Answer Engine Optimization That Earns the Citation

An answer engine does not rank ten results and hand you a list. It assembles one answer, and to do that it has to decide which entities it is confident enough to name in a sentence it takes responsibility for. That decision runs on three inputs, and understanding them is most of the job. The first input is retrieval. When a buyer asks a commercial question, the assistant usually issues its own search queries behind the scenes and reads a handful of pages from the results, plus whatever the model already learned in training. If your page is not retrievable for the reformulated query the model wrote, nothing else matters. This is why classic organic visibility is a prerequisite rather than an alternative: models overwhelmingly retrieve from the top of conventional results, then reason over what they read. The second input is corroboration. Models weigh claims that appear in more than one independent place far more heavily than claims that appear only on your own website. A firm described consistently across its Google Business Profile, an association directory, a local news mention, a podcast appearance, and two roundup posts is a firm the model can name with confidence. A firm whose only description exists on its own homepage is a firm the model hedges about. The third input is extractability. Even a well-corroborated business gets skipped when its pages bury the answer in a narrative. Models quote sentences, not pages. A page that opens a section with a direct declarative answer, then supports it, is dramatically easier to lift into a generated response than a page that warms up for four paragraphs. Every tactic below is downstream of these three inputs.

You cannot optimize for answers you have never read. The first deliverable in every engagement is a fixed prompt library of 25 to 100 real buying questions, written the way a buyer types them into an assistant, not the way a keyword tool phrases them. Build it in four groups. Direct vendor prompts: who are the best commercial roofing contractors near me, which firm should I hire to handle multi-state payroll compliance. Qualification prompts: what should I ask a marketing agency before signing, how much does commercial HVAC maintenance cost per square foot. Comparison prompts: is an in-house marketer or an agency better at our size, which is better for a law firm, local SEO or Google Ads. Problem-first prompts, which are the ones most firms forget: our inbound leads dropped 40 percent after a website migration, what do we do. Run every prompt across the engines your buyers actually use, and record the answer verbatim, the businesses named, the sources cited with URLs, and the position in which each business appears. That transcript file is your baseline. Everything you do afterward is measured as a delta against it, which is what separates a real AEO program from a list of best practices someone applied once.

An entity is a thing a model can identify unambiguously. Most small and mid-sized service firms are not entities to an AI model, they are strings of text that might refer to several different companies. Resolving that is the highest-leverage technical work in AEO, and it is mostly unglamorous. The checklist: one canonical business name used identically everywhere, down to the punctuation. Organization schema on the homepage with a stable @id, logo, founding date, and a sameAs array pointing at every profile you control. LocalBusiness schema with a single consistent NAP. Person schema for the founder or principal, linked from the Organization node, with a real author bio on the content they write. A Wikidata item when you qualify for one. Clean, matching profiles on Google Business Profile, LinkedIn, Crunchbase, industry associations, and licensing boards. The failure mode we see most often is drift: the legal name on the licensing board, the shortened name on the website, and a third variation on LinkedIn. Each variation splits the evidence the model would otherwise pool into confidence about one entity. Fixing name drift across a dozen profiles is a two-week project that frequently produces the first measurable mention-rate movement, before a single new page ships.

Answer-shaped content follows a specific structure, and it is different enough from ranking-shaped content that retrofitting is usually faster than rewriting from scratch. Open every H2 section with a one-sentence definitional answer to the question that heading implies. Put the number, the range, or the verdict in that sentence rather than three paragraphs later. Use real figures with context: not 'affordable pricing' but 'typically $1,500 to $4,500 per month for a firm running one market.' Publish honest pricing at all, which most competitors will not do, and which models reward because a page with numbers answers the question the buyer actually asked. Add a short summary block near the top of long pages. Keep paragraphs tight enough to quote whole. The formats that earn citations disproportionately are comparison pages, cost and pricing pages, checklists with a stated methodology, and pages that answer a question with a documented process rather than a promise. Publish the criteria you use to make a recommendation, including when you would tell someone not to hire you. Models pick that up as balance, and buyers who arrive from it convert at a materially higher rate than the average organic visitor. Our organic SEO and content program covers the ranking side of the same content, and the two workstreams share roughly sixty percent of their production.

