? AEO guide · part one · full transcript

The complete guide to answer engine optimisation in 2026, part one

This is the full transcript of part one of my video guide to answer engine optimisation. It covers what AEO is, where it came from, how it differs from GEO and SEO, why the click is dying, the four things that have to happen before an LLM will cite you, and how answer engines actually work. Part two covers what actually gets you cited and how to build a 30 and 90-day plan.

By Simon Young, Founder, Question.Marketing. Recorded August 2026.

What is AEO?

Answer engine optimisation is getting your knowledge selected, quoted and attributed inside a machine-generated answer, rather than getting a link ranked on a page of links in Google or Bing.

In reality, we are moving into a world where people are asking questions of AI assistants, whether that's Claude, Gemini, Google, Perplexity, Copilot or any of the others. ChatGPT would be the most obvious one, and that's where I started out using it a lot. Once you're asking a question, you expect an answer. And we're moving, very quickly, into a world where search is becoming no-click.

People get the answer from the AI and then don't move on to a website. That's significant. Publishers lost around a third of their Google search clicks in 2025 alone, and small sites have lost 60 per cent over two years (Chartbeat data, reported by Axios, March 2026). That is only set to increase. They might go on to ask other questions. I think that's where we end up: a marketing world where someone asks one question, then maybe a qualifying question, moves down a funnel, and then lands on a page and converts into a customer, an order or an enquiry.

The important thing is this. As we build out the question bank and the knowledge bank, moving people down that funnel becomes more and more important, and the people who land on your site, whether it's e-commerce or a normal website for an enquiry, are far more qualified. Think back to normal search, pre-AEO. What a lot of people would call SEO was driving people onto a website: rank a page in Google, get discovered, land on a page, and then take time to research at that point.

A lot of people are doing the research inside the answer engines now. They end up far more qualified, because by the time they've landed on your property or seen your brand, those people convert around four times better than a visitor from a normal Google search. Semrush put it at 4.4 times in June 2025 (Semrush, AI search traffic study), and that matches what I see with clients. It's the equivalent of somebody having searched, gone to three or four different websites, made a decision, and then enquired or bought. Think of answer engines as putting people in a position where they're finding out the information, pre-qualifying themselves, getting exposed to your brand, and then moving down a funnel to the point of enquiry.

So getting listed in the answer engines is incredibly important. Which raises the obvious question.

AEO, GEO or SEO: which one wins?

My personal belief: AEO is where the answers and questions will live, and that will be the marketing part of what becomes the new voice and AI engine web. GEO, generative engine optimisation, is the work that gets you into that space. SEO comes secondary to that now.

A lot of the SEO guys would say the opposite: that AEO and GEO are a secondary part of SEO. Let's see how that all plays out. Maybe I'll be playing this video back in five or ten years' time. Might be sooner. I'll try and settle some of that debate as we go.

How quickly can you get cited, and why is Reddit a warning?

It is possible to get cited really quickly in the AIs. It's different across all of them, and things are moving fast. Up until about a week ago, if you populated lots of content into Reddit, you'd find you got cited pretty quickly.

That's changed recently. A lot of the engines are now, not disavowing Reddit exactly, but taking much less of their information from it. Maybe they felt it was getting filled up with information they didn't want, or that wasn't trusted. I'll come back to Reddit later, because it's the clearest lesson in this whole guide about where not to build.

How do you measure where you are in the LLMs?

Badly, mostly, because the tools are in their infancy. We're in a wild west.

There's a lot of debate around measurement at the moment, and a lot of tools that supposedly do it. Because I'm in the space I get offered tool after tool, day after day. "This tool works, this is the latest, greatest tool that's going to measure everything." There are tools like Semrush and Ahrefs that have some level of visibility to show you where you sit. Google Search Console will give you indications of how many sessions you're getting from AI. It is in its infancy.

Which is exactly why answer engine optimisation is, I believe, the biggest opportunity and the most important change in the way search is working right now. You have a huge opportunity as a business to be listed, be found, be cited, and then become the answer.

