Search increasingly ends in an answer rather than a list. AEO is the work of being the source that answer is built from. This course covers what actually influences citation — and separates it from the noise the category has attracted.
⏱ 6 min read
Answer engine optimisation (AEO) is the practice of structuring content so that AI-powered answer systems can extract it, trust it, and cite it as the answer to a question. An answer engine is anything that responds with a synthesised answer instead of a list of links: Google's AI Overviews and AI Mode, ChatGPT with browsing, Perplexity, Claude, Copilot.
Classic SEO competes for a position in a list. AEO competes for inclusion in the answer itself — as the source being paraphrased, or as a named, linked citation. The two can come apart. A page ranked fourth can be the source an AI Overview quotes while the page ranked first is ignored entirely.
Behaviour moved. A large and growing share of searches now end without a click to any external site — one clickstream analysis put that figure around two-thirds of Google searches. Meanwhile a meaningful part of early research has shifted into chat interfaces, where there is no results page at all. If your visibility strategy assumes a blue link is the unit of discovery, you are measuring a shrinking surface.
| Term | What people mean by it |
|---|---|
| AEO | Answer engine optimisation — the umbrella term for being the answer |
| GEO | Generative engine optimisation — same idea, emphasis on generative results |
| AI SEO / AIO | Vendor labels for the same work |
| SEO | Still the foundation all of the above depends on |
The tactics under these labels overlap almost completely. Do not pay four times for one discipline, and be sceptical of anyone presenting AEO as a replacement for SEO rather than a layer on top of it.
Three outcomes, in descending order of measurability: a linked citation in an answer (sends referral traffic), a brand mention without a link (shapes the consideration set before anyone visits a site), and inclusion in the synthesis where your framing shows up unattributed. The last is real influence and almost impossible to attribute, which is the central measurement problem of this whole field.
This is a young discipline with more confident claims than evidence. Published figures vary widely by study, industry and platform, and vendors selling AEO tools produce most of the research. Treat specific percentages in this course, and anywhere else, as directional. What is well supported is the direction: structure, freshness, corroboration and crawlability influence whether a page gets used.
AEO pays off unevenly. It matters most where buyers research before they buy and where the question has a defensible answer: professional services, B2B software, healthcare, finance, education, and any considered purchase. It matters least for pure transactional demand where people already know what they want and simply need somewhere to buy it, and for brands whose traffic is overwhelmingly navigational.
It also matters more the earlier you are in a buyer's process. Someone asking "what is a fractional CMO" is forming a category understanding; being the source that shapes it is worth more than a click later on.
Surveys through 2026 consistently show the same pattern: a large majority of organisations expect answer engines to reshape their digital strategy within a few years, while only a fraction have started doing anything about it. Whatever the exact numbers, the gap is the opportunity. The structural work in this course is unglamorous and most competitors have not done it.
⏱ 6 min read
You cannot optimise a black box you have not looked at. Every answer engine runs roughly the same pipeline, and each stage is a place you can be filtered out.
Two consequences follow directly. First, retrieval still runs on classic search infrastructure, so crawlability, indexing and authority all still gate you. Second, because extraction works on passages, a brilliant page whose key claim is buried on paragraph nineteen, split across three sentences and dependent on earlier context, is a page that is hard to lift — and therefore easy to skip.
Research from several vendors converges on a few patterns worth knowing. A substantial share of AI Overview citations come from pages already ranking in the classic top ten, which is the clearest evidence that SEO underpins AEO. Analyses of cited pages also find citations skew heavily toward material near the top of the source page — CXL's review of 100 AI Overview citations put more than half in the first third of the page. And for commercial and evaluation queries, freshness dominates: AirOps found the large majority of citations came from pages updated within the last year.
| Engine | Retrieval basis | Tends to favour |
|---|---|---|
| Google AI Overviews / AI Mode | Google's index | Pages that already rank well; strong page structure |
| Perplexity | Its own crawl plus search APIs | Fresh, well-sourced, clearly-cited articles |
| ChatGPT with browsing | Search partner plus live fetch | Authoritative, in-depth pages on established domains |
| Claude / Copilot | Search partner plus live fetch | Clear, factual, well-structured sources |
Prioritise by where your buyers actually ask questions. For most B2B and local service businesses, that is still overwhelmingly Google.
