Subscribe
Share
Share
Embed
A bite-sized, charm-filled podcast that demystifies Answer Engine Optimization (AEO) for everyday content creators. Each 5-10 minute episode breaks down one key AEO concept in an accessible, entertaining way. Listeners will learn how to optimize their content for AI-powered search tools like ChatGPT, Siri, and Google Assistant - without the technical jargon that makes their eyes glaze over.
Alright then, hello hello, my lovely listeners, and you’re very welcome to AEO Decoded. If you’re a regular, I’m dead chuffed to have you back. If you’re new, you’re very welcome as well, you’re in the right place. This show is where we take Answer Engine Optimization, shake the jargon out of AEO, and leave you with something you can actually use, no waffle, and certainly no skulduggery. Today we’re talking about content auditing, and here’s the good news: you don’t need to start from scratch. You’re probably sitting on pages that are already close to working in AI systems. They just need a wee bit of tightening and a smarter structure, so they can get surfaced and cited. I’m Gary Crossey — kettle on — and we’ll crack on. Let’s get into the content audit. What I see all the time is teams spending months churning out brand new AEO content while completely ignoring the goldmine they’re already sitting on. If you’ve been publishing for a while, you’ve likely got hundreds, maybe thousands, of pages live, and the surprising part is that some of those pages already have traffic, authority, and decent traditional rankings — so you’re not starting from zero, you’re starting from “nearly there”. The key point is you can’t just wander around your site randomly changing paragraphs and hoping for the best; you need a simple, repeatable framework you can run every quarter without losing your mind. So let’s walk through my content audit framework, nice and simple. Start by picking your top 10 pages — and those 10 stay the same right through the whole framework. They’re your sample set for every step, so you’re comparing like with like. Paste the 10 URLs into a simple tracker — a spreadsheet or a Notion database — then run step one across all 10 URLs, step two across all 10 URLs, and keep going until you’ve completed the full run. When you’re done, you’ll have a clean, prioritised list of fixes, in the right order.
Step One: Content Inventory and Classification. Right, step one is your “get organised” moment — and I mean that in the best possible way. Before we start tweaking answers, adding schema, or testing ChatGPT, we need to know what we’re actually working with. Because if you don’t take stock, you’ll end up optimising random pages based on vibes… and that’s how audits turn into chaos. So step one is simple: build a small inventory of the pages you’re going to audit. For this run, we’re keeping it tight — the same top 10 pages you picked at the start. Those 10 URLs are your working list. Now, open your tracker — spreadsheet, Notion database, whatever you use — and paste in the 10 URLs, one per row. Then add a few basic columns so you’ve got context at a glance: URL, page title, content type, last updated date, and if you’ve got analytics access, include the last 90 days of traffic. That’s all step one is: a clean list, with enough detail to make the next seven steps faster and less painful. Now classify the content so you can spot patterns. You’re looking for types like: how‑to guides, product pages, FAQ sections, comparison articles, listicles, news or blog posts, landing pages, category pages. Different content types need different AEO approaches, so this wee classification step matters. Then tag each piece with its primary topic or entity. If you’ve got 50 articles about “email marketing,” they should all be tagged as such. This entity-based grouping connects directly back to episode 2.1 on entity-first optimization. You’re looking for topical clusters where you might already have authority but aren’t surfacing that authority properly for AI systems. Now — a tool to help. If you want to make this part faster, you can crawl your site with Screaming Frog. And honestly, Screaming Frog is one of the unsung heroes of this entire process. The free version alone is a bit of a powerhouse for getting a clear picture of what content is published, how the content is structured, and a clean export of your URLs with handy metadata like titles, word counts, publication dates, and last changed dates. And just to be crystal clear: I’m not sponsored by Screaming Frog, I just genuinely rate the app.