This is the part that cannot be automated and the reason cheap AEO offers do not work. What other credible sites say about you outweighs what you say about yourself, because a model that only has your own marketing copy has no independent evidence to reason with. The sources that carry weight, in rough order: a Google Business Profile with a steady, unincentivized review cadence and owner responses that add detail. Named mentions in local or trade press. Directory and membership listings from associations that verify membership. Inclusion in third-party roundups and comparison articles in your category. Podcast and webinar appearances with a written show page. Client case studies published on the client's own site. Speaking or teaching listings. We cannot and do not automate review generation. What works is a documented request workflow tied to the moment a job completes, run by a named person on your team, with our templates and follow-up cadence behind it. Everything else on this list is manual outreach: pitch, follow up, get published, keep the description accurate. Five real placements a quarter beat fifty directory submissions, and the difference shows up in citation rate within about a quarter.

You cannot be cited by a system that is not allowed to read you. A surprising share of the sites we audit are blocking at least one major AI crawler, usually because a plugin default, a security vendor, or a well-meaning developer added the rule and nobody revisited it. Check robots.txt for GPTBot and OAI-SearchBot from OpenAI, ClaudeBot from Anthropic, PerplexityBot, Google-Extended, Applebot-Extended, Bytespider, and CCBot. Note that Google-Extended controls Gemini and AI training use but does not remove you from AI Overviews, which is governed by ordinary Googlebot access, a distinction that trips up a lot of teams. Also check the layer above robots.txt: bot-management rules at your CDN or WAF frequently drop AI crawlers with a 403 while robots.txt says allow, and only a server log review or a fetch test will surface it. Then publish llms.txt and llms-full.txt at your domain root. Neither is a ranking factor and neither is officially required by any engine. What they do is give a crawler a clean, canonical summary of who you are and which URLs you consider authoritative, which reduces the odds of being described with stale or invented details. Treat them as low-cost hygiene, not as the strategy.

AEO reporting fails when it reports activity. Track outcomes on a fixed prompt library instead, and always against the same library, or the trend is meaningless. Mention rate: the percentage of prompts where your business is named at all. Citation rate: the percentage where your domain is actually linked as a source, which is the number most correlated with referral traffic. Position within the answer: first-named businesses receive a disproportionate share of the resulting contact, so moving from fourth-named to first-named matters even when mention rate is flat. Accuracy and sentiment: whether the description of you is correct, current, and favorable, because an inaccurate mention is a problem to fix rather than a win to report. Referral sessions in GA4 from chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, and claude.ai, segmented and tracked to form fills, plus the question-shaped queries in Search Console that indicate AI Overview exposure. Realistic timelines from the programs we run: entity and crawler fixes show mention-rate movement in roughly 30 to 60 days. Citation growth follows corroboration work and typically compounds over 60 to 120 days. AI-sourced referral traffic becomes a measurable, attributable channel around month three to four for most firms, and volumes stay modest in absolute terms while converting well above average, because the buyer arrives pre-qualified by the assistant's recommendation.

Buying schema markup and calling it AEO. Structured data makes a page machine-readable, it does not make you worth naming. Schema with no corroboration and no answer-shaped copy changes nothing, and it is the single most common thing sold under the AEO label. Optimizing for volume prompts nobody buys from. Ranking inside an answer to 'what is SEO' produces no pipeline. Build the library from purchase-intent questions. Treating one audit as a program. Model outputs shift week to week, and a competitor who ships a comparison page can displace you between two runs. Without a recurring cadence you have a snapshot, not visibility. Stuffing FAQ schema onto pages that do not display those questions. It is a mismatch between markup and page content, it risks a structured-data penalty, and it does nothing for citation rate. Spinning up thin location or service variants at scale. Near-duplicate pages dilute the entity signal and give the model no reason to prefer any one of them. Depth on fewer pages consistently outperforms breadth here. Ignoring conversion once the traffic arrives. Assistant-referred visitors show up already convinced and leave immediately when the page has no clear next step. Pair AEO with the conversion work on the pages the citations point at, or you are buying qualified visitors and letting them bounce.