How AI Is Changing Google Search and SEO — Google Search Central. Martin Splitt with Nikola Todorovic on the move from classic results to AI Overviews and AI Mode. Open on YouTube ↗
Fan-out means the engine decides what the sub-questions are. Ask "should I use WordPress or Shopify for a small store" and the engine may separately retrieve for hosting costs, transaction fees, ease of editing, and plugin availability. A page that covers one of those well can be cited even though the headline question was never its target.
The practical implication is to cover the neighbouring questions inside the same page rather than assuming your main heading is what gets matched. Each sub-question gets its own H2 and its own short, self-contained answer.
Every answer engine exposes its sources. Treat that list as free competitive research. For each of your priority questions, note which domains appear, whether they are competitors or publishers, roughly where on each page the quoted passage sits, and what format it took — a definition, a table row, a step. Repeat monthly and you have a visibility baseline that costs nothing but attention.
⏱ 6 min read
Almost everything carries over. The useful question is what changes at the margin.
| Classic SEO | AEO | |
|---|---|---|
| Goal | Rank in a list | Be quoted in an answer |
| Unit of competition | The page | The passage |
| Success metric | Clicks, position | Citations, mentions, share of answers |
| Content shape | Comprehensive, engaging | Comprehensive and liftable out of context |
| Freshness | Matters for some queries | Matters for most commercial queries |
| Authority | Links and brand | Links, brand, and corroboration across sources |
| Feedback loop | Days to weeks | Structural fixes can show in days; authority still takes months |
Self-contained claims. Write sentences that survive being cut out. "It typically takes six to twelve months for a new site to see meaningful organic traffic" works alone; "as we saw above, it usually takes that long" does not. Name the subject in the sentence rather than relying on a pronoun pointing at the previous paragraph.
Answer placement. Put the direct answer early — within the first screen — then expand. This inverts the storytelling instinct of most content writers and it is the single highest-leverage structural change available to you.
Corroboration. Models weigh whether a claim is supported in more than one place. A fact that appears only on your site reads as an assertion; one echoed by an industry body, a news article and a supplier page reads as established. That turns PR, partnerships and industry participation into visibility work.
If the page cannot be crawled, it cannot be retrieved. If it is not indexed, it is largely invisible to the systems that retrieve from a search index. If the site has no authority, it will lose to sources that do. If the content is thin, restructuring it into neat question headings makes it a well-organised thin page. Every foundation from the SEO course is a prerequisite here.
Treating AEO as a separate content programme. The best-performing AI-cited content is usually also the best-performing organic content. Running a parallel set of "AI pages" splits your authority, duplicates your maintenance, and creates exactly the kind of thin, machine-targeted material that quality systems are built to discount.
Be realistic with clients about the shape of the outcome. Being cited in an answer sends less traffic than ranking first used to, and the traffic it does send tends to convert better, because the visitor arrived pre-informed and chose you from a shortlist. The overall pattern for most informational content is fewer clicks, higher-quality clicks, and a growing share of influence you cannot see in a session count.
That argues for two changes in how you measure and what you publish. Report influence alongside traffic. And invest content effort where a click is still necessary — tools, calculators, bookings, detailed comparisons, anything the answer cannot fully substitute for.
A page written only to be quoted gives a reader no reason to visit. Keep something on the page that the answer cannot carry: the worked example, the template, the calculator, the full data set, the human you can talk to. The citation earns the consideration; the page has to earn the visit.
⏱ 6 min read
Structure is the part of AEO fully in your control, and the part that shows results fastest. The pattern is consistent across platforms.
Direct definitions. Numbered processes. Comparison tables. Short pros-and-cons lists. Specification-style facts with units. Dated figures with a named source. Narrative prose, however good, is the hardest format to extract cleanly — keep it, but do not make it carry the load-bearing claims.
Do not write question headings with no real answer beneath them. Do not stack twenty FAQ entries at the bottom of every page as an AEO ritual. Do not publish the same answer on six near-identical pages. Systems designed to identify unhelpful scaled content handle all three, and they read as gaming rather than helping.
Treat your heading outline as the index an engine reads first. Each H2 should be a question someone would actually type or say, and the two or three sentences directly beneath it should answer that question completely. If a reader could jump straight to any H2 and get a usable answer without reading anything above it, the page is structured correctly.
This also improves the page for humans, who scan rather than read. The two audiences want the same thing, which is the reason AEO structure is not a compromise.
Comparative information in a real <table> is unambiguous to a machine: rows are things, columns are attributes, cells are values. The same information in three paragraphs of prose has to be inferred. Analyses of AI answers consistently find lists and tables over-represented in what gets pulled, and it is the cheapest structural change you can make to an existing page.