Step Two: Question‑Answer Gap Analysis. Now we get into the part AI systems actually care about: clear questions, and simple answers — early, direct, and quotable. Remember episode 1.2, “Question-Based Content: The Secret Sauce of AEO,” where we talked about question-based content architecture? Now you’re going to audit whether your existing pages actually answer questions clearly. Go through those same 10 pages and ask yourself: what question is this page trying to answer? If you can’t immediately spot an obvious question, that’s your first red flag. AI systems retrieve content that directly answers queries, so vague, meandering pages without a proper question–answer structure get overlooked. For each page where you can identify a logical question, write the question as a clear heading, then put a direct answer immediately underneath — and make sure that direct answer appears within the first 150 words after the heading. If someone asked the question to ChatGPT or Perplexity, would the first 150 words after the question feel like a proper answer? If not, that’s a quick win: restructure the introduction so the page leads with the answer. In that same tracker from step one, add four key tracking columns. We’re simply capturing: the question, where the answer sits on the page, how clear that answer is, and the priority for fixing it. Start with the question — that’s the anchor for this whole check. Be sure to use normal human language, the way an actual customer would ask the question. Write the question down in your tracker exactly as an actual human would ask it. Then knock out the Answer location column and the Answer clarity score column — both columns have fixed response options, so the two columns are quick to fill. For the Answer location column, you have one of three responses. Start, Middle, or End. Ideally, you want the direct answer to land in the first 150 words after the question heading, because that placement makes it easy for AI systems to pair the question and the answer as a clean question–answer relationship. If the page gives the first direct answer in the first 150 words, the Answer location response is Start. If the page gives a direct answer, but the reader has to scroll before the first direct answer shows up, the Answer location response is Middle. If the page holds the first direct answer until the end of the main body content, usually the last section or closing paragraphs, the Answer location response is End. The goal for the Answer location column is to earn Start across the board. Middle and End mean the page needs a wee bit of structure work to move the answer up. For the Answer clarity score column, you’re grading the quality of the answer — not the placement.
A page can earn a 5 for the answer and still have an End in Answer location if the page hides the answer late in the content. A 5 means the page gives the right answer in clear, quotable language, usually in a sentence or two — and I love spotting 5s because a 5 tells you the message is already solid. Scores 4 down to 2 mean the page is on the right track, but the answer needs tightening, simplifying, or de-jargoning. A 1 means the page never properly answers the question, and the missing answer should be addressed right away. For the last column to wrap up the Question-Answer Gap Analysis, use the Optimization Priority column already in the tracker and mark each page High, Medium, or Low. Use High for pages that matter to the business and show a Middle or End answer location or a low clarity score. Once you've filled the tracker, the quickest wins are easy to spot: any page with Middle or End is usually one structural tweak away from Start, and getting that Start answer is the whole point of this step. Step Three: Schema Markup Assessment. Right, structured data — unsexy, but powerful — and this is where a few small tweaks can punch well above their weight. The reason schema matters is simple: schema is how you tell Google and AI systems what the page actually is, so the page can be understood, pulled into results, and surfaced properly. So here’s the check. Take the same top 10 pages you’re auditing and run each URL through Google’s Rich Results Test, then note the schema types that show up on the page. Most sites have the basics like Organization or WebPage, but quick wins usually sit in the missing types: FAQ, HowTo, Article, Product, or Event. Now cross‑reference the schema against the content type you already tagged in step one. If the tracker says “how‑to guide” and the page has no HowTo schema, that is a tidy win. If the tracker says “FAQ” and there’s no FAQ schema, that is an easy fix. If the tracker says “product page” and the page is missing Product schema with offers and reviews, that is a big opportunity. Now, to keep this as clean and repeatable as step one and step two, add three columns to the same tracker — one for what schema is already there, one for what’s missing, and one for how hard the fix will be.