What has to happen before an LLM will list you?

Four things: access, retrieval, extraction and trust. Miss any one of them and you have no chance of being featured in the answer engines.

Access. Your content has to be reachable. A lot of websites use Cloudflare, and by default the AI crawlers don't get through it. You need to make sure your Cloudflare is set to allow AI crawlers onto your site. Otherwise you're never going to get listed.

Retrieval. You need to be found when the question gets asked. If you physically ask Google a question, you'll probably get an AI Overview or an AI answer. Scroll further down and you'll see a section called People Also Ask. Google has been doing this for years and years. I was talking about it back in 2017 and 2018. There's a web of questions out there, tons and tons of content built off the back of what people have asked Google over the years. Treat it as a worldwide knowledge base and FAQ for everything. That's effectively how the answer engines got built.

Extraction. The engine has to be able to understand what you've put and pull the answer out of a page or a piece of content. It pulls from voice, image and text. If it can't pull the answer out, again, you have no chance. There's schema and markup within content to tell Google or any of the other engines: this is your question, this is your answer. Unless you format your content like that, you're never going to be found.

Trust. The last piece, and probably the most important as we go forward. Can the content you've published be trusted and cited? That means not just publishing it on your site but becoming the answer, the trusted answer. That gets measured by whether you're on other sites, whether the answer is understood and viewed as true when people receive it, and whether people interact with it and ask further questions around the subject. That tells the answer engine it's true.

Being first is really important as well. If there are gaps in the knowledge base of the web, or of the AI, you want to fill those gaps in. There are ways to find out where those gaps exist in your subject, and that's probably one of the most important things you'll learn from this video.

So: access, retrieval, extraction, trust. You'll have to make sure your site and your content fit all four, so you can serve the answer and have it trusted.

Was I wrong about voice?

Partly. I said back in 2018 and 2019 that we were moving into a world that was going to be voice driven. I still believe we'll get there. It's just not quite yet.

Search was bringing back a load of documents, and it was very difficult to get that into voice. I honestly thought voice was going to be the interface. But people didn't really want to talk to a speaker in the corner of their kitchen, did they?

I wrote a book about speaking to your toaster a long time ago. It was based on the fact that there are devices in your house listening to you all the time. My phone is sat here while I'm recording this, and it has WhatsApp on it, and WhatsApp is listening to my conversation. How do we think Meta paid 19 billion dollars for WhatsApp? They didn't do it to provide a free channel to the world so we could all communicate. They did it so they could harvest data from voice. So we've been in a world for however long Meta has owned WhatsApp, ten years I'm guessing, where voice is already harvested, analysed, put into a knowledge base and used to create data. So don't tell me we're not going to get to voice eventually. We absolutely will.

Why do you need to publish before anybody else?

Because the engines pick up unique content and reference it, and the first trusted answer becomes the one everybody else has to argue with.

AEO, GEO and SEO aren't really competitors, and we've touched on that already. This gets you retrievable, extractable and repeatable. It's creating content, and that's always been the same thing: creating content to become the listed and relied-upon answer, to get into a ranking position, to become the authority. Becoming the authority is exactly what you want. If you're talking around a subject and you have some unique knowledge, publish it. Get it out there as quickly as you can, before somebody else does. The generative engines will pick that content up and reference it, and you can then be in People Also Ask, in voice, in the AI Overview and in the chat assistant.

We're also moving quickly into a world where people talk to websites. A lot of chatbots are disappearing in favour of speaking to an AI assistant and having a conversation with it. You can't tell me that in three, four, five years' time you won't be talking to something that feels like a human but isn't. How quickly that happens, I don't know. Could be five years, could be ten. I think it'll be sooner, given the rate everything is developing.