A short FAQ answering questions the body genuinely did not cover is useful. A twenty-item FAQ bolted onto every page, repeating what is already above it, is padding — and Google explicitly restricted FAQ rich results for most sites, so it does not even buy the SERP real estate it once did.
⏱ 6 min read
AEO keyword research is question research. You are looking for the prompts your buyers type into a chat box, which are longer, more conversational and more specific than the keywords they typed into Google five years ago.
| Bucket | Example prompt | Page design that wins |
|---|---|---|
| Definitional | what is answer engine optimisation | Plain-language answer first, then context, distinctions, use cases |
| Process | how do I get cited by ChatGPT | Ordered steps, checklists, worked example, troubleshooting |
| Comparison | AEO vs SEO | Balanced table, explicit criteria, a recommendation per scenario |
| Decision | best AEO tools for a small agency | Honest shortlist with trade-offs, pricing, who each suits |
Ask the question in two or three answer engines and read what comes back. Who gets cited? What did those pages do structurally? Is the answer settled and stable, or does it change between engines? An unstable answer is an opportunity; a settled one dominated by a government source or a major publisher usually is not.
Three filters. Does an answer engine already answer this? If it does, there is a citation slot to compete for. Would being the cited source matter commercially? Brand-adjacent and decision-stage questions matter far more than trivia. Can you answer it better than the current sources? With data, experience or specificity they lack — if not, pick a narrower question where you can.
Fan-out means a single page can be retrieved for many related sub-questions. One thorough page covering a question and its neighbours usually outperforms eight thin pages, one per phrasing — and it is a fraction of the maintenance.
Prompts are longer and more conversational than keywords. People type "is it worth paying for managed WordPress hosting for a small nonprofit" where they once typed "managed wordpress hosting". Write headings and answers in that register — full questions, natural phrasing, the qualifiers people actually include — rather than compressing everything into a two-word head term.
Your inbox is the best question bank you own and almost nobody uses it. Pull the last fifty pre-sale enquiries and the last fifty support tickets, strip out anything client-identifying, and group the recurring questions. These are real prompts from real buyers, phrased in their language, weighted by how often they actually come up — which is more than any keyword tool can tell you. Ten pages built from that list will outperform fifty built from a volume export.
Before you change anything, run your priority questions through each engine and save the output — date, engine, answer, and the sources cited. Answers vary between sessions and users, so a single check is anecdote; the same ten questions run monthly is a trend line. A simple spreadsheet is entirely sufficient.
⏱ 6 min read
Structured data does not make an AI cite you. What it does is remove ambiguity about what your page contains and who you are, and ambiguity is a reason to pick someone else.
| Type | Use on | Why it helps |
|---|---|---|
| Organization | Homepage / about | Establishes the entity: name, logo, profiles, contact |
| Article / BlogPosting | Content pages | Author, publish and modified dates — the freshness signals |
| FAQPage | Pages with genuine Q&A | Pairs each question with its answer explicitly |
| HowTo | Step-by-step guides | Makes the sequence machine-readable |
| Product / Offer | Ecommerce | Price, availability, reviews as data rather than prose |
| LocalBusiness | Location pages | Address, hours, service area for local answers |
| Person | Author bios | Ties content to a real, describable expert |
Use JSON-LD. Two rules keep you safe: mark up only what is visible on the page, and validate before shipping. Markup describing invisible content is a guidelines violation, and rich result eligibility disappears along with the trust.
An entity is a thing with an identity — a company, a person, a product, a place, a concept. Search and AI systems reason about entities and their relationships rather than string matches. Practically, this means being consistent and explicit: one company name spelled the same way everywhere, author identities that resolve to a real person with a bio and profiles, and pages that state relationships plainly ("CLIKT is a web and content agency in Northeast Ohio specialising in WordPress, Shopify and Lovable builds") rather than implying them.
Use sameAs in your Organization and Person schema to link to the profiles that corroborate you — LinkedIn, Crunchbase, industry directories, Wikipedia where applicable. You are drawing the graph rather than hoping it gets inferred.
Monitoring Rich Results in Search Console — Google Search Central. Daniel Waisberg on validating and monitoring structured data — the same markup that disambiguates your pages for AI systems. Open on YouTube ↗
Google has said repeatedly that structured data is a clarity mechanism, not a ranking lever, and the same logic applies to AI citation. Schema on a thin page does not make it citable. Schema on a strong page removes doubt about what it says. Add it where it describes something real, skip it everywhere else, and do not let a plugin bury your site in markup nobody asked for.