So for each page you’re checking, you’re answering three quick questions: what have we got, what do we need, and is this a quick win or a bigger job? The reason behind these three columns is simple: the Schema present and Schema missing columns tell you exactly what to add, and the Schema effort column tells you how painful the fix will be. Once those three columns are filled, prioritize schema fixes the same way you prioritize everything else in this audit: pages with real traffic plus missing schema go to the top, because proper markup amplifies pages that already have momentum. And remember episode 2.2 on dynamic schema strategies — while you’re checking schema, also flag pages where conditional schema makes sense, like adding Event schema to pages about upcoming webinars or adding Recipe schema on cooking content. Those contextual schema upgrades are easy to miss, but they can be absolute gold in answer engines.
Mid‑Audit Summary: Focus on the Foundation. Right, before we get too deep into the weeds, let me give you a bit of perspective here. Steps one, two, and three in this process. Well, that is your inventory, question‑answer gaps, and schema markup — and those three steps should be the focus first. Run those three audits across the full list of 10 URLs, then use the tracker to correct the scores: move answers toward Start, tighten weak answers, and fill the schema gaps. After you’ve finished steps one through three for all 10 URLs and made the quick fixes the tracker points to, then you expand out to steps four through eight. Honestly, if you only did steps one to three well, you’d be miles ahead of most sites trying to crack AEO. Here’s why. Step one tells you what you have. Step two tells you if each page answers a real question clearly and early. Step three tells search engines and AI systems what each page actually is. So when you finish steps one through three, you’re no longer guessing — you’ve got a clear list of what to fix first, and why. So if you're short on time, resources, or just want to prove the concept internally before going all-in, focus there first. Get those three sorted, document your wins, then come back and layer in the remaining steps as you build momentum. The other steps (RAG optimization, evidence auditing, conversation testing) absolutely amplify your results. But you can also fold them into your process over time, or combine them with your quarterly content reviews. You don't have to do everything at once. Think of these steps one through three as the foundation. The remaining steps are the scaffolding that turns good into brilliant. But you need that foundation first. Alright, with that said, let's crack on with the full picture.
Step Four: AI Conversation Pattern Evaluation.**Right — step four is where you pressure‑test your pages in the real world. Steps one through three of this process tell you what’s on the page. Step four tells you whether an AI tool can actually use it — quote it, trust it, and cite it. Right, so here’s what we’re doing in AI Conversation Pattern Evaluation — and why this wee test matters. We’re not trying to “see if ChatGPT likes the page.” We’re trying to see if an AI tool can actually use the page as a source in a real conversation: pull the right answer, cite the page, and give your brand the attribution. Open ChatGPT — and if you want to sanity‑check the results, open Claude or Gemini alongside it. Paste the page URL into the chat, hit Enter, and then ask one real customer question on the next line. And a “real customer question” is just plain English. No jargon. The kind of thing someone types when they’re stuck or trying to buy — and it usually includes a few real nouns from the business. So instead of using the prompt “What is this page explaining?”, use sample questions that your customers would actually type into ChatGPT — plain English, with real nouns. Now, a quick note: this is the bit where a lot of my clients freeze. They know they should ask better questions, but the first two or three prompts feel awkward. And that’s normal. The trick is you don’t need perfect questions on day one — you just need a starting point. Once you run a few tests and see what the AI does with your page, you’ll find it way easier to tailor your prompts to your industry and your customers. So here are a few “starter questions” to get you moving: “Does Power Plumbing offer same‑day emergency drain cleaning in Belfast city center, and what’s the typical turnaround time?” “What’s included in HubSpot’s Marketing Hub Starter plan — and what features cost extra?”
