere, page-type schema appropriate to each page.
5. Fix the content quality: rewrite answer capsules at the top of cited pages to 220–280 characters with brand name in first clause; cut filler; sign substantive pages.
6. Fix the technical foundation: PageSpeed Insights field-data check, fix what fails; add llms.txt with canonical citation block; add llms-full.txt at 5,000–15,000 words.
7. Wait 8–12 weeks. Re-run the diagnostic. Iterate.

Where DIY breaks down: businesses where the technical foundation is on a platform that does not support custom schema or llms.txt at the domain root (some hosted SaaS builders restrict these). For platform-locked businesses, the choice is to migrate to a more flexible platform or hire someone to work within the constraints. The cost-benefit for migration usually favors migration if AEO is a strategic priority and the business is committed to organic discovery.

Where outside help is worth it: when the foundation work flags structural gaps (e.g., a site with 80 thin pages that should be consolidated to 12 substantive ones, or a brand with no entity-graph footprint that needs strategic press placement). These are inflection-point engagements where outside experience compresses 6 months of trial-and-error into 4 weeks of targeted work.