The visibility in the large language model, and how the response it generates develops, is really interesting to me. This is the nerdy piece I love. I can get a video ranked at the top of YouTube within 20 minutes. That's not a boast or a lie. I've done it repeatedly, and I'll show one in part two. The reason is YouTube wants fresh content all of the time, on any subject whatsoever, and it will serve it to people. If someone searches, and there's a new video on the topic, they'll be shown that video.

The important thing is: is the video any good? Do people watch it? Are they retained on it? Do they find it interesting and useful? Use that principle when you go through this process. Can I make useful content? Will people want it? Is it useful to the person reading it, or to the answer?

When you're optimising for answer engines, I personally believe a lot of content is served up across multiple LLMs to test it. Then, when it's cited, the AI engine measures how the audience reacts. Do they find it useful? Is it authoritative? Is it served from trusted sources? Does it answer the question? Think of it that way.

Hopefully I haven't gone too deep into the nerdy part. Technically, just think about what you're producing. Am I copying something that's already out there? Or can I give somebody additional knowledge that is worthwhile in this subject?

How much has click-through actually dropped?

Every study I'm looking at measures a huge decline, not a slight one, and it's increasing month on month.

Go to Google and search a question. Take your business as an example, and ask a question one of your customers might have asked. While we're on that: one of the best ways to create a knowledge base is to record all your phone calls with your customers. You'll hear all the questions, a bit like WhatsApp does with its customers. Record the questions, the answers, the feedback, how people talk to you and what they want to know. Feed that into your knowledge base. It can all be transcribed now, and that creates an enormous opportunity to create content.

People are not clicking through, and that's a huge problem for a lot of businesses and websites. If the answer comes from AI, if it's served by AEO or GEO, that is not SEO. SEO may well get your site listed for "recruitment in Manchester". But if somebody then asks, "Who is the best software or technology recruiter in Manchester?", or asks about a very specific job title, that is incredibly different to what would have been searched and served in Google.

Here's a little piece of research. It's not the case that a million searches on Google produced a million, two million, three million clicks, because people clicked on multiple sites and on ads. Now somebody searches, the answer is given to them, and a massive number of clicks simply don't happen. Pew Research tracked 900 US adults across 68,879 real Google searches in March 2025. When an AI Overview appeared, people clicked a traditional result on 8 per cent of visits. When it didn't, 15 per cent. That's nearly half the clicks gone. And only 1 per cent clicked a link inside the AI Overview itself (Pew Research Center, July 2025).

So you're seeing the shift in search intent and search activity, in how we all browse the web, changing very quickly in this latter part of 2026. Look at how position-one click-through rates have shifted. Ahrefs compared 300,000 keywords in Google Search Console, December 2023 against December 2025: position-one click-through down 58 per cent where an AI Overview appears, up from the 34.5 per cent drop in their April 2025 study (Ahrefs, February 2026). In Germany, SISTRIX ran over 100 million keywords and watched position-one CTR fall from 27 per cent to 11 per cent when an AI Overview is present (SISTRIX, 2026). Seer Interactive tracked 3,119 informational queries across 42 organisations, 25 million impressions, June 2024 to September 2025: organic CTR from 1.76 per cent down to 0.61 per cent, a 61 per cent fall, and paid CTR down 68 per cent (Seer Interactive, September 2025 update). Every single study measures a huge decline, and it's increasing month on month.

There is no debate, absolutely no debate, on where we're going. How the media is displayed doesn't matter. This is a pure shift in search activity because answer engines are providing the answers more quickly. It's efficient. That's the whole point. People are getting what they want faster through a different means, and the adoption is happening at scale.

If this had happened back in 2018 and 2019 with the answer engines of the time, meaning voice search, we'd have seen the shift then. The adoption just didn't happen. Maybe the knowledge base wasn't there. The compute power to serve answers quickly is there now.

Part of what I'll talk about to my audience as we go forward is local search. Nobody is really concentrating on local AI results being served yet, because everybody's gone crazy about how to get into the LLMs in the first place. If you're a local business and you want to be listed, start thinking about it now. The Manchester example is a prime example.