Each of those sample questions had two parts: the main question, then a bit of added detail or context. And honestly, that two-part structure is deliberate — it mirrors how real customers ask questions when they're trying to solve a problem or make a decision. But you don't need to stick to that format. Feel free to simplify your test prompts down to one clean question, or go the other way and add more complexity — multiple conditions, edge cases, or follow-up clarifications. The goal isn't to follow a script. The goal is to test whether your page can handle the kinds of questions your actual customers are asking, in the way they're actually asking them. So try a few variations. Start simply, then layer in complexity. See where the AI breaks down, where it gets vague, or where it pulls from a competitor's page instead of yours. Those breakdowns tell you exactly what to fix. Here are the three things we’re watching for in this step — in AI Conversation Pattern Evaluation. Three things: does the AI cite your page, does the AI summarize your page accurately, and does the AI pull the key facts correctly. Let me say that again. Does AI summarize your page accurately? Does it describe your page the way you’d describe it, or does it drift and make up its own version? And does the AI pull the key facts correctly? Does it get the important details right — names, numbers, steps, definitions — or does it twist them? Now, how do we record this in the tracker? Add three more columns to your existing spreadsheet: AI Citation Status: Mark as "Cited", "Not Cited", or "Partial" — did the AI reference your page when answering the question? Summary Accuracy: Mark as "Accurate", "Vague", or "Inaccurate" — did the AI describe your page correctly, or did it drift and make up its own version? Fact Extraction Quality: Mark as "Clean", "Twisted", or "Missing" — did the AI pull the key details correctly (names, numbers, steps, definitions), or did it get them wrong or skip them entirely? For pages marked "Not Cited," "Inaccurate," or "Twisted, "add a quick note in a fourth column called What Broke — just a sentence or two describing what went wrong. That note becomes your fix list. And here’s a wee bonus tip — this one’s a bit more service‑driven, because it helps you turn “AI testing” into something you can actually use. Pick one paragraph from the page — ideally the paragraph that should be doing the heavy lifting — and paste just that paragraph into the AI. Then ask one question that the paragraph should be able to answer on its own. Why? Because if a single paragraph can’t carry the answer, the full page usually won’t get quoted or cited either. This quick test tells you whether your copy is clear, self‑contained, and easy for AI to lift. After that, keep it simple in your tracker. For each URL, record one quick Step 4 note: did the AI cite the page, was the answer accurate, and what broke when it broke. And here’s the quick‑win rule to finish: fix the pages that are close but failing — the ones with the right answer but no citation, or the ones that are accurate but too vague to quote. Those are usually small edits that punch well above their weight.
Step Five: RAG Optimization Check. Right, step five is all about one thing: can your content be lifted in chunks and still make sense? I call that chunking — the way a system like ChatGPT pulls a small section of your page, not the whole page. And before we get too technical, let me give you a wee Irish story, because this idea makes more sense when you picture it. My granny had nine children. Nine. And if you’ve ever been in a house like that, you’ll know the rule: every child has a story, every child has an opinion, and somehow the entire room is talking at once. Now imagine you walk into that kitchen halfway through the conversation. If you only hear one sentence — one wee chunk — you might hear, “Aye, but that’s not what happened at all,” and you’d have no clue what the sentence means, because the sentence depends on everything that came before. That’s chunking. That’s retrieval. AI systems don’t “read” your whole page the way a human does — they pull a section, a paragraph, a chunk — and if that chunk is unclear on its own, you don’t get quoted, and you don’t get cited. So in Step Five, we’re not judging your writing style — we’re checking whether the page is structured in a way an AI can lift and reuse. Now, here’s what to check for. First, look for enormous walls of text. If a page is long — say, 1,500 words or more — and it has barely any subheadings, that’s a problem. Break it up with clear H2 and H3 sections so that each section has one job. Now, headings are one of the most misunderstood parts of writing on the web because people use them like decoration. AEO and chunking structure the headings. Your main headings should act like signposts — each one tells a human and an AI, “this section is about this.” And the subheadings underneath split that section into smaller, scannable chunks: steps, examples, definitions, FAQs. When you do that, your page stops being one long blob and starts behaving like a clean outline an AI can lift from. Here’s what can go wrong. People treat headings as a design tool — just