On WordPress, an SEO plugin handles Article, Organization and breadcrumbs with minimal configuration; add FAQ and HowTo deliberately, on pages that genuinely contain them. On Shopify, Product schema usually ships with the theme — check it renders the real price and availability. On a custom or Lovable build, write the JSON-LD yourself and validate it with the Rich Results Test and Schema.org's validator.
Two schema properties carry more weight than people realise: datePublished and dateModified. They should reflect reality and match the date visible on the page. Many themes emit a modified date that changes whenever any field is touched, which produces a site where everything was apparently updated yesterday — a pattern that is easy to detect and worth nothing. Check what your CMS actually outputs before assuming the signal is working for you.
⏱ 6 min read
Structure gets you extractable. Authority gets you chosen. When several pages can answer a question equally well, the tiebreakers are who says it, whether anyone else backs it up, and when it was last true.
Answer engines weigh whether a claim appears in more than one credible place. A statistic that exists only on your site is an assertion. The same statistic echoed by an industry association, a trade publication and a partner's case study is a fact with provenance.
What produces corroboration: original research others cite, quotes in industry press, association and chamber membership pages, conference and podcast appearances, partner and supplier listings, review platforms, and consistent public profiles. Most of this is ordinary business development — which is why AEO belongs with marketing rather than inside a technical SEO silo.
Recency weighs more heavily in AI citation than in classic ranking, especially for commercial and evaluation questions. AirOps' 2026 analysis of citation patterns found the large majority of citations for those queries came from pages updated within the previous twelve months, with a substantial share refreshed inside six.
The operational answer is an update cadence, not a publishing quota:
| Content type | Review every | What to change |
|---|---|---|
| Pricing, comparisons, tool lists | 3 months | Figures, names, availability, anything dated |
| How-to and process guides | 6 months | Interface changes, new steps, dead screenshots |
| Definitional and reference pages | 12 months | New distinctions, superseded terminology |
| Case studies and evidence pages | 12 months | Newer results, refreshed numbers |
Update the visible date honestly and mark it up with dateModified. Bumping a date without changing anything is a trust problem, not an optimisation.
Anonymous content is weaker in every framework now in play. Attach a real name, a real bio, and real credentials to anything making claims. Link the author to profiles that verify the claim of expertise. For Joe's kind of work, "written by the person who has run this agency for nearly two decades" is a stronger signal than any amount of keyword tuning.
Being genuinely known in a subject is the durable advantage. Structural fixes can change extraction within days; authority and corroboration compound over months. Plan accordingly, and be sceptical of anything promising the second on the timeline of the first.
The most reliably cited content is content that cannot be found anywhere else: your own survey, your own benchmark numbers, a before-and-after from real work, a method you can name. Synthesis of other people's material competes with every other synthesis of the same material, and the model has no reason to prefer yours. One piece of original research per quarter outperforms twelve summaries.
Say the same thing about yourself everywhere. One company name, one description of what you do, the same founding facts, the same service list across your site, your profiles, your directory listings and your press mentions. Contradictory descriptions create ambiguity about which entity is which, and ambiguity costs you at exactly the moment a system is deciding whether you are an authority on something.
⏱ 6 min read
Everything above assumes the machines can read your page. Some of that is familiar from SEO; some is new, and some of the new advice circulating is wrong.
Beyond Googlebot and Bingbot, several AI-specific agents fetch pages: GPTBot and OAI-SearchBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot, Google-Extended (which controls use in Google's generative products without affecting Search ranking). They are declared in robots.txt like any other user agent.
Blocking them is a legitimate choice with a straightforward consequence: you will not be cited by the systems you block. Publishers with licensing leverage sometimes block deliberately. A small agency or ecommerce business trying to be discovered almost never should. Whatever you decide, decide it explicitly rather than inheriting whatever your host or plugin shipped.
Many AI crawlers do not execute JavaScript, or do so less reliably than Googlebot. If your primary content only exists after a client-side fetch, an AI crawler may retrieve an empty shell. Server-side rendering or static HTML is the safe default. Check by fetching your own page with JavaScript disabled and reading what is left — if the answer is not in that text, assume it is invisible.