So, again, the traffic didn't move to a different channel. It's gone. It hasn't gone to AI search in terms of clicks yet, and the stats here are a little misleading. Sixty-eight per cent of US Google searches ended without a click in the first four months of 2026, up from 60 per cent in 2024 (SparkToro, using Similarweb clickstream data, June 2026). That's a huge shift. And AI referrals haven't replaced it: they're still under 1 per cent of publisher traffic (Chartbeat) and about 1.08 per cent across 13,770 enterprise domains (Conductor, 2026 AEO/GEO Benchmarks). The click has disappeared.

I am starting to see more links referenced in AI search, particularly in Claude and ChatGPT. More links are popping up within the results that get served. Maybe we move to a landing page world, where we're still optimising pages to appear in AI search and then seeing people click through. But the links appearing at the moment tend to be citations, so you can see where the knowledge was drawn from. That's interesting, but it won't necessarily be a page that converts. E-commerce will be a whole new ball game.

Being cited doesn't just replace the click. It improves the clicks you get. I cannot emphasise enough what this means. When you appear in those AI answers, what you get is much better intent within the click. When somebody ends up on your content after finding you in AI, they've probably already asked two or three questions and been coached, if that's the right word. They're more likely to convert. And brands cited inside an AI Overview earn 35 per cent more organic clicks and 91 per cent more paid clicks than brands on the same query that aren't cited (Seer Interactive).

You can move very quickly. I've seen websites that are 10 or 15 years deep in SEO, almost dominating their niches, and then new competitors springing up with new websites answering questions that simply hadn't been built yet. So build the content quickly, and build unique content as quickly as possible to fill the gaps. In the example where somebody searches a very specific job title in Manchester, if you build that content, it's a bit like back in the day when we used to build very long-tail keyword phrases and build content specifically for them. That's where we're at. Build the content that matches the question and the answer.

Why is AI slop the wrong shortcut?

Because unique content that you know and can cite is slower, and it wins. Absolutely it wins.

Citation is another matter, and it's going to be key to how your content is brought forward. Roughly two thirds of the time, content isn't retrieved live at all. Semrush's 17-month clickstream study found ChatGPT triggers a web search on about 34.5 per cent of queries, down from 46 per cent in late 2024; the rest is answered from what the model already holds (Semrush, reported by Search Engine Land, April 2026). That's why being in the training data matters as much as being on the page.

Building content used to be so slow. If I'd said to an agency, "Build me a hundred blogs around a topic," it would have cost a fortune. Nowadays we can build that content really quickly. What we don't want to do is do it completely with AI. The biggest term in marketing this year is AI slop. Everybody talks about it. LinkedIn even has an AI slop button now, in theory so people can report it. What we actually know is that LinkedIn is categorising content on whether it thinks it was written as user-generated content, as AI, or a bit of both. A lot of people are using the bit-of-both approach.

I strongly encourage you to use unique content that you know and can cite, so you become the cited source. It's a slower way of doing it, but you will win. Absolutely you will win if you're writing unique content and serving it into the answer engines in your own voice and your own brand. Think about share of voice. Are you cited based on your brand? Are you training that data correctly? That's what we want to make sure we're doing.

Do all the engines use the same sources?

No. They've all got different indexes and they all do different jobs.

ChatGPT, as it stands today, works mainly with things like Wikipedia, LinkedIn, and brand-owned content presented structurally properly. Google works with YouTube and still works a lot with existing SEO. I believe that will have to change, but a lot of it is built on Q&A, on People Also Ask, and on what we're all talking about.

Say I sat here today and talked about a Ford Ranger, because maybe I want to buy one for my wife. I'm probably going to get adverts for a Ford Ranger tomorrow in my feed, on Meta, on TikTok, when I'm browsing the web, on Google, on YouTube if people have tagged the content and built the ad the right way.