for making text look big — instead of a structure tool. They use vague headings like “Overview” or “More info,” which tell humans and AI nothing. Or they stick paragraphs under a heading that don’t match the heading — like a “Pricing” section that turns into a history lesson. Moving on to the second part of Step Five: the RAG Optimization Check. We have to test if each section stands on its own. And yes, I have a simple way to run that test. Pick one H2 section. Then pick one paragraph from the middle of that H2 section. Now pretend that paragraph is the only thing an AI system retrieved. Read the paragraph on its own and ask two quick questions: Does the paragraph make a complete point on its own, or does the paragraph reference “as mentioned above”, “as we said earlier”, or rely on a previous definition? And if a reader landed on that paragraph cold, would the reader still understand the point — or would the paragraph feel like the paragraph is missing the setup? If the paragraph fails either check, the fix is usually small: add one anchor sentence at the top of the section that names the topic and the goal, and tighten the first line of the paragraph so that the paragraph can stand on its own. If you want a fast shortcut, copy the paragraph into ChatGPT and ask: “What is this paragraph about, and what question does this paragraph answer?” If ChatGPT can’t answer clearly, the paragraph probably needs more context. And third, do a quick “anchor noun” scan across the page. Look for clusters of vague pronouns like “this”, “that”, “they”, “it”, “these” — and make sure each paragraph introduces a clear noun or entity early, so a retrieved chunk makes sense without the rest of the page. Now let’s wrap Step Five: RAG Optimization Check in a way that stays dead simple in the tracker. For each URL, record one Step 5 result and one reason — that’s it. Step 5 result: “Chunking: Clean” or “Chunking: Needs work”. Step 5 reason: one short phrase that explains why — “Long paragraphs”, “Headings missing”, “Sections rely on earlier context”, “Too many vague pronouns”, whatever fits. That single line becomes the fix list. And here’s the quick‑win rule to finish Step 5: fix structure before copy. Add or clean up H2/H3 headings, split long paragraphs, and add one anchor sentence at the top of any section that needs context. Those are the changes that make RAG retrieval easier fast — without rewriting the whole page.
Step Six: Evidence and Citation Audit. Right, step six is the trust layer — and this is where a lot of websites get a wee shock. Because AI tools are like a cautious mate. If a page makes a big claim but the page shows no proof, the AI might not cite the page… and the AI might grab a different source that sounds more authoritative. And this is where attribution becomes the entire game. A lot of brands don’t just want the AI to be right — brands want the AI to name the brand and link the brand as the source. No attribution means no visibility, no authority, and no downstream trust. So step six is simple: give the AI a reason to trust the page. So how do we check for those trust triggers? One quick and easy method is to fire up your platform of choice — ChatGPT, Claude, Gemini — paste the page URL, and ask one normal human question that forces a factual answer, a number, a requirement, a policy, a guarantee, a step‑by‑step, a date. Pick a question where the AI can’t waffle. Then look at the answer and check two things: did the AI cite the page, and did the AI pull the proof correctly — sources, stats, references, certifications, awards, testimonials. Start with the website pages that are close but failing — pages that have the right answer but don’t get cited because the proof is missing. Add lightweight evidence first: link the source for each stat, add one testimonial, and add one actual photo with a descriptive filename. Then check the awards section. If the awards section is “logos only,” add one short paragraph per award that explains what the award is, what the award recognizes, and why the award matters? Those small proof upgrades can move citations fast — without rewriting the whole page. Look, content auditing isn’t glamorous. It’s not the fun “new shiny thing”. But it’s where the easiest wins are hiding — and honestly, it’s where a lot of confidence gets built as well. I’ve seen people come into AEO thinking they need to “do AI” by rewriting the whole website — and then a week later they move one answer up the page, tighten one definition, add one bit of schema, and suddenly the page gets pulled, quoted, and credited.
And the funny thing is, the folks who’ve been doing schema for years? You’re finally getting rewarded for being the boring, disciplined one. You were right all along. So if you're brand new to AEO or SEO, don’t panic — start small and ship a few tidy wins. And if you’ve been at this for ages, enjoy the moment: the world is finally catching up. Next week we’ll get into the dual optimization approach — how to write for humans and for AI systems without turning your site into robotic nonsense. Until next time, I’m Gary Crossey, helping you make your content speak AI fluently. May your content always earn answers, not just clicks!