Two ideas circulate widely and deserve a clear answer. The first is converting your site, or publishing a parallel copy, in Markdown so that models can read it more easily. The second is llms.txt, a proposed file listing your content for large language models.
On the Search Off the Record podcast in 2026, Google's John Mueller and Martin Splitt addressed both directly. HTML has been the web's standard for longer than Markdown has existed, every crawler is built around it, and converting HTML to plain text is trivial. Splitt noted that stripping to Markdown actually removes link structure, navigation and heading relationships that help systems understand context. On llms.txt, Mueller pointed out that a file the site writes about itself is self-reported information with no discovery mechanism, which limits how much any system can rely on it. As of 2026 no major model vendor has committed to honouring it.
The practical reading: llms.txt costs almost nothing if you want it, and it is not a strategy. Parallel Markdown copies carry the same maintenance and duplication problems as every parallel-version scheme before them. Spend the effort on clean, server-rendered, well-structured HTML instead.
<div> soup.<table> elements; key figures as text, not baked into images.Almost everything that makes a page usable with a screen reader also makes it usable to a crawler: real heading hierarchy, descriptive link text, alt text that describes the image, tables with proper headers, and content that exists in the markup rather than being drawn by script. If you already do accessibility work, most of the AEO technical layer is already done — and if you do not, this is a second reason to start.
A crawler that times out does not retry indefinitely. Slow server responses, aggressive rate limiting and bot-protection rules that challenge unknown user agents can all quietly remove you from the candidate pool. If your host or firewall blocks non-browser traffic by default, check what it does with the AI user agents you have decided to allow — permitting them in robots.txt means nothing if the edge is returning a challenge page.
⏱ 6 min read
Measurement is the least mature part of this discipline. Be honest about that with clients rather than presenting estimates as facts.
| Signal | Source | Reliability |
|---|---|---|
| AI-feature impressions and clicks | Search Console generative AI reports | Good, where available — first-party Google data |
| Referrals from AI platforms | GA4 traffic acquisition, by source | Good for platforms that send a referrer |
| Manual citation checks | Asking your questions in each engine | Directional — answers vary by user and session |
| Branded search volume | Search Console, Google Trends | A lagging proxy for AI-driven awareness |
| Vendor citation tracking | Third-party AEO tools | Sampled; useful for trends, weak as absolute truth |
Google launched dedicated generative AI performance reporting in Search Console in June 2026, rolling out to a subset of properties first. If your property has it, it is the most trustworthy measurement available. In GA4, add AI platforms as a channel or segment in Traffic acquisition — the volumes are typically small, and the conversion rate is often notably high, because someone who clicks through from an answer has already been pre-sold.
Some visibility is uncountable. A model that paraphrases your framing without naming you still shaped a buying decision. This is not an excuse for unfalsifiable reporting — it is a reason to report AI visibility alongside classic organic performance, and to judge the programme on the combination.
Week 1 — baseline. List your ten most commercially important questions. Ask each in ChatGPT, Perplexity and Google AI Mode. Record who gets cited and what those pages look like. Check whether your Search Console property has the generative AI view.
Week 2 — structure. Take the three pages closest to winning and rebuild their openings: question H1, 40–60 word answer, evidence, sub-questions in their own sections. Add a table where one is warranted. Fix any content that only exists after JavaScript runs.
Week 3 — trust. Add or upgrade author bios and Person schema. Add Organization schema with sameAs profiles. Refresh dated figures across your top pages and update dateModified honestly. Review robots.txt and decide your AI crawler policy deliberately.
Week 4 — corroboration and cadence. Pursue two external mentions of your key claim — an association page, a partner, a trade publication, a podcast. Set the update cadence from module seven. Re-run week one's prompts and log what changed.
Then repeat monthly. Structural changes can show up within days because answer engines re-crawl frequently; authority and corroboration compound over a quarter or more. Both are needed, and neither substitutes for the SEO foundations underneath.
The AEO tool category is young, expensive, and full of vendors whose own research is their marketing. Citation trackers sample prompts and report frequency, which is genuinely useful for trend direction and unreliable as absolute measurement, because answers vary by user, session and location. Before recommending one, ask what it samples, how often, and whether the number would change a decision. For most small businesses, Search Console plus a monthly manual check of ten questions is enough, and the money is better spent on the content itself.
⏱ ~6 min
Answer all ten, then score yourself. Explanations appear after scoring.
All consulted September 2026. Search and AI products change fast — check the primary documentation before acting on anything time-sensitive.