I did a huge piece of work on YouTube years ago, ranking videos at the top and building Q&A videos. Think of a scenario where you build a blog page. New knowledge in the blog: good. Unique statistics in the blog: yes. Your knowledge in the page: yes. Structured correctly with schema and JSON: right. Content near the top, answers you want to be found for: great. Getting it found: good. Linking to relevant parts of your site on topic: excellent. What can we do on top of that?

I ran an agency that specialised in YouTube ads and building content around answers in video. The problem was clients didn't want to pay to make the video, or have the time to go on camera, or want to go on camera. Now we have the opportunity to make a lot of video content using AI. I'm not suggesting you build tons of videos that are entirely AI driven, because I think the search engines are going to pick that up too. Authentic content on video, with Q&A, a bit like what I'm making here. This video is, after all, an experiment in answer engine optimisation, to see if it performs and is found by the LLMs and cited as a source for how to rank in answer engines.

Google uses YouTube a lot. Google owns YouTube, for those of you who don't know, and a lot of classic SEO sits within YouTube. Ask yourself: would you rather watch a video or read a web page? Ninety-nine per cent of people would rather watch the video and have it shown to them properly. Making that content is difficult. It's expensive, it's time consuming, and it's knowing what to put in and how to structure it. That's what stopped YouTube doing as much as it could in search. Facebook's Nicola Mendelsohn said in 2016 that Facebook would probably be all video within five years (CNBC, June 2016), and Cisco was forecasting that video would be 82 per cent of consumer internet traffic by 2020 (TechCrunch). I said Google on camera. It was Facebook and Cisco, and it never quite happened. But they foresaw a Q&A world where video answers questions.

So can you see a scenario where you're talking to an AI person? Let's call him Claude. Or her Claude. I don't know whether Claude is a man or a woman. Could be Claudine. But Claude becomes a person, an avatar, looks like a real human, pops up on your phone, and you ask it questions and it answers and finds that knowledge. Where's it getting that knowledge from? It's going to search all of these different sources, and it doesn't want to be wrong. It doesn't want to hallucinate.

Unless you put some really good guardrails on it, AI will go and make up information, because it cobbles together pieces from sources that may not all be correct and then takes its best guess. A bit like a human. If you'd only ever learned there were 99 pence in a pound, you'd repeat that back. You'd be wrong, but if there were a million sources telling you a hundred and only fifty telling you 99, you'd go with the million.

Within Q&A there are multiple answers. Certain questions will have many, many answers, and some of them contradict each other. So there is no single source of the truth at that point. But to be found in the answer engines, you need to become that single source of the truth. And it wants real content. It reads where content comes from and asks: is it trusted? I'm doing a lot of work on that with Claude, with Trustpilot and G2 and some of the other review sites, to make sure that if a piece of content is found, the AI can go and check its validity.

These engines are all doing different jobs. They're all competing. They'll all have different revenue models and different ways of answering us. But at the moment the crossover between all the sources is the web, voice search, and everything being laid down through a mobile phone. You don't know where that data is going. I believe personally that the majority of our lives are now being listened to or recorded. Somebody tell me I'm wrong. But we're in a world where all of this data is being harvested, and it's your opportunity to build knowledge for the AI. They are eating it for breakfast. They love it. The fresh new meat, the unique perspective. Look at any of the social channels: write a unique piece of content with statistics in it, and if people start digesting it and finding it interesting, the reach is huge.

What actually gets you cited?

The evidence. And nobody actually quotes it alongside the advice. This is where it gets fascinating, and it's where part two starts.

That's the end of part one of this AEO tutorial. In the second video I'll give you a good idea of what you need to be doing to get found in the LLMs, how to appear in AEO and in AI search overviews, and how to build a plan across 30 and 90 days. I look forward to helping you with that, and thanks for watching so far.

Before you go: open ChatGPT, Claude or Gemini, and ask it the question one of your customers asked you last week. Does your name come up?

Does your name come up?

Open ChatGPT, Claude or Gemini and ask it the question one of your customers asked you last week. If you aren't in the answer, that's what we fix. It starts with a measured baseline, not a contract.